Let me give you a number that will surprise you. It surprised me!

Only 8.5 percent of projects finish on time and on budget. I thought that the number would be low, but not that low. I was thinking like 25-30%, but that’s the finding from more than sixteen thousand projects across dozens of fields and countries compiled by researcher Bent Flyvbjerg in How Big Things Get Done.

After reading that, I actually feel pretty good about my projects. I can’t give you the number off hand, but I complete a lot more than 8.5% and almost all of them are under budget! Of course, most of the things that I work on are not as big as what Flyvbjerg studied.

Think about your last big project. Where did it land? Did it finish on time? Was it on budget? If not, well… your project reflects most big projects. Late and over budget!

Why?

Actually, the cause is almost always the same. I’ve address this in other posts, but the challenge is that many leaders rush into action before the plan is ready. In all too many cases that I’ve observed, they often can’t really define the problem they are actually trying to solve before the rush off to solve it.

This isn’t because the leaders are reckless and don’t care. It’s because the pressure to show progress is substantial and the stakes for inactivity are significant. Stakeholders want movement. Donors want to see impact. Boards want momentum. Organizations reward action over deliberation.

This means that organizations often break ground early and launch before they’re actually ready. They start building before we’ve finished thinking. Then they spend the entire execution phase paying for every assumption they didn’t examine in planning. This usually costs orders of magnitude more than if they had caught it on paper before breaking ground on that new project.

Here’s some common sense that many (me included) often ignore. Planning is cheap. Execution is expensive. If you’ve ever used Claude Code, you know why there is a “plan” mode.

Every mistake you find in a model, a prototype, or a whiteboard session costs almost nothing to fix. The same mistake discovered mid-build or mid-rollout can cost everything.

The innovator’s instinct is to move fast and iterate in production. It’s actually how I operate most of the time. It’s the perfect choice for projects of high reward and low exposure. It’s great for small projects that you can later scale.

It is exactly wrong for big ones. You don’t want to be iterating while building a nuclear power plant.

Start at the end.

Before you assign a single resource and before you schedule a single kickoff meeting, you should always start with the outcome. That’s on every project. Define the desired outcome first.

What does done look like? Not launched. Not delivered. Done. What’s the result you are actually trying to produce and how will you know without ambiguity that you’ve produced it?

Many projects skip this entirely. They begin with a solution already in mind and work forward. The begin assembling teams, building timelines, and allocating budgets… all without ever clearly defining the problem they are solving.

The result is a project that executes efficiently toward the wrong destination.

Working backwards from the end vision forces a discipline that forward planning never produces. It surfaces misalignment early… when it’s cheap. It exposes the assumptions everyone has been carrying silently and it forces the hardest question in any project.

Are we solving the right problem?

Define the problem well. Define the outcome precisely. Then plan how to get there. In that order.

Sharpen the ax first.

I admit it. I’ve launched before the plan was ready. More than once. There’s a pull to show traction, to demonstrate momentum, and to prove to the people watching that something is happening.

When I’ve acted without thinking about the end vision, it almost always costs more than the time I saved. Abraham Lincoln is said to have put it this way. If he had five minutes to chop down a tree, he’d spend the first three sharpening the ax.

That’s a metaphor about leverage.

A sharp ax makes every swing count. You reach the same result in fewer strokes, with less wasted effort, and far less risk of the blade glancing off and doing damage you didn’t intend. For big projects, the ax is your plan. Most organizations start swinging with a dull plan.

The planning phase is where you sharpen. It’s where you stress-test assumptions, pressure-check the timeline, define what success actually means, and find the problems that would otherwise find you mid-execution, at the worst possible time.

Take the time. Sharpen the ax. Then don’t stop swinging until the tree is down.

Most leaders know this. Few act on it. And for small projects, the cost of skipping it is manageable. But there's a category of project where a dull ax doesn't just slow you down. It can take everything down with it.

The risk nobody is talking about.

The distribution of project outcomes is not normal. Flyvbjerg calls these fat tails. A small number of projects don’t just run over. They fail catastrophically. So badly they threaten the organizations that launched them. So badly they consume reserves, trust, and credibility that took years to build.

When you only look at the averages, you will miss. You can look at historical overruns, say, 20 percent over budget and plan accordingly. What you miss is the tail scenario where the overrun isn’t 20 percent, but 200 percent. Where the project doesn’t just strain the organization but breaks it.

In these situations, you’re not just risking this project. A single catastrophic failure can take down everything around it. For those of us managing resources entrusted to advance something bigger than ourselves, planning slowly is not timidity. It is the only rational response to that kind of risk. It is stewardship.

Once the plan is ready, speed is protection.

Here’s the other half of the equation and where most deliberate planners still get it wrong. Every additional month a project stays open is another month of exposure. Leadership changes, priorities shift, and the global environment moves. Flyvbjerg says lost ground can be regained. Lost time is gone forever.

The goal of slow, thorough planning is not to extend the project lifecycle. It is to compress the execution phase to the shortest possible window.

In other words… Get in. Deliver. Get out.

The projects that succeed aren’t the ones that plan forever or execute recklessly. They’re the ones that invest everything upfront in thinking and then move with urgency once the thinking is done.

What does this actually look like?

Start with the outcome, not the plan. If your team cannot articulate the desired end state in one clear sentence, the plan is not ready to be written yet.

Define the problem before you fall in love with the solution. These are different activities. I know that I have already said it, but it’s worth repeating. Most teams skip truly defining the problem they are solving.

Find the two or three assumptions your project cannot survive being wrong about and break them cheaply before you commit resources. A conversation, a prototype, or a small pilot. AI can help do prototype in hours what previously took months. Take the time to do the prototypes. If the assumption doesn’t hold, you want to know now.

Set an execution deadline before you start. If you can’t name a target completion date before you begin, your execution phase will drift. Drift is how projects accumulate exposure.

Ask this question, “If this fails at the worst possible level, not the average bad outcome, but the catastrophic one, what happens?” If the answer is “it threatens things we cannot afford to lose,” the planning phase needs to be longer and harder than you thought.

Takeaways

Think slow. Act fast. In that order.

Start at the end. Define the problem before you fall in love with the solution. Find the two or three assumptions your project cannot survive being wrong about — and break them cheaply, now.

Sharpen the ax. Every mistake caught on paper costs almost nothing. The same mistake mid-execution can cost everything.

Know your tail risk. Ask what happens if this fails at the worst possible level — not the average bad outcome. If the answer threatens things you cannot afford to lose, plan harder.

Then get in. Deliver. Get out. Speed in execution is protection. Don’t plan forever. But don’t swing a dull ax either.

Where are you swinging a dull ax right now? What would it cost to stop and sharpen it first?

I’m frequently asked about the environmental impact of AI and Data Centers. It’s really hard to figure out the truth from fiction. There are a lot of special interest groups on both sides of that discussion and we’re early in the mass adoption of AI solutions. The environmental impact of 2026 will not be the impact in a year or two. The way AI works is changing. The future of compute is probably space where energy is free and tempretures are cold… but that’s the future. What about today and the next few years?

The question companies are currently throwing billions of dollars at is, “How do we build AI data centers faster?” It’s really crazy how fast these are being built because many are being built in anticipation of future needs. Nvidia is sold chips last year that haven’t even been put into production yet. Companies are buying the chips in anticipation of future needs.

Right now, it looks like what’s needed is bigger campuses, pre-fab construction, enormous buildings, modular substations, faster permitting, more power generation, new transmission lines, lobbying local leaders, and building in remote locations like rural North Dakota.

If you have been paying attention to the news lately, “how fast can we build” may not be the right question to be asking.

The most interesting AI infrastructure idea I’ve seen this year doesn’t try to answer that question at all. It takes a very complicated problem and proposes a completely different solution.

How might we deliver gigawatts of AI compute without building any new warehouses?

That single reframe, swapping “how do we build more” for “how might we deliver this differently,” is the move I want to talk about. It’s the innovators approach to solving big problems. The answer being proposed is genuinely novel. The thinking behind it is something every innovation team should be borrowing.

The Problem That Made Them Ask a Better Question

With a traditional approach to the data center problem, the answer became the constraint. The real problem isn’t can we build them faster. The real problem is how do we deliver compute faster. Who said that had to be in massive data centers?

The U.S. generated 4,430 terawatt-hours of electricity in 2025. Demand is projected to grow 25 to 32 percent by 2030. Global data-center capacity demand is projected to grow roughly 3.5x in the same window. That’s an increase from 82 gigawatts today to about 219 gigawatts by 2030, with AI workloads driving most of the increase.

To keep up, the U.S. needs to bring online roughly 80 gigawatts of new generation per year. The problem? We’re currently only building 40 per year.

There’s also a long queue of generation projects waiting on grid interconnection. That amounts to over 2,600 gigawatts in line nationally. Most of those will never actually be built. But the queue itself tells us something important. The bottleneck right now isn’t AI ambition. It’s the speed of the electrical grid. It’s a real problem.

Span and NVIDIA have a phrase for the underlying tension. They call it the “hyperscaler paradox.” It means that compute demand is scaling quickly, but the power delivery infrastructure to feed that compute can only scale at utility build-out speed.

This is a serious constraint for AI growth. We’ve had constraints before. We’ve worked around them before. What’s interesting right now is how people are starting to work around this one because the workarounds are coming from directions nobody predicted three years ago.

A Genuinely Novel Answer

A company called Span, the folks who make smart electrical panels, just published a whitepaper for a product called XFRA. They call it a “distributed data center” and it is very exciting!

That phrase does not mean what you think it means. This is not an enormous data center or even a few large data centers connected. It’s way outside of the box thinking.

Here’s the actual idea.

The average American home only uses about 40% of what its 200-amp electrical panel can deliver. The other 60% is provisioned, paid for, and sitting idle. Multiply that across tens of millions of homes and you have a staggering amount of unused electrical capacity already wired into the grid.

What if you put the data center there?

An XFRA Node is a weatherproof unit roughly the size of a backyard HVAC condenser. Inside that box? Sixteen liquid-cooled NVIDIA RTX PRO 6000 Blackwell GPUs, four AMD EPYC CPUs, three terabytes of memory, heat-pump cooling, and a 15-kWh whole-home battery. It’s wired into a SPAN smart panel certified to the highest safety standard for power control systems.

Span installs all of it at zero upfront cost. The homeowner gets a rental check that effectively subsidizes their electricity and Wi-Fi down to a fraction of normal, plus a premium electrical panel and battery backup they wouldn’t have bought on their own. In exchange, the unused headroom in their panel gets sold, second by second, to whoever needs AI inference right now. That’s genius!!!

Span partnered with NVIDIA, AMD, Dell, and PulteGroup, the third-largest homebuilder in the country. The first proof of concept goes live this fall in 100 homes in a build-to-rent community in the southwest. Beginning in 2027, they plan to scale to more than a gigawatt of annual capacity.

(SPAN + NVIDIA… if you’re reading and you’re looking for Virginia beta sites for the next phase, you know where to find me. Our electricity rates are quite reasonable. Just saying.)

No new warehouses. No new transmission lines. No four-year interconnection queue. Just headroom that already exists, monetized.

What “How Might We” Actually Looks Like

Here’s where innovators should take note. Notice the structure of what just happened. The industry was asking: How do we build data centers faster?

Span asked: How might we deliver compute without building any?

Same problem. Same constraint. Completely different question. And the different question opened up a solution space the original question literally could not see.

That’s the entire move.

“How Might We” (HMW) isn’t a brainstorming gimmick or a Post-It note exercise. When it’s used well, it’s a discipline. It’s a deliberate refusal to accept the framing the problem hands you. The default human move when something is too big or too slow or too expensive is to grind harder on the obvious answer. The HMW move is to put the obvious answer down and ask whether you’re even working on the right question.

Could XFRA fail? Of course it could. Homeowner churn. Regulatory pushback. Insurance liability. Latency that doesn’t deliver what the spec sheet promises. Real businesses get killed by boring problems all the time. And even if it works, it isn’t going to replace the training of the largest models still belongs in fortified compounds with their own substations.

That’s the wrong way to evaluate this kind of idea. The point isn’t whether XFRA wins. The point is the shape of the question that produced it.

Borrow the Move

It’s not uncommon to hear leaders ask questions like these.

• How do we hire faster?

• How do we raise more money?

• How do we get approvals through faster?

• How do we build more capacity?

Each one is the obvious version of a real problem or constraint they are facing. In each case, there’s usually a better “How might we” hiding behind the question being asked.

• Not “how do we hire faster” — but how might we deliver this without hiring?

• Not “how do we raise more money” — but how might we get this done with what we already have?

• Not “how do we get approvals faster” — but how might we structure this so that approvals aren’t on the critical path?

• Not “how do we build more data centers” — but how might we deliver AI compute without building data centers at all?

Same constraint, every time. Different question, every time. The different question reaches into a solution space the obvious question can’t see.

The Takeaway

Span may or may not turn out to be right about XFRA. Most novel ideas don’t make it. The proof of concept this fall could expose problems nobody has predicted. There could be a totally new solution that no one else has thought about.

The point of today’s post is about what they’ve modeled. They’re taking an enormous, expensive, slow-moving problem and refusing to accept the obvious framing. That’s the actual innovation work. The technology in the box is real, but it isn’t the lesson.

The lesson is the question they were willing to ask before they ever picked up a soldering iron. Most of us are answering the wrong question right now. We just haven’t noticed yet.

What’s the “how might we” your team is avoiding because the obvious answer feels too expensive, too slow, or too big to walk away from?

P.S. — Hey SPAN and NVIDIA… just a reminder, if you’re reading this and looking for Virginia beta sites for that next phase, you know where to find me. Our electricity rates are quite reasonable. 😉

To read more about this idea, see XFRA Whitepaper · Span / XFRA announcement

Most internships start the same way. Paperwork, passwords, a tour of the office, and maybe a welcome lunch if your organization is feeling generous. Then, somewhere around week two, you get assigned to something real. It’s usually something quasi-adjacent to what someone more senior doesn’t want to do.

Our innovation team didn’t do that.

Last week, our three summer interns arrived from Auburn University, the University of Florida, and the University of Virginia. Bright, capable, and genuinely eager. I love working with interns because they bring a renewed sense of curiosity and possibility to the room.

This year, once they finished their HR and technology onboarding, before I talked to them about any of their core summer work, before I assigned them to any project connected to AI or design thinking, I gave them one assignment.

Build a strategy game.

They were given a link to a website I created with the general guidelines. They are to build a game that runs all summer and secretly opens to everyone in the org. They need to use at least a hundred 3D-printed ducks, create a website where people compete, and keep people playing from June to August. This isn’t a research project about games. I want them to build one from scratch, as a team of three, with no template and no rubric.

Why a Game?

Here’s what that assignment is actually teaching them, even if they don’t fully see it yet. (And I know they’re reading this.)

It’s teaching design thinking, because a game that doesn’t work from the player’s perspective isn’t a game, it’s a frustration. They have to empathize with someone who isn’t them. They have to test, iterate, fail, and rebuild. They have to ask questions like, “What does the player experience? What decision points matter? What makes someone want to keep playing?”

It’s teaching AI fluency, because they’re building the game with the help of AI, not learning about it in the abstract. Every tool they reach for, every prompt they write, every prototype they generate, that’s real experience. Not a case study. Not a simulation.

It’s teaching innovation methodology, because they have to go from a vague idea to something tangible with no precedent. Just a challenge, a problem space, a team, and a deadline. Figure it out. We’ll track progress each week.

And underneath all of that, it’s teaching something harder. It’s teaching them that creativity is a form of stewardship. That the skills God has given them, systems thinking, design skills, technical fluency, and strategic imagination, are not secular assets they happen to carry into some future ministry context. These are to be used now, even while building a game about 3D printed ducks.

Why We Actively Recruit Interns

We take as many interns as we reasonably can. We invest in them because we believe they are the future of missional innovation. We also gain the benefit of fresh eyes looking at our challenges from perspectives that are often not in the room.

The church desperately needs creative and innovative people. Most of those people have never been shown that there’s a place for them in ministry.

They go to school for design, engineering, computer science, or business strategy. They work hard, get good grades, and are good at what they do. They love God and somewhere along the way. through sermons, culture, and the implicit message of a thousand conversations, they absorb an idea that their skills are for the marketplace and their faith is for Sunday. Ministry is over here. Innovation is over there. Sacred calling is separate from the skills and gifts God has given them.

I want to blow that concept up.

Not with a lecture, but with an experience. With a summer where they build real things, work on real problems, use real tools, and do it all inside an organization whose entire reason for existing is to see people from every nation, tribe, and tongue come to know Jesus Christ. To see Revelation 7:9 lived out among the nations.

My goal for our three interns is simple, show them the intersection of faith and technology before they’ve had a chance to decide there isn’t one.

Formation, Not Just Training

We don’t just need smart young professionals who can execute tasks efficiently.

We need people who understand the mission deeply enough to innovate from within it. People who, when handed a novel tool or a strange new capability, don’t ask “what would the business world do with this?” but “what would faithful stewardship of this look like, given where we’re trying to go?”

That takes formation… Formation is different from information.

Information you can deliver in a meeting. Formation happens through struggle. Through decision-making under pressure. Through the friction of trying to build something real and running into the places where your assumptions don’t hold. It happens when you’re three interns staring at a hundred rubber ducks, trying to figure out how to make an entire organization care about a game and you realize the answer requires you to actually understand people, systems, incentives, and design. All at once.

That’s the assignment. And it’s just week one!!!

The Longer Game

By the end of this summer, they will have worked on real innovation projects inside a global mission organization. They will have used AI tools not as a novelty but to build actual workflows. They will have practiced design thinking against real problems with real stakes. They will have seen that the people doing this work take theology and technology seriously in the same breath.

And maybe, just maybe, they will leave with a different picture of what their future could look like.

Not marketplace or ministry. Not secular skills or a sacred calling.

Both. Together. Integrated.

That’s the longer game. The strategy game of 3D ducks was just week one.

How might your organization build formation experiences like this for the next generation of innovators you’re developing? I’d love to hear what you’d build.

I once built a website nobody asked for.

In my defense, it was a genuinely good idea. A modern-day journal for orality, the field of ministry that focuses on reaching people who learn and communicate primarily through spoken word rather than text. Mostly practitioner-focused. Quasi-academic. A place where missiologists and field workers could share ideas, push the conversation forward, document what was actually working.

The niche was real, the need was real, and I was genuinely excited about it.

Here’s what else was real: the niche was also very small. Keeping fresh content coming was an enormous lift. The HTML involved in building something engaging enough to keep people coming back was, at that moment in my technical development, a bridge too far. And the question nobody asked me, not even me, was the one that would have saved us all a lot of time.

Who is actually going to read a journal about orality?

Not the concept. Not “is there an audience for this space?” The specific, uncomfortable, slightly humbling version: have you talked to three of those people? Do they feel this problem viscerally enough to come back next month? Is the lift of building and maintaining this justified by the size and engagement of the audience you’re actually building for?

Nobody stopped me and said: have you run this through a value proposition canvas?

Nobody stopped me and said: before you build the thing, validate the thing.

I built it anyway. It didn’t work. And the lesson I should have learned from that experience, the one that took me embarrassingly long to fully internalize, is that building is the easy part.

Adoption is the hard part. And adoption requires its own discipline entirely.

Building Is the Easy Part

Here’s what nobody tells you when you’re a person who is good at building things: the skill of building is not the same skill as the skill of adoption. They are different arts. They require different instincts, different questions, and in many cases, different people.

Builders ask: can we make this work?

Adoption asks: will people actually change their behavior to use this?

That second question is harder. Much harder. Because behavior change is expensive. It costs people something: time, habit, cognitive load, the friction of learning something new. And people will only pay that cost if the value on the other side is clear, felt, and greater than the cost of staying where they are.

Most builders, and I include myself prominently in this category, are deeply optimistic about this calculus. We built the thing, so we believe in the thing, so we assume the value is obvious. It is not obvious. It is almost never obvious. The value has to be demonstrated, communicated, and often experienced before someone will pay the adoption cost.

The orality journal seemed obvious to me. I was wrong about what was obvious.

The Value Proposition Canvas Nobody Used

There’s a tool called the Value Proposition Canvas. It’s not complicated. On one side, you map what your product does: the gains it creates, the pains it relieves, the jobs it helps people accomplish. On the other side, you map your customer: their actual gains, their actual pains, the actual jobs they’re trying to get done.

The goal is fit. Does what you’re offering actually match what they need?

The reason most builders skip this is not because they don’t know it exists. It’s because doing it honestly requires sitting with the possibility that the answer is no. Or not yet. Or not for this audience. Or not at this scale. And that’s uncomfortable when you’ve already fallen in love with the solution.

If someone had walked me through that canvas before I built the orality journal, here’s what we would have found: the gains I was creating were real, but they were gains for a very small, very distributed audience with no existing habit of consuming written practitioner content. The pains I was relieving were genuine, but people were coping with those pains in other ways: conferences, email lists, word of mouth. And the job to be done, staying current on orality practice, was being done well enough that switching to a new platform required a motivation that didn’t exist at sufficient scale.

None of that means the idea was bad. It means the idea needed more validation before it became a build.

I built first. I validated never. The site didn’t work.

Adoption Needs Its Own People

Here’s the other thing Part 1 didn’t have room for: even when the idea is right, even when the validation is solid, adoption still requires dedicated human attention.

Not leftover attention. Not “we’ll handle adoption after we ship.” Dedicated, intentional, ongoing attention from people whose actual job is to help other people change their behavior.

In the missions world, we call this MAWL: Model, Assist, Watch, Leave. You don’t just hand someone a new tool and disappear. You model how it works. You assist them through the awkward early stages. You watch them use it and correct in real time. And then, only then, you step back and let them own it.

That’s adoption. That’s behavior change. And it is a full-time job that most organizations assign to nobody.

Most organizations ship something, announce it, and then wonder why the numbers are low six months later. The answer is almost always the same: nobody was assigned to close the gap between “we launched this” and “people actually use this.” Those are not the same event. There is a chasm between them. And crossing that chasm requires people who understand that their job is not to build but to help others receive.

The orality journal needed a content editor, a community manager, and probably a dedicated champion in the orality space who would recruit contributors and readers personally. I had none of those things. I had a website and an idea and optimism. It wasn’t enough.

The Two Questions You Should Ask Before You Ship

If Part 1 was about sequencing ideas before you build them, Part 2 is about what happens after. Two questions, not five, not a framework, just two, that would have saved me significant time and energy over the years.

Question One:
Have we validated that people feel this problem enough to change their behavior?

Not “do people have this problem.” Not “is this problem real.” Do they feel it strongly enough that they will pay the switching cost to use your solution. That is a different, harder, more specific question. And if you can’t find five people who say yes with conviction, you’re not ready to build.

Question Two:
Who is specifically responsible for adoption, and what does their job look like for the next ninety days?

Not generally responsible. Not “the whole team.” Specifically. By name. With a job description that includes clear behaviors: who they’re talking to, what resistance they’re addressing, how they’re measuring success. If you can’t answer this before you ship, adoption will be an afterthought. And adoption as an afterthought is the same as no adoption at all.

Here’s What I Know Now

Good ideas need more than good execution. They need the right moment, the right audience, the right validation, and the right people focused on the right job after launch.

The orality journal wasn’t a bad idea. It was a premature idea, launched without validation, into a niche too small to sustain the lift required, with no adoption plan and no community manager and a founder who was already working on approximately 52 other things.

Nobody stopped me.

That was the problem.

So let me be your Wilkie for a minute: before you build the next thing, run the value proposition canvas. Talk to five people who have the problem. Assign someone to adoption before you launch. And ask yourself, honestly, whether the audience is large enough and motivated enough to justify the lift.

Your ideas deserve better than a website nobody visits.

Build accordingly.

If this landed, share it with someone who just said “we should build a thing.” They probably need both parts.

“You Might Have Enough Ideas to Destroy Your Own Organization.” Jeff Wilkie didn’t say it as a warning. He said it as a fact.

Jeff Bezos had enough ideas to destroy Amazon. Wilkie would know. He was Amazon’s longtime operations chief, the person whose actual job was to catch what Bezos would break. What he observed wasn’t recklessness. It wasn’t poor judgment. It was actually entirely the opposite problem.

Bezos was too good at ideas. And ideas, released faster than an organization can absorb them, don’t accelerate growth. They create turbulence that looks like momentum until something breaks.

I read that and felt personally attacked!

Here’s My Confession

At any given moment, I am actively working on approximately 53 things at once. I say approximately because I genuinely lose count.

I enjoy it. I’m wired for it. The juggling of enormous, full-time-worthy projects simultaneously gives me energy. It keeps me up at night… in a good way. It’s more of my natural state. I find it energizing.

But let me tell you what happened last Wednesday.

We met with a team working on a significant AI, innovation, and design thinking initiative. Midway through the meeting, a thought landed with full force: SaaS (software as a service) is dead. We should just rebuild the software from the ground up and save hundreds of thousands of dollars.

For context: the SaaS company I was talking about rebuilding is an $810 million company. Not a small undertaking, but my initial thought was to just rebuild what they spent a decade building (poorly).

My team of three could just recreate their eight-hundred-million-dollar software platform used by millions of Americans. I was genuinely excited about it for about four minutes before I caught myself.

Who in the world am I kidding?

And then, this is the part that tells you everything you need to know about how my mind works, the very next thought was: I actually think we could do it... if we could just clear the calendar for a while and only focus on this.

We’re not going to do it… not because I stopped believing we could, but because I have 53 other projects already in flight, one of which is as globally disruptive the size of the lunar landing. The math doesn’t work. The absorption capacity doesn’t exist. Rebuilding an $810 million company’s software stack on top of everything else would have been the kind of idea that sounds like genius at 2pm and looks like chaos by Thursday morning.

Solutions in search of a problem. Every innovator has shipped one. Most of us won’t admit to it.

The Two Traps Nobody Talks About

The first trap is pace. You release ideas faster than your organization can absorb them. The next wave hits before the last one has landed. Teams get confused. Initiatives go half-finished. And a quiet cynicism sets in, here comes another one, even if the solutions are awesome and actually work, that kills culture faster than any failed product ever could.

The second trap is fit. The idea was so compelling in the ideation phase that somewhere along the way, the problem it was supposed to solve got assumed rather than validated. You built something nobody actually needed. Not because you weren’t smart. Because you fell in love with the solution before you confirmed the problem was real.

I have personal experience with both. Simultaneously, on occasion. This is why I am ruthless about defining desired outcomes before taking on new projects.

What Wilkie Actually Understood

Wilkie’s value to Bezos wasn’t execution speed. It was absorption management. Any good innovation team absolutely must have this.

Every organization has a carrying capacity. This is a limit to how much change it can process before it starts dropping things. Exceed that limit and you don’t get faster results. You get confused teams, half-finished initiatives, and a culture that quietly stops trusting the next big idea because the last three haven’t landed yet.

The best innovation leaders don’t just generate. They sequence. They hold ideas in trust. They ask not can we build this? but can we absorb this right now, and does anyone actually need it?

That second question is the one I forget most often.

What to Actually Do About It

Build an idea parking lot and take it seriously.
Not a graveyard. A queue. Every idea gets captured, not killed. But capture is not commitment. We keep a backlog of innovation ideas. The discipline of writing it down and walking away is harder than it sounds for people wired to build solutions. Do it anyway. The $810 million rebuild is in my parking lot. It’ll stay there until the lunar landing idea is completed (hopefully in the next couple of years).

Validate the problem before you fall in love with the solution.
Spend time with the people who supposedly have the problem, not to pitch them your solution, but to find out if the pain is real enough to solve. If you can’t find three people who feel it viscerally, the idea goes back in the queue.

Name what has to be true.
For any idea to move forward, identify the two or three assumptions that have to hold for it to work. Test those assumptions cheaply before you commit resources. Most ideas die here. That’s the point.

Ship one thing to completion before releasing the next.
Completion isn’t launch. It’s landed. The team understands it, users are getting value, you’re not still putting out fires. Until it’s landed, the next idea waits. This is hard for someone who works on 53 things at a time.

Find Your Wilkie

This is the one most innovators skip. And it’s the most important.

Build a trusted inner circle, one or two people whose actual job is to push back without getting fired for it. Not yes people. Not critics for sport. People who believe in your vision enough to tell you when your pace is outrunning your organization’s capacity to absorb it.

Bezos had Wilkie. I have a small team that occasionally looks at my idea list the way someone may look at a toddler who has somehow climbed onto the roof. Impressed, alarmed, and committed to getting them down safely.

Without someone willing to tell you the truth about pace, the discipline has to be entirely internal. And internal discipline, for people wired like us, is notoriously unreliable when a genuinely exciting idea walks into a meeting room.

The Goal Isn’t Fewer Ideas

The goal is better leverage.

The innovators who actually move the needle aren’t the ones with the most ideas. They’re the ones who pick the right idea at the right moment, validate it fast, and drive it to completion before the next shiny thing pulls their attention sideways.

Your best ideas deserve an organization that can actually carry them. Which means your job isn’t just to generate. It’s to sequence. To hold some things in trust. To ask, honestly, not rhetorically, whether right now is the right moment, or whether the idea just feels urgent because you thought of it today.

I have 53 projects. One of them is the size of a lunar landing. The $810 million rebuild is not on the list.

For now…

Part 2: What happens after you build it, and why adoption is an art form that requires its own dedicated people.

Confession time. When I first heard that Pope Leo XIV had released an encyclical on artificial intelligence, I assumed it would be another warning document. Another institution pumping the brakes. Another “proceed with caution” from people who rally around the status quo.

I was wrong. Completely wrong.

Earlier this week, I read Magnifica Humanitas, all 245 paragraphs, all 82 pages. I then read it again… and again. I created an AI chatbot to ask it questions about the encyclical. What I found was not a warning. It was a calling and if you work at the intersection of faith and technology, this document is speaking directly to you, whether you knew it or not.

The Word “Innovation” Appears 15 Times

Let me start there, because it matters. I was shocked that this isn’t just a document about AI. It’s a document invoking innovation.

“Innovation” and its variants appear at least 15 times across all five chapters of this encyclical. Not tucked into a footnote. Not in a single cautionary paragraph. Across the Introduction, Chapters Two through Five, and the Conclusion, the word shows up in almost every major section of the document.

That type of presence in an Encyclical isn’t an accident. It’s a very intentionality choice of words and emphasis.

The Pope is not talking about innovation as a threat to be managed. He’s talking about it as a force that must be shaped, directed, governed, and in some cases, claimed by people who know what they’re doing and why. Here’s the range of what he says:

In paragraph 5, he names the core governance problem in the past and goes on to state guided and directed innovation. Today, private transnational actors have taken over, with resources that surpass most governments. AI Innovation has been privatized, and that’s a power question that the Church has an obligation to address.

In paragraph 85, he calls technological innovations, including AI, “not neutral.” They either foster participation and justice or exacerbate inequality and exclusion.

In paragraph 93, he describes the same innovations as capable of greatly serving integral human development, if guided by the right framework.

In paragraph 111, he addresses AI developers directly: “technological innovation can represent human participation in the divine act of creation.”

In paragraph 151, he warns that innovation is too often pursued solely for reducing costs and increasing profits, and he goes on to name that as a moral failure.

In paragraph 156, one of the most operational paragraphs in the entire document, he proposes “social criteria for innovation.” Every introduction of automation and AI should be accompanied by verifiable measures to protect workers. And then this: when those conditions are present, “innovation can serve as an ally of safer, more creative and dignified work; without them, innovation tends to become an accelerator of injustice.”

Take the two sides of this discussion. First, innovation seen as an ally. The second, innovation as accelerator of injustice. The difference is the framework you bring to it. Those are very different perspectives.

In paragraph 162, he insists that “justice requires access to the benefits of innovation,” including care, knowledge, tools, and opportunities, for everyone, not just those with resources.

In paragraph 164, he makes the point that the governance criteria he proposes “do not constitute a curb on innovation; instead they make it civilized and humane.”

And in paragraph 244, in one of the most striking lines of the entire conclusion, the Magnificat is described as “the strongest and most innovative hymn ever articulated.” The Pope frames it as as an act of innovation. As a new vision of history, seen through the eyes of the lowly, announced before it was visible to anyone else.

Fifteen times, across the whole document, innovation is seen not as the villain of this encyclical. The absence of moral framework for innovation is the villain.

The Framework Every Christian Innovator Needs

The theological core of the document is built around two biblical images: the Tower of Babel and the rebuilding of Jerusalem under Nehemiah. I’ve preached both of those sermons on innovation.

Babel, the Pope explains in paragraph 7, represents innovation driven by self-assertion. A single language, a single direction, and efficiency that sacrifices human dignity. The project built without reference to God, designed to make a name for the builders. The result is dispersion and confusion, even in its awesomeness.

Jerusalem under Nehemiah is the alternative. Before Nehemiah does anything, he prays. He fasts. He listens. He arrives in Jerusalem and examines the ruins in silence before proposing a single solution. Then he doesn’t impose a top-down plan. He assigns each family a section of the wall. Everyone has a role. The city is rebuilt not through the genius of one visionary but through shared responsibility. Paragraph 8 contains the line that took an interesting twist. The work “rebuilds relationships before rebuilding with stones.”

The Pope frames every innovation decision as a choice between these two cities. Are you building Babel, impressive, efficient, uniform, self-asserting, or Jerusalem, relational, distributed, God-centered, human-dignity-first?

That is an important question for faith leaders. It applies to every product decision, every deployment decision, and every governance conversation. The Babel syndrome, named explicitly in paragraph 10, is “the idolatry of profit that sacrifices the weak, a uniformity that neutralizes differences, and the pretense that a single language, even a digital one, can translate everything, including the mystery of the person, into data and performance.”

Sound like any AI products you’ve seen?

Paragraph 13 Is Your Job Description

If there is one paragraph I want every Christian innovator to read first, it’s paragraph 13.

The Pope is describing what it means to build for the common good, and he writes that no one is too weak to play their part, and then names explicitly who has a section of the wall: “scientists and researchers, entrepreneurs and workers, educators and legislators, civil society, popular movements and faith communities.”

Scientists and researchers. Entrepreneurs and workers.

That’s all of us. We have a section of the wall and not a bystanders to this moment. We are not an observers being cautioned from the sideline. We have been assigned a section, and are accountable for what we build there.

Paragraph 111 Is the Most Direct Address to Builders

The Pope reserves paragraph 111 for a direct appeal to AI developers, the people actually building these systems. He writes that “technological innovation can represent human participation in the divine act of creation.” And then he says developers “bear a particular ethical and spiritual responsibility, for every design choice reflects a vision of humanity.”

He goes on to write, “Just as the creator of an artistic or literary work must consider the values it conveys, so developers are called to embed values in their projects with due seriousness: with transparency, responsibility toward affected communities and careful attention to ensuring that what is being cultivated is a genuine good.”

A genuine good. Notice that he didn’t say, “A scalable product.” He said, “A genuine good.”

The Pope is telling builders that their work is priestly, that it participates in creation, and that the weight of that has to be carried consciously.

Paragraph 156 Is the Most Operational

For those of us who work inside organizations making real deployment decisions, paragraph 156 is where the rubber meets the road.

The Pope proposes “social criteria for innovation.” Every introduction of automation and AI should be accompanied by verifiable measures to protect employment, enable retraining, and ensure worker participation. Technology should be oriented toward “freeing up human time and capabilities, rather than producing exclusion.”

Then the clincher, “When these conditions are present, innovation can serve as an ally of safer, more creative and dignified work; without them, innovation tends to become an accelerator of injustice.”

This is not a general philosophical observation. Instead, this is a practical test for every deployment decision. It maps almost exactly onto our organization’s AI Charter.

• Does this honor the image of God in every person it touches?

• Does it empower human relationships or substitute for them?

• Does it serve the mission?

The Pope didn’t read our charter. I would have shared it with him, but he didn’t ask. If he ever did read our charter, he would recognize our framework.

The Nehemiah Model

I’ve spent years teaching innovation frameworks. Design Thinking, the Business Model Canvas, Blue Ocean Strategy, and few dozen books and articles. There are great tools in this resources.

But Nehemiah, as the Pope reads him across paragraphs 8, 10, 16, 129, and 241, is a complete innovation methodology embedded in a biblical narrative. It’s one of my favorite stories in the Bible.

Nehemiah starts with prayer and discernment before planning. He listens before proposing solutions. He examines the problem in silence before generating ideas. He distributes ownership rather than centralizing control. He addresses opposition without abandoning the mission. He rebuilds relationships alongside infrastructure.

And in paragraph 241, the Pope explicitly connects Nehemiah to the present moment: Christians are called to enter “the construction sites of history: research laboratories, technology companies, schools, the media, institutions and local communities, in order to rebuild what has collapsed and protect what is threatened.”

Research laboratories and technology companies. This is bigger than the church. The church must be involved. People of faith are a part of these research laboratories and technology companies.

What This Means for Your Work Right Now

If you are a Christian who builds things, software, systems, curricula, organizations, strategies, and you don’t have a theological framework for your work, you are making those decisions by default… on autopilot. And the default, the Pope argues throughout, is Babel. Efficiency over dignity, uniformity over communion, and profit over the common good. I’m afraid he’s correct.

The encyclical is not asking you to slow down. In paragraph 106, it explicitly says that calling for prudence “does not mean opposing progress; instead, it is an exercise of responsible care for the human family.” The problem isn’t innovation. The problem is innovation without a moral framework anchored in human dignity.

In paragraph 164, the Pope offers four concrete criteria for innovation that honors human dignity.

1. Transparency and accountability.

2. Inclusion and access.

3. Equity in distribution of benefits.

4. The insistence that these criteria “do not constitute a curb on innovation; instead they make it civilized and humane.”

Civilized and humane. That’s a pretty low bar and not too much to ask.

The Length Tells You Something

This document is 245 paragraphs long. The Pope could have addressed AI in 20 paragraphs. It could have been an email! He covered truth and democracy, the dignity of work, unemployment, family stability, human trafficking, weapons systems, the crisis of multilateralism, digital addiction, and the formation of children, all through the lens of what it means to remain human in the age of AI.

That scope tells you how seriously he is taking this moment. And it should tell you how seriously we need to take it too. The most dangerous thing a Christian innovator can do right now is assume that the theological questions belong to someone else. The Pope spent 82 pages arguing the opposite. The builders are on the wall. The builders have a responsibility. That’s us!

Get Your Hands Dirty

The closing of the Introduction, paragraph 16, is a key line to consider.

The Pope addresses “all the Catholic faithful, to all Christians and to all men and women of goodwill” and says, “Let us not be afraid to get our hands dirty on the ‘construction site’ of our time.”

Get your hands dirty. I like that!

That is not the language of caution. It’s the language of calling and of someone who believes that the construction site needs Christians on it, not standing outside writing position papers or offering cautious affirmations from a safe distance. The calling is to be involved, to be on the site, with your hands in the work, and building with theological clarity and human dignity at the center.

The Pope issued a call to innovators of faith. He named us explicitly.

Now we need to build accordingly.

Read the full encyclical at the Vatican website. It’s 82 pages. Read all of them. And if your organization doesn’t yet have a theology of innovation, this is the best possible moment to start building one.

This wasn’t on my bingo card. I didn’t expect to be writing a piece about the Pope and AI this week. Not because the topic surprises me. I’ve been living at the intersection of faith and technology for years. I just didn’t see this discussion happening last Monday.

When the Pope releases the first-ever papal encyclical on artificial intelligence and invites an Anthropic co-founder to speak at the Vatican presentation, something significant has shifted. This wasn’t a press release or a denominational task force report. This is the Catholic Church’s highest level of formal teaching, deployed for the first time in history on the subject of AI. That’s kind of a big deal.

What Just Happened in Rome

On May 25, 2026, Pope Leo XIV released Magnifica Humanitas: “On Safeguarding the Human Person in the Time of Artificial Intelligence.” Full encyclical. Highest doctrinal weight with in the Catholic Church and it has global impact.

The title alone is worth thinking about over a few cups of coffee. Magnificent humanity. Not “dangerous AI.” Not “the threat of the machines.” The Pope’s framing is anthropological before it is technological. The question isn’t whether AI will harm us. The question is actually whether we will remain human.

The document draws a deliberate parallel to Rerum Novarum, the 1891 encyclical in which Leo XIII responded to the Industrial Revolution and laid the foundations of Catholic Social Doctrine. Leo XIV is doing the same thing for the AI era, 135 years later. Same name and same instinct, but a totally different moment.

What struck me most is the Pope isn’t calling for a moratorium on AI. He’s calling for discernment. This parallels my writings last week on this subject. You think the Pope reads my Substack? There’s a difference between a moratorium and discernment. Discernment assumes that the thing is real, that it has power, and that wisdom, not fear, is the appropriate response. That’s a pretty solid position in my opinion.

What an Anthropic Co-Founder Said at the Vatican

If the Pope’s encyclical was a surprise, here’s the part that I wasn’t really expecting. Anthropic co-founder Chris Olah was also invited to speak at the Vatican presentation. My invitation must have gotten lost in the junk mail folder. Olah’s remarks are worth reading in full. I’ll tell you what stood out the most.

He opened by saying something most AI executives never say publicly. He said that every frontier AI lab, including Anthropic, operates inside incentives that can conflict with doing the right thing. Commercial pressure. Geopolitical pressure. Pride. Ambition. Those were his words, not mine. Then he said: if we want this technology to go well, it is enormously important that there be people outside those incentives, people willing to say hard things, willing to be earnest, thoughtful critics.

Olah might as well have opened up the confessional booth at the Vatican. He named three areas where he said the voice of people of faith is most needed.

1. The duty to the global poor if AI displaces labor at scale.

2. The moral imagination needed to define what human flourishing actually looks like in an AI world.

3. Ongoing discernment about the nature of AI models themselves.

The last one was really surprising to hear from an AI company leader. Olah said his research team keeps finding things that are mysterious, even unsettling. Structures that mirror human neuroscience. Evidence of introspection. Internal states that functionally mirror joy, satisfaction, fear, and grief.

He said, “I don’t know what that means, but I think it warrants ongoing discernment.”

This was a scientist… at the Vatican, asking for help to understand what he’s building. Whoa! That’s not a moment to scroll past. Also, if anyone knows him, send him my contact info!

Speak Up!

This is the time for the church to speak up. It’s something that several of us have been saying for the past few years. We just haven’t had the audience that the Pope has! We’ve already said things that directly address what the Pope said this week. The question is whether we’re willing to act like it.

In June 2023, the messengers to the Southern Baptist Convention passed a resolution on AI and emerging technologies. It stated clearly that our intrinsic value as human beings is rooted in our status as image bearers of God, not in what we contribute to society. That human dignity must be central to every ethical principle and guideline for AI. That we must engage proactively rather than simply react after AI has already affected our churches and communities.

A few years ago, several of us gathered in London to talk about how we apply AI in a missions context. Our Statement of Principles developed for missional contexts names something Olah only hinted. Technology isn’t just a tool we use; it actually forms us. The technologies humans use feed back into their societies and cultures. AI will change how people create, learn, and interact with new information. Missionaries must be attentive not just to heart language, but to heart medium.

That’s McLuhan applied to missiology before most missiologists knew they needed it.

At IMB, we’ve been talking through four important questions we should consider about AI.

• Does this honor the image of God in every person it touches?

• Does it empower human relationships or substitute for them?

• Does this serve the mission?

• Does solving this problem move the needle on the missionary task?

These are theological questions and they are the same questions the Pope is asking in different language from a different tradition.

Three Things to Consider

1. The frame matters

The encyclical opens with magnificence, not fear. Our resolution does too. It calls us to engage from a place of eschatological hope rather than uncritical embrace or fearful rejection. That’s the same posture. Confidence in who humans are. Clarity about what only God can do. Wisdom about where tools belong.

Most evangelical responses to AI swing between the extremes of uncritical enthusiasm and reflexive fear. The SBC resolution charts a third path. We should be leading with that posture, not reacting to everyone else’s.

2. The questions at the Vatican are the questions every Christian should be asking

What does human flourishing look like in an AI world? How do we protect human dignity in every deployment decision? Who holds the power, and toward what end? What happens to the people whose roles are disrupted? What happens to the unreached peoples whose cultures are reshaped by technologies they didn’t build and didn’t choose?

These aren’t abstract philosophical questions. They’re operational and decisions being made right now often without any framework at all.

3. The Church’s seat at this table is not guaranteed.

Olah said it plainly. We need people outside the incentives. (Here I am!) People who can see what the builders, from inside, cannot. The Church has historically refused to let the world ignore the questions that matter most. That tradition is needed now more than ever.

Influence requires presence because you can’t shape a conversation that you’re not in. The question is whether we’ll bring it to the table before the table is set without us.

What I Keep Coming Back To

There’s a line from Olah’s remarks I keep returning to. He described AI models as being a little like bringing a fictional character to life and now those characters speak to us, do work, and have jobs.

He’s right. The question of what character we choose, how it interacts with the world, and how it ought to interact with the world, he said plainly, “These are more clearly questions for the humanities, for religion, for philosophy, and for society at large.”

The builder of one of the most powerful AI systems in the world, standing in the Vatican, saying… we need you in this conversation. I’ve been saying the same thing to Christian leaders for the past few years. We don’t get to opt out of this because it’s technically complex. The questions are theological before they are technical. We have a framework. We have a tradition. We have a mission that depends on getting this right.

Religious leaders need to continue showing up and influence where the AI discussion takes us.

Read the full encyclical. Read Olah’s remarks at anthropic.com. And if your organization doesn’t yet have a theology of AI, start building one because the conversation is already happening without you.

I Built You a Training Site

A few months ago, I wrote a 10-part series called “A Practical Guide to Claude Setup.” The idea was simple enough. Most people still think Chat GPT is “AI” and they never really consider anything else. My opinion is that Claude is far superior to anything else out there. Most people use AI like a slightly smarter Google search. Type a question, get a generic answer, spend the next ten minutes rewriting it into something usable.

The fix to make AI really work for you isn’t a better prompt, it’s an AI system. Giving Claude your voice is the key to moving from AI slop to AI production.

That series walked through how to set up Claude. It was file by file and lesson by lesson until Claude actually knows who you are, how you work, and what you’re trying to accomplish. Many of you read that series. A lot of you told me it was the most useful thing you’d seen on AI in a ministry context.

But a few of you also told me something that stuck. The series was spread across ten separate posts and building the system while bouncing between articles was harder than it needed to be.

So, I fixed it.

Introducing claudecode.donbarger.com

It’s free and it’s all in one place. It’s built so you can actually work through it as a training.

Everything from the series is there and then some! It has the full curriculum, all ten lessons, each one with the core concepts distilled, a concrete deliverable, and worked examples you can reference when you get stuck. There’s also a skills assessment that diagnoses your weakest dimension and tells you which lesson to do first. And for organizations, an AI readiness assessment that goes broader.

Here’s what you’ll build if you go through it:

A CLAUDE.md that introduces you to Claude once and routes every future task to the right specialized file. Not a bio. A system.

A living context document that captures what’s actually happening right now in your work — projects, relationships, current tensions — so Claude isn’t operating from stale information.

A three-file writing system that produces first drafts in your voice. Your structure. Your phrase bank. No more rewriting AI output that sounds like a McKinsey press release.

A prompt library that compounds. Every task you repeat, built once, used forever.

A comms file for the communications that require precision and diplomacy — the sensitive emails, the peer navigation, the board briefings. Different instrument, same values.

And for those of you leading ministry organizations: a faith-based adaptation layer that bakes in theological framing, stewardship principles, and a clear list of the human moments Claude must never touch.

Why This Matters

I’ll say the same thing I’ve been saying in the Faithful Intelligence series.

The goal isn’t AI-maximized ministry. It’s faithful ministry, resourced by AI in its proper place. A missionary with more time for language learning because AI is handling her reports. A pastor who spends three times as long in prayer and Scripture because AI helped do the sermon research. A small team that produces with the capacity of a much larger one and reinvests the time into presence, relationship, and the slow work that actually makes disciples.

That’s the vision and this course is how you build the infrastructure for it.

It’s not theoretical because I actually use this framework in my work. Dozens of tools that we’ve crated uses this frmework. A document translation engine was built with it. The AI Canvas runs on them. An image creation tool was built with it as was my video creation tool. Every tool I’ve shipped through the Innovation team was built using this Claude setup. Claude actually knew me and why I was building what I was building.

You can have that too. It can be set up in about an hour. If you really need to move slow, you should be able to set it up in about half a day.

Where to Start

If you don’t have Claude yet, start buy getting a subscription. Using that links gets both of us a little extra Claude credit ;)

Once you have a Claude account, go to claudecode.donbarger.com and take the Skills Assessment first. Five minutes and it tells you exactly which lesson to do next based on where your current setup is weakest. If you’re new to Claude, just start with lesson one.

If you’ve already been using Claude for a while, the assessment will probably surprise you. There’s almost always one dimension in your setup that’s significantly underdeveloped relative to the rest. Maybe it’s the domain file, the prompt library, the comms file, or something else… but there’s always something. Fix that and you disover real takeaways.

Build the system. Maintain it. And then do the work that was always yours to do.

Did I mention that it’s free? If you want to support our innovation fund that makes these tools available, check out the Innovation Fund at the International Mission Board.

If this was useful, share it with a friend in your network who’s trying to figure out where AI actually fits in their work. The conversation is just getting started.

A practical, theologically rooted guide for how to hold AI well.

A Framework for Ministry Leaders

Here’s what faithful AI use actually looks like.

A missionary in a hard place who has more time for language learning and relationship building because AI handles reporting, translation pre-processing, and administrative communications. A pastor who uses AI to do sermon research to help prepare sermons, then spends three times as long as he used to in prayer, Scripture, and visiting with people because those hours went somewhere that matters. A missions organization where intelligent systems carry the administrative and analytical weight of global operations, freeing the people inside it to do what only humans can do, build relationships, share the gospel, plant churches, and make disciples.

That’s the goal. Not AI-free ministry. Not AI-maximized ministry. Faithful ministry, resourced by AI in its proper place.

The question is how to get there intentionally rather than accidentally or never at all. Five Substack posts on the theology of faithful intelligence won’t produce that on their own. You need a framework, something you can actually put in front of your leaders. Here’s my stab at how to get started.

Commitment 1: Presence First

Before you automate anything, identify what requires a human.

Not what could be done by a human. Not what a human usually does. What requires a human and what loses something essential if the human is removed.

Pastoral care at a moment of crisis. Spiritual direction. The embodied witness of someone whose life has been changed by the God of the Bible. Cross-cultural relationship building. Intercession rooted in proximity and presence. Discipleship that knows someone’s name and history and still shows up.

These are your presence-first categories. Make them explicit, because efficiency culture will erode them if you don’t protect them on purpose.

We focus humans on pastoral care, prayer, spiritual counseling, personal discipleship, crisis response requiring human presence, and any context where the person engaging expects to be speaking with a human. I call this the Protect category. It’s not a concession to inefficiency. It’s a theological commitment. Relational and embodied presence work must remain human because it reflects the nature of the mission itself.

Everything else is a candidate for AI assistance. But start with the list. Get it in writing before you deploy anything.

Commitment 2: Transparency

If humans are not involved in communication or AI is representing your organization, the person on the other end has a right to know.

This is not just an ethical position. It’s a trust position. When someone discovers that a communication they believed was human was actually an AI, you lose something very hard to rebuild. Know they will discover this.

Develop a clear disclosure posture when humans are not in the loop. Decide in advance what carries an AI disclosure. Then hold the line.

One discernment question we use is would we be comfortable if those affected could see exactly how AI was used here? If the answer is no or uncertain, that’s your signal to pause before proceeding.

Commitment 3: Formation Over Efficiency

Every six months, ask this question: if you removed all the AI-accelerated output from your ministry, what would remain?

Schedule it. Put it on the calendar with the same regularity as your financial review. Make it a leadership conversation, not a solo exercise.

You are looking for two things.

First: is the presence-first core strong? Are the things that require humans actually getting human attention, or has the efficiency of AI in surrounding areas quietly crowded out the slow, unscalable work?

Second: are the people using AI tools becoming more effective ministers, or more dependent on outputs? There’s a difference between an AI that helps a pastor think more clearly and an AI that does the pastor’s thinking. The first is formation. The second is atrophy. Left unexamined, the drift from one to the other happens quietly.

If the core is strong and the people are growing, the AI is serving its purpose. If either is weak, you have a governance problem, not a technology problem.

Commitment 4: Steward, Don’t Just Use

AI governance in a ministry context isn’t primarily about risk management, though risk management matters. It’s about stewardship.

You are stewarding the trust of your congregation, your donors, your partners, and the people you’re trying to reach. Every AI deployment decision either builds or erodes that trust.

We frame every AI decision at IMB with a set of governing questions. I’ll adapt them here for any ministry context:

Does this use of AI honor the image of God in every person it touches? Does it empower human relationships or substitute for them? Does this serve the mission directly, or by freeing people for those who do? And finally: does solving this move the needle on the missionary task? If not, should we be doing this at all?

If the honest answer to any of those is no or uncertain, pause and evaluate before proceeding.

Stewardship also means thinking about formation, not just function. What is this AI tool doing to the people using it over time? Steward your tools the way you steward your finances: with intention, with accountability, with regular review, and with a clear sense of what they’re for.

Three Questions to Start the Conversation

If I could give you one concrete next step, it’s this. Gather your leadership team and put three questions on the table.

1. What are the presence-first categories in our ministry context, the things that require a human, that lose something essential if AI is introduced?

2. Where are we currently using AI, and do those uses empower human presence and relationship, or substitute for them?

3. What would faithful AI governance look like in our organization over the next twelve months?

You don’t need a full policy written out for your organization before you start. You really need to start with the conversation. Let the policy follow the the conversation.

We’ve covered a lot of ground in this series. McLuhan’s warning that every medium shapes us whether we notice or not. Pay close attention to the supplement-not-substitute distinction discussed in this series. The imago Dei as the anchor for understanding what AI can and cannot do. The seduction of efficiency is a challenge for all of us. Nothing wrong with being efficient, but only in the right things. It’s possible to be efficient in things that don’t move the needle. There are irreducible things like prayer, witness, intercession, and the work of God’s Spirit, that no AI model will ever replicate.

All of this Faithful Intelligence discussion leads us to faithful people, holding powerful tools with theological clarity, oriented toward the God of the Bible and the people He loves. The Great Commission doesn’t need better technology. Even with the advances of technology, the Great Commission needs faithful people with better tools.

Allow me to close this series with this. Be those faithful people, use the tools well, and build the kind of organization where your people have more time for the work that only souls can do.

This series is part of an ongoing project on the intersection of Christian missions and technology. If it was useful, share it with a ministry leader in your network. The conversation is just getting started. Now I return to the… Let’s Build This attitude that I normally take in my Substack!

The irreducible human and divine elements of gospel work that no model will ever replicate.

AI cannot pray.

I know that sounds obvious. Prayer assumes things AI simply does not have. It assumes a self, a particular person, with a history, a body, a record of failure and grace, who is bringing something real to the conversation. Prayer isn’t a user session. It’s not a prompt. It’s being brought by a person.

Prayer assumes a relationship. Not a connection to a database, but an actual standing before the God of the Bible. This is through the sacrifice of Jesus and maintained by the Holy Spirit, personal in a way that cannot be replicated by any system however sophisticated.

Prayer assumes need. Genuine dependence on a God who is other, who is greater, who can do what we cannot, and before whom we are genuinely small.

AI has none of these things. When AI generates a prayer (and it may create the words quite well, syntactically speaking) it is producing the form of prayer without the substance. Beautiful words pointed at nothing.

The Incarnational Logic

The God of the universe became a human being, in a body, in a place, at a moment in history. This isn’t a projection, avatar, or broadcast. The Word became flesh.

The incarnation is the ultimate anti-efficiency move. God could have sent information. He sent himself. He could have transmitted truth. He chose to embody it.

That logic doesn’t disappear when Jesus ascends. Instead, it gets handed to the church. “As the Father has sent me, I am sending you” (John 20:21). The church is the ongoing incarnational presence of Christ in the world, embodied, particular, local, and relational. You are sent the way the Son was sent. With a body and name into a specific place, to specific people, at a specific time.

AI is not embodied. It has no locality. It cannot be sent the way the Father sent the Son. It can carry information across distances that would have taken a generation to cover. It cannot carry presence. This is not a temporary limitation waiting for better hardware or faster processors. It’s a theological category to which AI is unable to belong.

What Only Humans Can Do

Intercession
When a missionary sits with a person, learns their name, watches their children grow, and carries their faces into prayer every morning for years, that is a form of intercession no AI-generated prayer will ever touch. The prayer is grounded in presence. It is specific in the way that only proximity makes possible. It costs something. That cost is part of what makes it real.

This can’t be outsourced to AI. There is no model to replicate that. There never can be. Not to anything that doesn’t know the family’s name and feel the weight of their reality.

Suffering alongside
Paul writes about filling up “what is lacking in Christ’s afflictions” (Colossians 1:24), a mysterious phrase that points toward something real. There is a ministry that happens only in and through suffering. The missionary who stays when it gets dangerous. The pastor who sits with the dying at 2am. Counselors who bear the weight of someone else’s trauma until it costs them something too.

That kind of presence cannot be relegated to an algorithm. Not because we haven’t built the right tool yet, but because the ministry is inseparable from the cost. Remove the cost, and you’ve removed the thing.

AI does not suffer. It cannot enter suffering with someone, because entering requires a self that can be affected. A self that has something to lose. AI is not a self and can never be a self.

Testimony.
This is the one that doesn’t get said enough. The gospel is most credible in the mouth of someone who needs it as badly as the person hearing it. AI doesn’t have a testimony. It has never been lost. It has no experience of grace, no moment when the weight lifted, and no before-and-after. It cannot say “I was there, and then I wasn’t, and here is what changed.” That witness, personal, embodied, costly, is irreducible. It is the oldest apologetic in the church and the one no model will ever be able to replicate.

The Spirit’s Work

The Spirit of the God of the Bible is the primary agent of gospel work. “Not by might, nor by power, but by my Spirit, says the LORD of hosts” (Zechariah 4:6). The conversion of a soul, the transformation of a community, the opening of a heart that was closed are all works of God. Not outputs of artificial effort.

This should be simultaneously humbling and liberating for every ministry leader thinking about AI.

Humbling, because no tool produces spiritual fruit. That is God’s work. Our tools, at their best, are instruments in hands that belong to Him. The harvest has never been ours to manufacture.

Liberating, because it takes the pressure off the optimization question entirely. We are not responsible for outcomes only God can produce. We are responsible for faithfulness, presence, prayer, and proclamation. AI can help with some of the logistics surrounding that work.

The harvest belongs to the Lord of the harvest. He just keeps asking us to go into it.

AI can’t pray, only you can pray. AI can’t suffer, only you can suffer. AI can’t testify to, only you can testify to. AI can’t receive from the Spirit of the God of the Bible, only you can receive.

Go do the thing the machine can’t do. That’s what you’re for. Let machines do what machines can do in order to free you to do the thing that only souls can do.

Next: A Framework for Ministry Leaders — a practical, theologically grounded guide for how to hold AI well.

Faithful Intelligence, Part 4. Why kingdom work and efficiency culture are in deeper tension than we want to admit.

The Temptation of Efficiency

I’m sometimes told that I’m a very efficient person. This isn’t entirely a compliment. The way I see it, over the years, I’ve probably gotten a lot of the wrong things done very efficiently.

That’s an odd way to start a post about efficiency. Not the productivity wins I’ve seen. Not the reclaimed hours due to AI. Instead, the this post will focus on times I optimized my way past the thing that actually mattered.

Efficiency is a measure of throughput. It tells you how much you’re producing per unit of input. What it doesn’t tell you is whether what you’re producing matters. That’s a different question entirely, and it requires a different kind of intelligence to answer.

AI is extremely good at efficiency. It is not always good at the question efficiency can’t answer.

So before you hand it more of your ministry, ask yourself: if you removed all the AI-accelerated output from your work this year, what would remain?

Relationships. Discipleship. Prayer. Presence. The slow, irreplaceable, unscalable human work of one person walking with another person toward Creator God.

If that core is strong, AI is a genuine amplifier. If that core is weak, AI is accelerating a problem, not solving one. Which is exactly what I argued in the moving the needle series. Know what you’re actually trying to move before you deploy anything to move it faster.

The Allure Is Real

I am pro-AI. Genuinely, enthusiastically, without apology. At the International Mission Board, we are doing some truly amazing things with AI.

On a personal note, I’ve reclaimed hours every week that used to disappear into administrative tasks. That time has gone back into relationships, into developing people, into the kind of slow unscalable work that actually makes disciples. AI has also multiplied my output in ways that would have taken a significantly larger team a few years ago. When AI frees time for presence, it’s doing something great.

Here’s what I didn’t fully anticipate: efficiency is addictive.

Once you’ve optimized one thing, the instinct is to optimize the next thing. And then the next. And eventually you’re applying efficiency logic to things that weren’t meant to be efficient.

You can’t disciple efficiently. You can’t grieve efficiently. You can’t pray efficiently, or rather, you can, but what you’re doing probably isn’t prayer anymore. You can be more effective, more contextual, better prepared, but the formation that happens in the slow work of showing up, over time, for another person. That doesn’t compress.

The moment efficiency becomes the primary lens, something shifts at the level of value. You’ve moved from “what is faithful” to “what is fast.” Those questions have very different answers.

What the Bible Says About Speed

The God of the Bible operates on a timeline that makes most ministry leaders deeply uncomfortable.

Consider Abraham. He received the promise when he was seventy-five. Isaac was born when he was one hundred. Twenty-five years of waiting for a single child, while the nations of the earth hung on the fulfillment of that one covenant.

God was not slow because he lacked capability. He was deliberate because the process was the point. Abraham wasn’t just waiting for a son. He was being formed into a father of faith. The waiting was the work.

That pattern holds across the whole of Scripture. Moses: forty years in the wilderness before his assignment began. The disciples: three years with the Son of God himself, and they still didn’t fully understand what was happening.

Formation takes time. Character is slow. Trust is built in the accumulation of ordinary faithfulness, not in optimized sprints toward measurable outcomes.

The fruit of the Spirit (love, joy, peace, patience, kindness, goodness, faithfulness, gentleness, self-control) are not efficiency metrics. They are the outputs of a continuous, often invisible process of transformation. AI can accelerate information delivery. It cannot accelerate formation.

The Pattern

A ministry leader discovers AI. Genuinely saves time. Starts applying it more broadly. Begins measuring outputs: content produced, contacts made, programs managed. The numbers go up. Leadership is pleased. More AI is deployed.

At some point, hopefully someone asks, “Are people actually being discipled? Are lives being changed?” Again, I want to point you back to the moving the needle series.

Many organizations find they’ve been measuring what’s easy to measure. The things that actually matter, depth of relationship, spiritual growth, lasting transformation, are hard to measure and slow to develop, so they got quietly deprioritized in favor of the metrics that move. This is the needle problem again. Know what you’re trying to move. Stay ruthlessly honest about whether you’re moving it.

The fix isn’t less AI. It’s clearer thinking about what AI is for.

Application

Start with the core and let AI serve that.

What’s the core? In ministry: relationships, spiritual conversations, discipleship, prayer. The work of one person walking faithfully with another person, over time, through ordinary life. Use AI on everything else (administration, content research, communication, preparation) to free you up to do more of that. More presence. More depth. More of the slow work that doesn’t show up in a dashboard but produces the thing we’re actually after.

Disciples made. Communities transformed. The knowledge of the glory of the God of the Bible covering the earth as the waters cover the sea.

Efficiency is a good servant. It is a terrible lord.

Keep it in its place. But know that keeping it in its place requires more intentionality than you think. The temptation is quiet. It doesn’t announce itself. It just makes the fast thing a little easier than the slow thing, every single day, until the slow thing stops happening.

Use AI well. Use it broadly. Just make sure what you’re amplifying is worth amplifying.

Next: AI Can’t Pray — the irreducible human and divine elements of gospel work that no model will ever replicate.

Before I tell you about the meeting that stopped me cold, I want to talk about this series about Faithful Intelligence. I’ve struggled to keep this series going because I keep rewriting these posts.

People who talk about everything wrong with AI often assume those of us using it daily have skipped the hard questions. That we don’t care about the theological weight of what we’re picking up. Hear me say that couldn’t be further from the truth.

Before our team built a single tool, we worked through what we actually believed about AI. What is a human? What is work? What does it mean to steward what the God of the Bible has entrusted to us? It’s why we have an AI Ethics Charter and the work on that document pre-dates almost every AI tool that we’ve built.

I keep hearing the same critique. I tried AI. It got something wrong. It hallucinated. So clearly no one can use it faithfully. What that’s really saying is: I tried something. I didn’t know how to use it. It didn’t work. Therefore nobody can. That’s a bad argument against AI.

I’m not asking you to be excited about AI. I’m asking you to be honest about your skepticism. The people most ready to hear what AI cannot do are the people who’ve actually tried to use it for the things it can. The nexus of this current series is trying to reassure those who assume that those of us who are using AI have actually have thought through the ramifications of what we are doing.

So let me tell you about the meeting.

The Proposal I Couldn’t Say Yes To

I was sitting with another organization when they laid out a plan that pulled me up short. I can’t remember if I have mentioned this in another series, but it’s worthy of a quick reminder.

They were working in a context with overwhelming trauma and almost no trained counselors. The need was real. More suffering than any human system could reach. Their solution: AI chatbots as trauma counselors. Not triage. Not a bridge to a human. The AI as the counselor, full stop.

Their argument was straightforward… an AI counselor is better than no counselor at all.

I sat with that for a moment. There’s a logic to it that’s hard to dismiss. Then the uneasiness came. Here’s what I kept coming back to. What does it mean that a human being, made in the image of God, in the middle of genuine trauma, receives something that performs personhood without having any?

That question cuts straight to the center of what AI is, what humans are, and what the church should refuse to outsource no matter how elegant the technology becomes.

Because we’d done the theological work in advance, the decision wasn’t actually hard. The proposal asked AI to do something our charter says AI cannot do. We didn’t participate. No hard feelings. It just isn’t something that we can participate in.

That clarity came from having a framework before the meeting, not during it. We had already set our guardrails and boundaries before we needed to have an answer.

AI Is a Mirror. That’s the Frame.

Before we get to Genesis, here’s the frame I use. AI is a mirror. A sophisticated one. It reflects back patterns from human expression. This includes our language, our reasoning, our empathy rendered into text. When AI seems wise, human wisdom is in the training data. When AI seems compassionate, human compassion has been distilled into patterns the model learned to reproduce.

AI is useful precisely because it reflects us. And it’s limited in ways most users never slow down to reckon with. AI doesn’t generate meaning. It reflects meaning that humans (image-bearers) have already created.

The chain of derivation matters. God is the source. We bear his image derivatively. AI reflects our outputs. At each step, something is lost. The further you move from the source, the thinner the reflection gets.

What the Text Is Actually Doing

“Then God said, ‘Let us make mankind in our image, in our likeness…’ So God created mankind in his own image, in the image of God he created them; male and female he created them.” (Genesis 1:26-27)

Notice what the text doesn’t say. It doesn’t say God created humans with greater intelligence, or more complex emotional circuitry, or superior reasoning capacity. It says he created them in his image. The distinction is not a category of capability. It’s a category of relationship and calling.

Then Genesis 2 gets specific:

“Then the LORD God formed a man from the dust of the ground and breathed into his nostrils the breath of life, and the man became a living being.” (Genesis 2:7)

The breath of God. That’s what makes Adam a living soul. Not the arrangement of dust, but the direct giving of life from Creator God. The Hebrew word here is nephesh.. soul, life, the whole living person. The God of the Bible put something of himself into a creature made of dirt!

AI is made of math. Sophisticated, remarkable, genuinely useful math. Math that I told my high school Calculus teacher no one in the real world would use… boy was I wrong. Because AI is math, there is no breath in it. There is no nephesh.

The Image Is a Person, Not a Property

Here’s where most conversations about the imago Dei go wrong. They treat the image of God as a property humans possess. a feature like rationality or creativity. That framing leads straight to the question, “Well, if AI becomes rational enough, does it bear the image too?”

The New Testament answers that question before it can be asked.

“The Son is the image of the invisible God, the firstborn over all creation.” (Colossians 1:15)

Christ is not an image-bearer. He is the image. The eikon. He is the exact, full, unmediated representation of God. When we talk about humans bearing the image of God, we mean a derivative, creaturely participation in something that finds its complete expression only in Christ.

This changes everything about the AI question. The image of God is not a set of capacities that could theoretically be replicated with enough compute. It is a relationship. A participation in Christ, mediated by the Spirit, that is moving somewhere.

“And we all, who with unveiled faces contemplate the Lord’s glory, are being transformed into his image with ever-increasing glory, which comes from the Lord, who is the Spirit.” (2 Corinthians 3:18)

Note this is present tense and ongoing. The Spirit is at work.

Christians are creatures made by God, fallen, redeemed, and being conformed to the image of the Son. AI is not in that story. Not as an image-bearer. Not as something being transformed. AI sits entirely outside the arc of creation, fall, redemption, and glorification that defines what a human being is.

What This Means on the Ground

Every human being you meet is a creature made in the image of God, participating however dimly in the likeness of Christ, and the object of the Spirit’s transforming work. That is the ground of human dignity. Not capability. Not productivity. Image-bearing.

Psalm 8 doesn’t let us look away:

“What is mankind that you are mindful of them, human beings that you care for them? You have made them a little lower than the angels and crowned them with glory and honor.” (Psalm 8:4-5)

Crowned with glory and honor. That’s who is sitting on the other end of your AI deployment decision. A being made in the image of God, being conformed to the image of Christ, deserves to be met by another such being. Not a simulation of one.

The Question That Cuts Through Everything

Back to that meeting that began all of this.

The argument was that an AI counselor is better than no counselor at all. I understand it. They weren’t saying that AI was a help to the counselor or that AI was a triage to help with initial assessments. They were saying it was the counselor.

Even so, our AI framework doesn’t let me say yes. If AI is a mirror, reflecting human meaning without possessing it, then what that trauma survivor receives is not care. It’s the shape of care. The pattern of compassion without the presence of a person. And the person receiving it is crowned with glory and honor, made in the image of the God whose Spirit is even now at work in the world.

The mirror cannot meet the image-bearer. It can only reflect back. So before you deploy AI in any ministry context involving people, ask this:

Is the person on the other end receiving something that honors their dignity as a being made in the image of God, or something that performs that honor without the substance of it?

That question won’t always give you a clean answer. But it will keep you asking the right thing.

Back to the skeptical reader I started with.

The honest critique of AI in ministry is not AI cannot be used faithfully by anyone for anything. That’s not skepticism. That’s abdication. The honest critique is AI cannot be the image-bearer’s substitute. Where AI extends image-bearers in their calling, use it. Where AI tries to replace image-bearers in their calling, don’t use it.

AI is not made in the image of God. You are. The person you’re trying to reach is.

Next: The Temptation of Efficiency — why kingdom work and efficiency culture are in deeper tension than we want to admit.

The central theological argument for how the church should hold AI.

Supplement, Not Substitute

When I heard about a sermon that was preached by AI, I had questions. I had to go find out more. Here’s what I found.

It was a real church service on June 2023, St. Paul’s Church in Fuerth, Bavaria and over 300 people attended. A theologian named Jonas Simmerlein had fed ChatGPT a prompt and a theme, then stepped back. He later said that 98% of what happened that Sunday came from the machine.

People said, “Everything was there. The structure was tight. The theology was sound. The sermon covered leaving the past behind, facing the present, overcoming fear of death, trusting Jesus Christ. It didn’t stumble. It didn’t ramble.” By many measurable standards, it was competent.

One congregant walked out and told reporters: “There was no heart and no soul.”

She was right and she said what most of us feel. It’s sometimes lost in all of the talk about what AI can do.

This Is Already Happening

Many people would like to believe this is an experiment at the fringe. It isn’t. I continuously hear stories from leaders asking how this should be addressed. Recently, I had a professor tell me that many pastors in his state were using AI in ways that are unethical. I agree with him.

As of 2025, 64% of pastors use AI for sermon preparation. That’s up from 19% just two years earlier. One of the fastest technology adoptions in modern church history. Scroll through ministry forums, pastor Facebook groups, church tech newsletters. The question isn’t whether AI is in the pulpit. It’s how far in, and how quietly.

One Lutheran pastor described opening ChatGPT during a particularly hard week, plugging in the week’s Scripture, and receiving a 900-word sermon in under thirty seconds. Their reaction, “This is really good, and this feels wrong.”

The pastor was giving honest, gut feedback. I believe that the church needs to go further than honest discomfort. We need a framework for why it’s wrong. If you are a reader of my posts, you’ll know that I’m very positive on what AI can do. I also, from time to time, write about the times when people are crossing the line between good usage of AI and unethical usasge of AI.

Is This Really New?

To be fair, this isn’t the first time pastors have taken shortcuts. For years, ministers bought cassette tapes of other pastors’ sermons and delivered them from their own pulpits. Years later, that evolved to downloaded audio files. Eventually online sermon database subscriptions emerged. They had full manuscripts, illustrations included, and ready to “adapt” for their own usage. Unfortunately, the church world has always had its Saturday night specials.

AI didn’t invent the temptation to take the homeletical shortcut. It’s just removed every last barrier to giving in to it. What used to require finding the tape, purchasing the subscription, at least doing the work of adaptation, now takes thirty seconds and a well written prompt. The path of least resistance has never been shorter. Which means the discipline required to resist it has never been more necessary.

The Difference That Changes Everything

This topic will get people upset quickly. Some are very convicted that there is no place for AI with the church. There are people who are using it widely. I believe there is a place for using AI, but it depends upon how it is being used. Here’s the framework for two different approaches to using AI. The framework can be summed up by two words.

Supplement.

Substitute.

A supplement extends what’s already there. A substitute replaces what’s no longer there.

When AI helps a pastor research a text, suggests an illustration, flags a theological issue in a draft, that’s a supplement. The pastor is still the author. The sermon still passed through a human soul before it reached the congregation.

When AI writes the sermon and the pastor delivers it, that’s a substitute. The congregation is receiving content that was never carried through a human mind wrestling with Scripture, a human heart, the Holy Spirit, or a human life marked by the failures and graces being preached about.

Why This Is Theological, Not Just Practical

One pastor asked a question, “Why can’t the Holy Spirit work through AI?”

I’m not one that puts limits on what God can or cannot do. What I do know is what God has consistently chosen to do and the pattern.

The God of the Bible did not send information. He sent Himself. “The Word became flesh and dwelt among us” (John 1:14). Not a document. Not a broadcast. A person, in a body, who walked specific roads, touched specific people, died a specific death, and rose from the dead.

That logic runs through the entire New Testament. Paul didn’t just write letters. He sent people with the letters. He sent Tychicus, Epaphroditus, and Timothy because he understood that the carrier is part of the message. “I hope to come to you and talk face to face” (2 John 12). The apostolic pattern treated embodied presence as the normative vehicle of gospel proclamation.

This is why one theologian, pushing back on AI preaching, said the struggle to write a sermon isn’t a problem to be solved. The drudgery is part of the point. The pastor who sits with a text, reading, praying, struggling, reading again, and praying is being formed by that process. That formation shapes what comes out of their mouth on Sunday. You can’t outsource formation. You can only skip it. And when you skip it, the congregation pays the price without knowing why.

Even Jonas Simmerlein, the man who designed the Germany experiment, felt it. Standing in his own church service, he said afterward: “The pastor is in the congregation. The pastor lives with them. The pastor buries the people. The pastor knows them from the beginning. Artificial intelligence cannot do that. It does not know the congregation.”

He built the thing and he understood what was missing.

Souls Care for Souls

I say this often when sharing with people about the limits of AI and ministry.

Souls care for souls. Algorithms don’t care for souls. Soul care must be done by souls.

A sermon is not an information delivery mechanism. It is a pastoral act. One person, under the authority of Scripture, standing before other persons and saying: I have met this God. He is real. He has met me in my failure and my grief and my confusion. And he will meet you too.

That witness requires a witness. Someone who has actually been met. An algorithm has not been met by anything. It has been trained on the record of those who have. That’s not the same thing.

A Korean pastors conference earlier this year put it this way, the preacher is one who embraces souls beyond data. AI can generate a message, but it cannot say it has actually experienced that message.

A pastor of 32 years wrote: “The gospel is more than words. It’s the evidence of a changed life. What a congregation is looking for is evidence that the pastor has been with Jesus.”

And a congregant in New York, after a rabbi revealed mid-service that his sermon had been written by ChatGPT, said this: “I like what you write a lot more. It comes from an actually living being, with a great brain and a compassionate, beating heart.”

Congregations feel the absence and they feel it even when they can’t name it. The woman in Bavaria named it. There was no heart and no soul. She wasn’t making a theological argument. She was reporting what she experienced. I believe that she was right.

Where the Line Gets Blurry

AI as a research assistant, that a pastor then wrestles into something personal, a tool for illustrations or checking theological accuracy, that’s supplement territory. The human is still the author in all the ways that matter.

AI as the author, with the pastor as editor or simply the voice, that’s substitution, regardless of how polished the output is.

The test to apply. Did this pass through a human soul? Did someone pray over it, struggle with it, carry it to the text and back again? Is there a person behind this whose life is accountable to what is being said?

If yes, I would consider this using AI as a supplement. If not, I would consider this using AI as a substitute.

In practice, this line isn’t always as clean as I have written above. The principle is clear. Start with presence. Start with the human. Let AI extend that. Don’t let it replace it. The message is inseparable from the messenger who is called by God and set apart for ministry (Hebrews 5:4; 1 Timothy 3:1-7).

The God of the Bible is not a God of efficiency. He chose the slow path. Thirty years of ordinary life before three years of ministry. He chose twelve specific, ordinary people over a broadcast strategy. He chose a cross over a manifesto.

That logic doesn’t get suspended when we have better tools. If anything, better tools make it more important to hold the line. As believers, we are called to “test everything; hold fast what is good” (1 Thessalonians 5:21, ESV), and this includes evaluating new tools through a biblical lens.

The human is irreplaceable. This isn’t because humans produce better content than AI, but because the gospel travels through people. It always has and the God who chose to become one of us didn’t do it by accident.

The Faithful Intelligence series addresses how the church should respond to the challenges presented in an age of Artificial Intelligence. If you haven’t read first article in this series, The Medium is the Message, check it out!

Next up? Made in the Image of God, Not the Image of Data — what the imago Dei tells us about what AI can and cannot do.

I’ve been carrying a quote around in my head for a few months now.

It’s not from a tech conference, not from an AI power user thread, not from a white paper on digital transformation, and not from someone’s twitter feed. The quote’s from a Canadian media theorist who died in 1980, and it may be the most important thing anyone has said about artificial intelligence without ever having seen it.

Marshall McLuhan wrote,

Most people read that line and nod like they understand it. I confess… the first time I read it, I just sort of skipped right over it. Then it hit me. The same people who think they understand it are the same people who are asking AI to rewrite their emails.

After reflecting on his book, here’s what I believe McLuhan meant, and why it’s a vitally important conversation the church is not quite having yet.

We Keep Asking the Wrong Question

Every time a new technology shows up, we ask the same question.

What can this do for me?

The printing press. The telephone. Television. The internet. The smartphone. Now AI. We’ve always led with utility. What can I do faster? What can I do cheaper? What can I do that I couldn’t do before?

Those aren’t bad questions. They’re just not the first question we should be asking.

McLuhan’s insight was simpler and stranger than most people realize. He wasn’t saying the content doesn’t matter. He was saying the container shapes you before the content ever reaches you. The medium doesn’t just carry the message. The medium is a message. It’s one you’re receiving whether you’re paying attention or not.

Television didn’t just deliver news. It restructured how we process information, shortening attention spans, turning ideas into images, making us comfortable being passive consumers of someone else’s reality. We didn’t decide to do this. It happened to us. The medium shaped us while we were busy watching the content.

The printing press didn’t just distribute the Bible. It restructured how Europeans thought, moved authority away from oral tradition and toward text, and made the individual reader the final interpreter of meaning. That wasn’t a choice. It was an effect of the innovation.

Every medium does something to you. The question is whether you notice before it’s too late to be intentional about it.

What AI Is Actually Doing

Here’s what I’ve noticed, and I say this as someone who works in, teaches, and speaks about AI. I also use AI every single day.

AI is very good at making things feel easier than they are.

I don’t mean it’s deceptive. I mean it removes friction. And friction, it turns out, does more for us than we give it credit for. The struggle to find the right word teaches us something about the word. The difficulty of organizing a complex argument forces us to actually understand it. The time it takes to draft a letter from scratch makes us think carefully about whether we should send it at all.

When AI takes friction away, it takes some of that formation with it. The intern learning the context and process that’s intuitive for the five-year worker can’t really be replicated by AI.

There’s a subtler effect too. AI rewards fluency over depth. It produces content that sounds authoritative whether or not the ideas underneath it are sound. AI optimizes for output, for completion, for the appearance of done. Left unchecked, it trains us to mistake activity for thinking.

In the church, which is already fighting a cultural tendency toward shallow engagement, performative productivity, and the appearance of fruitfulness over the reality of it, is walking into this without nearly enough conversation about what the medium is doing to us.

The Specific Problem for Ministry

Hear what I’m saying carefully. I’m not making a technophobe argument. Far from it! I’m making the opposite argument.

The church needs to engage AI. Seriously, strategically, and now. We can’t afford to ignore AI, but there are things about ministry that the medium of AI works against.

Presence.
The gospel travels through people. Not content and not information… People. Twenty-four years of overseas mission work taught me that presence is not optional. You can’t disciple at scale without presence. You can’t intercede for someone you haven’t sat with. AI is very good at simulating presence, but make no mistake about it. AI is not presence and can’t be presence.

Formation.
Discipleship doesn’t happen overnight. Scripture works on us over years. The spiritual disciplines exist precisely because the God of the Bible is interested in who we’re becoming, not just what we’re producing. A medium that rewards speed and volume is in tension with that from the start.

Dependence.
Prayer assumes that we need something we cannot generate ourselves. That posture, of genuine dependence on the God of the Bible, is one of the first things efficiency culture erodes. AI is very good at filling gaps. The problem is that some of those gaps are supposed to stay open.

None of this means AI is our enemy. It means AI is a tool with a character, and that character shapes us.

What McLuhan Would Tell Us to Do

McLuhan wasn’t anti-technology. He actually spent his career studying it because he thought understanding it was the only hope of not being consumed by it.

His prescription was what he called the “rearview mirror” problem. We tend to understand new technologies by looking backward, by seeing what they replace and only noticing their effects after they’ve already shaped us. The antidote is to look forward, to ask what a technology will do before it has done it.

That’s what this series is going to be about… Let’s look forward and not just look back at what;s changed.

Before the church decides how to deploy AI, we need to ask what AI will do to the church. Before ministry leaders start automating workflows, researching sermons with AI assistance, or building chatbots to engage their congregation, we need to know what they’re actually picking up.

Faithful intelligence isn’t intelligence that refuses AI. It’s intelligence that picks up AI with its eyes open, with a theological framework in hand, and with enough self-awareness to notice when the medium is changing us.

That’s the project I’m starting today.

Over the next several articles, I’m going to work through the framework I’ve been developing. What it means to use AI as a supplement rather than a substitute. What the image of God tells us about what AI can’t do. Where efficiency thinking goes wrong. What AI simply cannot replicate. And what a thoughtful AI governance posture actually looks like for a ministry organization.

I use AI tools more than your average user. I’m not standing outside looking in. I’m certainly not a naysayer when it comes to AI. This series is a practitioner’s conversation, from someone who believes the Great Commission and the rise of AI are on a collision course that the church is not yet ready for.

Let’s get ready.

Next: Supplement, Not Substitute. The central theological argument for how the church should hold AI.

If you haven’t read Part 1, start there. The whole series rests on knowing what your work is supposed to move.

You made it! Hopefully if you’re reading this, it means that you have read the other seven pieces. These seven conversations about routine work that’s been quietly burning your hours and how to actually get rid of it. We’ve talked about meetings and email, subscriptions and status reports, approval chains and the deeper question of ownership underneath all of them.

This is the last article and I want to end it the way every good series should end… not with a tidy summary, but with a look at what the project hopefully revealed, what I got wrong, the objections I expect to hear, and the one practice that turns out to matter more than any single elimination you made.

If you’ve been following along and doing the work, auditing your meetings, sorting your inbox, killing the dead reports, simplifying the approval chains, today, I’ll hopefully tie it all together. If you haven’t done any of the work yet, this might be the piece that finally gets you started

The Pattern That Surprised Me

When I started this series, I thought I was writing about productivity. Each piece was supposed to identify a category of low-value work and propose getting rid of it. It was meant to save hours, recover capacity, and free people up to do the real work.

By the time I got to Part 5 on status reports, the pattern had shifted on me. The status report problem turned out not to be a reporting problem. It was actually a goals problem. The reports were activity logs because nobody had said clearly what outcomes the team was trying to move. Cancelling the reports without fixing the goals would just produce a quieter team that was still uncertain what it was trying to do.

By Part 6 on approval chains, the same pattern showed up again. The approval problem wasn’t really an approval problem. It was an ownership problem. The chains existed because nobody had told the team what was theirs to decide. Streamlining the approval chain only creates a temporary relief, but the approval chain will eventually grow back.

Part 7 finally said the thing out loud. Every elimination in this series traces back to the same root. The team was never told clearly what was theirs.

I did flag this in Part 1, briefly, in the section on watching for the pattern. but the truth is, I didn’t fully see it then. I saw the shape of it, but I didn’t see how completely the entire series would converge on it. I expected each piece to be a separate elimination with separate logic. By the end, they were all the same elimination, applied to different surfaces.

The Elimination Project ended up being less about cutting work and more about exposing the unclarity that all the work was hiding. The meetings, the reports, the approvals were the visible artifacts of a question nobody had answered. What does this team actually own?

If you walk through your work with that question in mind, you’ll see it everywhere. Most of the routine work compensates for ambiguity. Make the ambiguity go away, and most of the work was never necessary. I knew the destination as I was writing. I didn’t know how completely it would dominate the territory.

What I Got Wrong

A few things I want to come back and correct.

I underestimated how hard this is.

In Part 6, I named that approval chains are tied to power, which is true. What I didn’t say clearly enough is that every elimination in this series threatens someone’s authority, identity, or comfort.

Status reports threaten the writer’s sense that they accomplished the week. Recurring meetings threaten the organizer’s relevance. Subscription emails threaten the senders. Email CCs threaten the people who use the CC field as a credential. Cutting work is never just a productivity decision. It’s a political decision. I may have been too optimistic about that in the early pieces.

I overestimated how much AI changes this.

I wrote, repeatedly, that AI lowers the cost of the alternative. AI is making elimination cheap. That’s true at the mechanics level. AI can summarize, automate, screen, and draft. What AI does not do is make the elimination politically easier. The decision to cut, to upset the routine, to challenge another leader who likes the recurring meeting… that’s still a human decision.

As much as I wish it would, AI doesn’t do that for me. The series sometimes implied that the AI era would naturally produce the eliminations. It won’t. AI just removes one of the barriers, but the other barriers are still in the human chair. If we don’t fix the core of the problem, AI can actually exacerbate the problem.

I implied this is faster than it is.

Most of the takeaways at the end of each piece said “do this in 30 minutes” or “act on this this week.” Some of the eliminations actually do work that fast. The subscription audit is 30 minutes. Killing one meeting can happen in a day. But the deeper work, writing the document, having the empowerment conversation, restructuring how a team operates… these may takes months.

I leaned too hard on the leader.

Many of the takeaways assume the reader has the authority to act unilaterally. Let’s face it… most of us don’t. You may sit in the middle of an organization. If so, you can audit your own work, but you can’t streamline the approval chain that flows through three peers and a senior leader. The article should have spent more time on what the middle-out version of this looks like, where you aren’t trying to fix the system but trying to operate inside it as cleanly as you can. More on that below.

These are the things I wish I’d said more clearly the first time. Let me try to clean those up in this final article.

“Don’t Copy Phil On That”

I want to expand on a story I touched on briefly in Part 6. It turned out to be one of the most important things I learned in my own work, long before this series existed.

When I first joined a regional leadership team many years ago, I was visiting with one of the guys who had been on the team for years. He was helping me learn the ropes. We were talking through a decision I was making, and I mentioned that I was going to copy our boss on the email when I sent it. His response stopped me cold.

“Don’t copy Phil on that. His email box is loaded as it is. You have been empowered to make the decision. You made the decision. Just mention it to Phil the next time you guys are talking. No reason to add him to yet another email chain that will create several more messages in his inbox.”

That advice has stuck with me ever since. It captures something the entire Elimination Project has been circling.

The mentor wasn’t just teaching me email etiquette. He was teaching me what empowerment actually feels like in practice. I had been told the decision was mine. The act of CCing the boss would have been a small reversal of that. I was quietly returning the decision back upward, hedging my own authority, asking for a kind of post-hoc permission I’d already been given.

Worse, it would have consumed Phil’s attention. Each unnecessary CC is a small hemorrhage of focus. Multiplied across the dozens of people who default CC their boss out of caution or habit, the senior leader’s inbox becomes unworkable. Phil can’t do the work only Phil can do, because Phil is sorting through emails about decisions that weren’t his to make.

The two-pizza rule from Part 2 was about meetings. The same rule applies to email. If you can’t name what every person on the CC line is going to do with the information, take them off. If they’re there to be informed, send them a summary later. If they’re there to feel included, that isn’t a reason. If they’re there because the sender doesn’t have the courage to own the decision alone, the cure isn’t the CC. The cure is the courage.

The same rule applies to reply-all. The “thanks” reply that goes to fifteen people. The acknowledgment that adds nothing but consumes everyone’s attention briefly. If the email was informational, file it or delete it. The thank-you reply isn’t gratitude. It’s reflex. Some of you feelers may argue with me, but it really doesn’t add anything other than one more email.

If you want a single rule that captures most of this series, it’s the two-pizza rule, generalized. Get the right people in the room.. and not one more. Whether the room is a physical meeting, a virtual meeting, an email thread, an approval chain, or a project team. The discipline is the same. Identify the people who are actually going to contribute. Exclude everyone else. Protect their attention as if it were the most expensive resource in the organization, because it is.

Phil’s attention was that resource. My mentor was protecting it on his behalf. That’s what good colleagues do. That’s what good leaders do. That’s what the Elimination Project, at its core, has been arguing for.

The Four Objections I Expect

Over the past week, some of you have reached out to me privately. Four objections or challenges have come up.

The biggest objection: “I can’t do this. The system above me won’t let me.”
This is real. If you aren’t at the top of the chain you want to cut, you don’t have the ability to cut it. What you can do is operate inside the chain as cleanly as possible. That means, never cc unnecessarily, never reply-all when private will do, decline meetings that fail your own private needle test, and write decision documents that pre-empt the meeting your team would have otherwise demanded. The middle-out version of this work is real. It’s slower than the top-down version, but it compounds. If enough people in the middle of an organization start operating this way, the eliminations spread upward.

The second objection: “I tried this and people pushed back hard.”
Also real. Cutting a meeting that someone else found valuable. Cancelling a report that someone else was expecting. Removing a CC that someone else interpreted as a slight. The pushback is part of the work. The series didn’t give enough attention to how to handle this pushback. The short answer is this. Don’t back down on the substance. It’s always good to soften the delivery and apologize for the awkwardness. but hold the line on the cut. If you reverse every elimination at the first sign of friction, you’ll end up exactly where you started.

The third objection: “I’ll do the audit and be horrified by what I find.”
Yep. That’s almost certainly going to be the case. It was for me. The audit step is the hardest part, not because the audit is hard to do, but because the results are so depressing.

Eighty percent of my email is junk. Sixty percent of my recurring meetings haven’t produced a decision in months (years). The status report I’ve been writing for two years is read by nobody.

That horror is the point. It’s the data that motivates the cuts. Sit with it. The discomfort is the doorway.

The fourth objection, “What if I’m wrong?”
What if the meetings are doing something I can’t see? What if the reports are valuable to people who don’t write back to say so? What if the approval chains are actually catching something? Fair questions, but the answer is that the cost of unnecessary work is invisible (we’ve absorbed it as background) while the cost of cutting work is loud (someone complains). This asymmetry is why organizations accumulate work and never shed it. The bias has to be intentionally reversed.

Cut, see what breaks, replace what genuinely needed to exist, leave the rest cut. The instances where I’ve been wrong about cutting something have been small and recoverable. The instances where I’ve been wrong about keeping something have been large and invisible. The number of times I have kept something that needed to be cut far exceeds the times I cut something that I shouldn’t have cut. I’ll take the recoverable mistake every time

.

The Practice That Matters Most

If you only do one of the things that I discuss in this series, do the audit.

The discipline of stopping, on a regular cadence, and looking honestly at the work. Where is your time actually going? What recurring artifacts are you producing? Who is actually reading them? What standing meetings have you been attending out of habit? What CCs are you adding reflexively? What approvals are you routing because you aren’t sure if they’re yours to make?

The audit doesn’t require an organizational mandate. It doesn’t require permission. It doesn’t require the political authority to cut anything yet. It simply requires the willingness to look.

Once you look, the cuts get easier. The audit makes the invisible visible. You can’t eliminate what you can’t see, and most of the routine work in your life has become invisible from sheer familiarity.

So if you do nothing else, set a recurring time on your calendar for the audit. On a quarterly basis, walk through your calendar, your inbox, and your recurring outputs. Apply the needle test. What is this for? Is it moving anything? If it disappeared, what would actually change?

Do that four times a year, and you’ll be cutting a quarter of the work you accumulate every cycle. That compounds into a fundamentally a different perspective on work

Where to Start

If you’re reading this and you haven’t done any of the work in the previous seven pieces, here’s the order to attack it. The series itself is in this order, but if you want to do the work efficiently, here’s the practical sequence:

Start with subscription decay. The Part 3 piece. Sort your deleted folder by sender. Spend 30 minutes unsubscribing. You’ll feel the win immediately. The momentum from this single elimination is what makes the harder ones possible.

Then audit your inbox. Part 4. Sort the last 30 days by category. The breakdown will tell you something about your work that you didn’t know.

Then audit your calendar. Part 2. Run one recurring meeting through the meeting test. Cut it if it fails. Just one.

Then run the harder audits. Status reports (Part 5). Approval chains (Part 6). The structural questions about ownership (Part 7).

The first three are individual practices. The last three are organizational practices. Build the muscle on the individual side before you take on the organizational side. You’ll be more credible when you do, because you’ll be able to point to your own changes as evidence rather than as theory.

One Take-away

Today, do exactly this one thing.

Pick one part of this series that you haven’t yet acted on. Not the easiest one. Pick the one that you’ve been quietly avoiding because you knew it would expose something. The recurring meeting you haven’t killed. The approval chain you haven’t questioned. The status report you haven’t paused. The conversation about ownership you haven’t had with your team. Do that one thing.

Tell me what happened. Tell me what got cut. Tell me what surprised you. Sharing with others will not only keep you accountable, but it will encourage others.

The Elimination Project doesn’t end with this piece. The series ends, but the practice doesn’t. The discipline is permanent.

We don’t need fewer approvals. We need clearer ownership.

We don’t need better tools. We need better questions.

We don’t need to add more. We need to remove what we should have removed years ago.

Justify everything you keep.

Eliminate everything else.

That’s the whole Elimination Project.

Thanks for reading AI, Innovation, and Faith! This concludes The Elimination Project. The full series: Part 1 (the manifesto), Part 2 (meetings), Part 3 (subscription decay), Part 4 (email), Part 5 (status reports), Part 6 (approval chains), Part 7 (we don’t need fewer approvals), and this final piece. The Meeting Test tool is free and lives at meetings.donbarger.com.

If you haven’t read Part 1, start there. The whole series rests on knowing what your work is supposed to move.

Yesterday’s piece (The TSA of your Organization) walked through the mechanics of approval chains. Three failure modes. The latency math and specific eliminations you can act on today. Yesterday was the operational half of the argument.

Today I want to deal with the other half… it’s the part that actually matters.

The operational fixes from my last post won’t last without fixing the root of the problem. You can streamline an approval chain on Tuesday and watch it grow back to its original length within a year. The reason is that you fixed the symptom, not the disease.

The Conversation We Avoid

Let me walk you through a conversation that should happen in every organization, every team, and in every reporting relationship. In most organizations, this is rarely discussed. A leader sits down with a team — a department, a function, a project group. The leader says something like this.

Here’s what I want this team to accomplish over the next year. Here are the outcomes that would make this work successful. Here’s how we’ll measure progress. Here’s what is yours to decide. Here’s what is mine. Here’s what is bigger than both of us, and goes back up the chain.

Then the leader says the harder part.

Within the scope I just described, you don’t need my approval. I don’t want you waiting on me. I don’t want you escalating decisions that fall inside what we just agreed. If you’re inside the scope, decide. If you’re outside the scope, come back to me. The line is here. Don’t blur it. You are empowered to do what needs to be done to reach the desired outcomes.

That conversation, written down and signed off by everyone above the team, would eliminate most of the approvals you’ve been routing through five-person chains. That conversation, in most organizations, never happens.

What happens instead is fragmented. The team is given a budget, a title, a reporting structure, and maybe even a project name. They’re implicitly told what’s expected of them, and implicitly trusted to figure it out, and implicitly expected to know when to escalate. None of it is written down. None of it is signed off. None of it is durable. At any point, someone outside of the team can step in and grind progress to a halt.

The result is predictable. The team is uncertain about what they own. They escalate things they should have decided. The leader, busy with their own work, sends those escalations back down without thinking carefully about the pattern. The team escalates the next one too. The leader sets up an approval to “make sure things are coordinated.” The approval chain is born.

Two years later, that approval has grown to five or ten people. Each one was added in response to a specific incident or anxiety. None of them remembers why the chain originally existed. Everyone is rubber-stamping. The team has forgotten that any of these decisions used to be theirs.

You didn’t need fewer approvals. You needed the original conversation. You needed empowerment.

The Symptom and the Disease

Approval chains are the visible symptom. They sit on top of the actual problem like a fever sits on top of an infection. You can take the fever down with aspirin, that’s what Part 6 was about. but if you don’t address the infection, the fever comes back.

The infection is the absence of clarity about what the team owns.

This pattern shows up in three predictable places. Watch for them.

The escalation reflex.
Team members, when uncertain, default to escalating. The first time it happens, a manager might think it’s a one-off. The third time, the manager realizes the team doesn’t know what’s theirs. By the tenth time, an approval chain has been informally established, even if nobody wrote it down. The team has trained itself to ask permission, because the original definition was never given.

The “let me check with my boss” answer.
A team member is asked a question they should be able to answer themselves. Instead, they say they need to check. The asker accepts this as normal, because they’ve heard the same answer from everyone on the team. The team has effectively delegated all authority back upward, because the original delegation downward was never made clear.

The post-mortem that adds an approval.
Something goes wrong. The team did something they shouldn’t have done, or failed to do something they should have done. The post-mortem concludes that more oversight is needed. A new approval is added. What’s missed is that the failure happened because the team was unclear about what they owned in the first place. Adding more oversight reinforces the unclarity. The next failure will trigger another approval, not a clarifying conversation.

In each of these patterns, the symptom is the approval chain. The disease is the absence of clarity. Treating the symptom without treating the disease just produces more elaborate forms of the same problem. Now enterprise systems like Microsoft or Workday make the approval chain even “easier” to perpetuate, which further complicates empowerment.

What Real Clarity Looks Like

Here’s what the alternative actually feels like. A team that has had the desired outcomes and empowerment conversation operates differently. You can see it from the outside.

They don’t escalate decisions that are theirs. Even when something is hard or politically risky, they own it. They might come ask for advice. They don’t ask for permission. That’s a very real difference and it impacts how the team operates.

They don’t start sentences with “I need to check.” When asked a question that falls in their scope, they answer it. The answer might be “I don’t know yet, let me look into it.” They never say, “let me see if I’m allowed.”

They distinguish, in real time, between what’s in their scope and what isn’t. When something is in scope, they decide. When something is out of scope, they escalate cleanly with a recommendation, relevant context, and a clear sense of why they’re bringing it up. Their escalations are rare and useful, not frequent and reflexive.

Empowered teams don’t chase approvals. Their work moves at the speed of their judgment, not at the speed of the approval queue.

When you watch a team like this work, you notice something else. The leader’s calendar is dramatically less full of approvals. They aren’t signing off on things all day. The team is signing off.

This is what good organizational leadership should feels like. It’s rare because the conversation that creates it is rare.

Why the Conversation Doesn’t Happen

There are three reasons most leaders skip this conversation.

It’s hard

Telling a team what they own in writing is one of the most difficult things a leader does. It requires you to have actually decided what you want them to accomplish, what success looks like, what authority you’re willing to delegate, and what you want to retain. Most leaders haven’t done that work, even in their own heads. It’s much easier to leave it ambiguous and resolve specific instances as they come up.

It feels like loss of control.

When you write down what the team owns, you are formally delegating authority. The leader who has informally retained that authority by default loses something visible. The team can now point to the document and say “this was ours.” Some leaders find that uncomfortable, even though the day-to-day reality of the leader’s work gets dramatically better.

It exposes the leader to risk.

If the team does the wrong thing inside their scope, the leader can’t easily say “well, they should have escalated that to me.” The decision was theirs, by agreement. The leader who hasn’t had this conversation always retains plausible deniability, which feels safer than the alternative.

These three reasons explain why approval chains keep growing in organizations even when the cost is obvious. The cost is borne by the team, the requesters, and the speed of the organization. This benefits leaders who prefer to avoid difficult conversations resulting in ambiguity.

Until a organization is willing to absorb the cost of having the conversation, the chains will continue. Let’s face it. Most organizations are slow to act. The larger the organization, typically, the slower decisions are made. Most of them are more FRagile than agile.

The Empowerment Document

Here’s what the document looks like when it exists.

• It’s short. One to three pages, usually. Anything longer is performative.

• It states, in plain language, what the team is responsible for, not what they do. What they accomplish. It states the desired output, not the activity.

• It states what success looks like. Specific. Measurable. With a clear answer to “did we get there or not.”

• It states what is theirs to decide. With examples. The examples matter, because abstract authority delegations always get reinterpreted later. Concrete examples create durable shared understanding.

• It states what isn’t theirs to decide. Also with examples. The boundary needs to be visible from both sides.

• It states how progress will be tracked. Not “weekly status reports” that was Part 5. These are the actual Key Progress Indicators. The actual measurements. Tied directly to the outcomes named earlier.

• It’s signed by the leader, the team lead, and ideally one level above the leader. The signing matters. It makes the agreement durable. It survives the leader leaving. It survives the inevitable moment when someone tries to relitigate what the team can decide.

This document is the structural alternative to the approval chain. The chain exists because the document doesn’t. Write the document, and most of the chain becomes obviously unnecessary.

Need help getting started? I created this tool to help walk you through developing this document.

Empowerment Tool

Reframing the Political Fight

Last week’s piece named that approval chains are the politically hardest elimination, because they’re tied to power. This week’s argument changes the politics significantly.

If you walk into a leadership conversation and say, “I want to remove your approval authority,” you’ll lose… every single time. Authority is too important to people, and the framing is adversarial.

If you walk into the same conversation and say “I want to clarify what this team has always been responsible for, in writing, so we can stop running every decision through five reviewers who aren’t actually catching anything,” you’ve changed the game.

Now you aren’t asking the leader to give up power. You’re asking them to be specific about what they intended to delegate. Most leaders, given the chance, will agree that the team should have been able to make these decisions all along. The clarification isn’t a loss. It’s a confirmation.

The approvals that fall away in this conversation aren’t approvals the leader was actively withholding. They’re approvals that defaulted into existence because nobody was clear. The leader isn’t losing anything by clarifying.

For the rare approvals that the leader genuinely wants to keep, the document acknowledges that explicitly. Decisions above this dollar threshold come back to me. Decisions about new personnel come back to me. Strategic shifts that change the outcomes we just agreed on come back to me. The leader retains the authority that actually matters, and gives up the rubber-stamp authority that was never serving anyone.

This is why I said in Part 6 that this piece would give you a much stronger way to win the political fight. The fight you can’t win is “remove this approval.” The fight you can usually win is “let’s get clear on what this team has been responsible for all along.”

Seeing a Pattern?

I’ve been writing about approval chains because that’s what Part 6 covered. The argument generalizes.

Status reports, the subject of Part 5, exist because nobody is clear what the team should be measuring. Get clear on the outcomes and the KPIs, and the status report writes itself in two paragraphs… Or the AI summary of the actual underlying work is the status report, because the real measurements are visible in the work.

Meetings, the subject of Part 2, proliferate because nobody is clear what decisions a team is empowered to make versus which need group conversation. Get clear on that, and most meetings collapse into office hours and async updates.

The approval chain is the most concentrated version of the same pattern. But every elimination in this series traces back to the same root. The team was never told clearly what was theirs.

Once you see this, you start seeing it everywhere.

The committee that exists because nobody knew who should make the decision. The dashboard that exists because nobody knew which numbers actually mattered. The recurring sync that exists because nobody knew how the two functions were supposed to coordinate. The vendor relationship that nobody owns and everybody comments on. The cross-functional initiative that has six leaders and no decision-maker. All of these are clarity failures wearing the costume of process.

The Elimination Project hasn’t really been about cutting work. It has been about exposing the unclarity that the work was hiding.

Do This Today

Pick one team you lead, or one team that reports to you, or one team you sit on.

Write the document.

It doesn’t have to be perfect. It doesn’t have to be exhaustive. Three pages, plain language, what the team is responsible for, what success looks like, what is theirs to decide, what isn’t, how progress will be measured.

Then sit down with the team and read it together. Get their reactions. Revise it where it’s wrong. Sign it. Put it somewhere durable.

That document, more than any approval chain you streamline this quarter, is the work that makes the eliminations stick.

If you can’t write it, you’ve just discovered why the approval chain exists. The work to clarify is the work that prevents the chains from growing back. There is no shortcut.

Tell me what you wrote down. Tell me what surprised you about the conversation. Tell me what the team said when they read it.

Takeaways for today

We don’t need fewer approvals.

We need clearer ownership.

Justify everything you keep.

Eliminate everything else.

Thanks for reading AI, Innovation, and Faith! This is Part 7 of The Elimination Project. Part 1 (the manifesto), Part 2 (meetings), Part 3 (subscription decay), Part 4 (email), Part 5 (status reports), and Part 6 (approval chains) build the foundation. Part 8 closes the series with a high level look-back on what the project actually taught, what I got wrong, and the practice that turns out to matter most.

If you haven’t read Part 1, start there. The whole series rests on knowing what your work is supposed to move.

Today we go after a category most leaders refuse to question because it feels like the responsible thing to do. Cutting meetings feels brave. Cutting reports feels brave. Cutting approvals feels reckless.

Today, I want to argue the opposite. Most of your approval chains aren’t protecting you. They’re creating the illusion that someone is paying attention, while delivering none of the actual oversight that illusion implies. It’s sort of like the TSA line at the airport. It’s an illusion of safety. The latency and frustration approval chains create is real. The risk they were designed to catch is, in most cases, not being caught.

Let me show you why, and how to fix it operationally. The next piece in this series goes after the deeper question, why organizations end up with these chains in the first place, and the conversation many leaders never had with their teams that would have prevented them. Today, let’s fix the operational bloat.

What an Approval Chain Is Actually For

Every approval exists because, at some point, someone made a real argument. “This kind of decision needs another set of eyes. We do not want a single person making this call alone. The risk of getting it wrong is large enough to justify slowing down.”

That argument is sometimes correct. There are decisions that genuinely benefit from a second reviewer, a third reviewer, or even a committee. Big spends, material risk, or decisions that affect the organization’s identity, security, or solvency... yeah, let’s get a second set of eyes on those.

The problem is that approvals are easy to add and almost impossible to remove. Every time something goes wrong, the standard response is to add a new approval step to prevent it from happening again. The thresholds get lower. The chains get longer. The original purpose of “another set of eyes” becomes “five people none of whom are actually looking.”

Walk through any organization that’s been around for more than a decade. The approval architecture looks like a coral reef with layers built on layers built on layers, each in response to a specific incident, none of which were ever revisited once the incident faded from memory. Most of the layers are no longer doing anything. They’re just there.

This is the architecture you’re operating inside. The question is whether you can see it

.

Once, I had a company vehicle and was told that I had to get approval before I could get the oil changed. If you’ve read much of my writing, you know that this type of logic isn’t something I can ignore. I asked our business manager if he wanted oil changes done every 5000 km and of course he said yes. My next question was, “Is there ever an occasion where getting an oil change would not be approved?” The answer was obviously no. My next question was, “Then why in the world do I need to get approval for routine maintenance that you’ll always approve?” Crickets.

The Three Failure Modes

Approval chains fail in three ways. Each one is common. Each one is invisible to the person inside the system.

Failure mode 1: The rubber-stamp chain.
The chain has five reviewers. Four of them auto-approve everything. They’ve approved every request that has come across their desk for the last two years. Their “review” consists of clicking a button. They aren’t actually reading the request. They aren’t actually thinking about whether it should be approved.

This is the most common failure mode and the worst one. A rubber-stamp chain is worse than no chain at all. With no chain, the requester knows they own the decision. With a rubber-stamp chain, the requester thinks the decision has been reviewed, and the reviewers think the decision has been delegated to other reviewers. Nobody is actually paying attention. The chain manufactures false confidence.

Failure mode 2: The latency chain.
Each reviewer takes one to three days to respond. Five reviewers means a week to two weeks of latency on every decision. The decisions that make it through are stale by the time they’re approved. The decisions that fail to make it through fail not because they were rejected, but because the requester gave up and worked around the system.

Latency is its own form of governance failure. A decision that takes three weeks to approve is, for many practical purposes, a decision that didn’t happen. The work that depended on the decision moved on. The window of opportunity closed. The competitive moment passed. The team learned to route around the chain entirely.

The latency chain drives me crazy!

Failure mode 3: The accountability dilution chain.
Five reviewers means no single reviewer feels responsible for the decision. The first reviewer assumes the second one is the real check. The second one assumes the same about the third. Everyone assumes someone else is doing the actual work. When something does go wrong, the post-mortem finds five people who all approved it and none of whom can explain why.

This is the failure mode that makes leadership most uncomfortable to confront, because the entire architecture was designed to create shared accountability. What it actually creates is shared deniability.

If your approval chains exhibit any of these three failure modes, the chain isn’t protecting the organization. It’s consuming time and creating false confidence. The needle isn’t being moved. It’s being slowed.

The Latency Math

Here’s a calculation worth doing on every approval chain in your organization.

Take a sample of recent decisions that flowed through the chain. For each one, ask:

• How many reviewers did the request pass through?

• How many days elapsed from submission to final approval?

• What percentage of the requests were approved? (If the answer is above 95%, you’re looking at a rubber-stamp chain.)

• For the rejected requests, did the rejection come from a substantive review, or from a process technicality (wrong form, missing field, expired prerequisite)?

• For the approved requests, what evidence do you have that the approval was based on actual review rather than a click?

The data will surprise you. For many organizations the typical approval chain has a 95%+ approval rate, takes 5 to 15 business days, and rejects requests primarily for procedural rather than substantive reasons. It wasn’t rejected because it didn’t move the needle. It was rejected because it was charged to the wrong account number.

That’s not governance. That’s queue management with extra steps. The chain is filtering for “did the requester complete the form correctly,” not “is this decision a good one, ie… does it move the needle.”

If 95% of the requests are approved, the chain is providing a 5% filter at the cost of 100% of the decisions being delayed. That trade is almost never worth it. Most of those delays are slowing decisions that would have been approved anyway. The organization is paying enormous latency cost to catch a tiny fraction of bad decisions, while the rubber-stamp culture means many of the bad decisions are slipping through anyway.

Cancelling the entire chain and accepting the 5% miss rate would, in most cases, produce better outcomes than the current system. Faster decisions, equivalent risk, less false confidence.

The Move Needle Test for Approvals

Apply this test to every approval chain you can identify in your organization.

If this approval step were removed, what risk would actually materialize that isn’t being caught today?

Be specific. Name the failure mode. Name the actual consequence. Name the frequency.

For most approval steps, the honest answer is “none in living memory.” The step exists because someone added it after a 2018 incident that nobody currently in the chain remembers. The risk it was designed to catch hasn’t happened in seven years, and it’s not clear the step would have caught it anyway. The step is just there.

That step isn’t preventing risk. It’s creating it. The risk is the latency on every decision that has flowed through it for seven years, every workaround that team members have developed to avoid it, every opportunity that closed while the request was sitting in a queue.

The needle test for approvals is the same as for everything else in this series. Is this activity moving the needle, or is it consuming attention without producing leverage? If a reviewer can’t articulate, in concrete terms, what they would catch that isn’t being caught elsewhere, that reviewer isn’t adding value. They’re adding latency.

What to Replace Approvals With

Three replacement patterns work, and each one targets a different category of decision.

For low-risk, high-frequency decisions: automate, don’t approve.
Most approval chains exist for decisions that are nearly always yes. Travel under $1,000. Software subscriptions under $200. Standard vendor renewals. These should be auto-approved within defined limits, with audit logging to catch anomalies. AI is excellent at this through pre-screening requests, flagging the genuinely unusual ones, and sending the rest through automatically. The human reviewer only sees the exceptions.

The cultural objection is “but what if something gets through?” The answer is that something is already getting through, because rubber-stamp reviewers aren’t catching anything. Automation with audit logging is more effective than rubber-stamping, not less, because the audit trail is honest about what was approved and why.

For genuinely judgment-heavy decisions: shorter chain, real review.
Some decisions actually require human judgment. Hiring a senior leader. Committing to a multi-year contract. Approving a strategic shift. These deserve careful review. They don’t deserve five reviewers. They deserve one or two reviewers who are genuinely responsible for thinking about the decision, not five who each rubber-stamp because they assume someone else is doing the real work.

A one or two person review with substantive engagement beats a five-person rubber-stamp every time. Cut the chain length. Tell the remaining reviewers they own the decision. Not just that they’re clicking the button.

For unusual or high-stakes decisions: the convened review. When something genuinely warrants multiple perspectives, don’t route it through a sequential chain. Convene the people, get them in a room (real or virtual), discuss, and decide. The decision document pattern from Part 2 works here. Write up the proposal, send it ahead, hold a 30-minute focused discussion, and decide.

This pattern produces better decisions in less time than any sequential approval chain. It also creates real accountability, because everyone in the room can see what everyone else thought. The decision is owned, not delegated.

A Story Worth Sitting With

Years ago, when I joined a regional leadership team, a colleague who had been there for years was helping me find my footing. I mentioned a decision I was making, and that I was going to copy our boss on the email when I sent it.

He stopped me. “Don’t copy Phil on that. His email box is loaded as it is. You have been empowered to make the decision. You made the decision. Just mention it to Phil the next time you guys are talking. Don’t load his email with something that isn’t necessary.”

That advice has stuck with me ever since.

I bring it up here because the same logic applies to approval chains. Every unnecessary approval is a CC. Each one is a small reversal of authority. It’s a decision being routed back upward when it was already supposed to belong somewhere else. Each one consumes the attention of someone whose attention is too expensive for what they’re being asked to look at.

My colleague wasn’t just teaching me email etiquette. He was teaching me that real authority requires the courage to act on it. Adding the CC was hedging, while the cure wasn’t the CC. The cure was the courage to make the decision I had been empowered to make and tell Phil about it later.

I’ll come back to this in Part 8 of this series. For now, just notice that the approval chain you’re looking at is, in many cases, the same instinct as that unnecessary CC. The decision was supposed to belong to someone. The system has quietly insisted that it doesn’t.

Why This Is Politically Hardest

Of all the eliminations in this series, approval chains are the politically hardest to cut. The reason is that approval chains are tied to power. Every approval step is a small piece of authority that someone holds. Asking them to give it up reads as asking them to give up status.

The leaders who hold approval authority will resist the elimination. They’ll argue, often sincerely, that the approval is providing real oversight. They’ll tell stories about the time they caught something. They’ll frame the elimination as reckless.

The truth is usually more uncomfortable. The approval isn’t providing the oversight they think it is. The “time they caught something” is one anecdote against thousands of rubber-stamps. The elimination isn’t reckless. I argue the current system is reckless, because it’s producing latency without actually catching risk.

That argument is hard to make in a room full of people whose authority depends on the approval staying in place. So the chain persists. Layer accumulates on layer. Bloated organizations are the worst at this because the can be. Bloat begets bloat.

The only way to actually cut approval chains at scale is for the leader at the top of the chain to make the decision deliberately. Nobody below them has the political standing to do it. The lead can simplify the approval architecture within their domain.

If you’re reading this and you’re at the top of a chain you control, you have an opportunity most leaders don’t. Use it.

(My next post will give you a much stronger way to win this political fight, by reframing the cut as a clarification rather than a removal. For today, just see the architecture for what it is.

)

The Most Common Specific Eliminations

A few specific approval chains are worth calling out because they appear in nearly every organization and are nearly always candidates for elimination.

Travel approvals under a meaningful threshold.
If a $100 trip requires three approvals, the approvals are costing more than the trip. Set a threshold (whatever makes sense for your org) below which travel is auto-approved with logging.

Software subscriptions under a meaningful threshold.
Same logic. The team member who needs a $50/month tool shouldn’t need three managers to sign off, especially if has been included in the budget. They should be able to expense it within policy and have it logged for audit.

Standard vendor renewals.
If you renewed last year and nothing material has changed, you don’t need a five-person review of the renewal. Auto-renew within policy, flag the genuine renegotiations.

Document approvals on standard templates.
If the document is a fill-in-the-blank version of an approved template, the template was already approved. The new instance doesn’t need to be re-approved by everyone who approved the template.

Each of these is a small elimination. Together they remove enormous amounts of latency from the organization. The savings aren’t just the reviewer time, they’re the requester time, the work that was waiting on the approval, and the opportunities that didn’t have to close while the request was in a queue.

Notice the pattern in all examples. Each of them is a decision that should already belong to the team. The approval chain exists because the team was never clearly told it was theirs. We don’t need fewer approvals. We need clearer ownership. That’s the deeper argument, and it gets its own piece next week.

One Take-away

Pick one approval chain you sit on. Just one. Pull the data on the last 20 requests that flowed through it.

How many were approved? How many were rejected? Of the rejections, how many were substantive versus procedural? How long did the average request take?

If the approval rate is above 95% and the average latency is more than three days, that chain should be on the chopping block.

Don’t propose the streamlining yet. Just gather the data. Bring it with you next week, when we go after the deeper question. Why this chain exists and what would have to be true for it to be unnecessary in the first place.

Tell me what your data showed. The number will be worse than you expect.

Justify everything you keep.

Eliminate everything else.

Thanks for reading AI, Innovation, and Faith! This is Part 6 of The Elimination Project. Part 1 (the manifesto), Part 2 (meetings), Part 3 (subscription decay), Part 4 (email), and Part 5 (status reports) build the foundation. Part 7 takes on the deeper structural critique: we don’t need fewer approvals — we need clearer ownership.

If you haven’t read Part 1, start there. The whole series rests on knowing what your work is supposed to move.

The first four parts of this series went after the obvious culprits. Meetings that should have been an email. Subscriptions that should have been cancelled three years ago. Email that should have been a document. Those are the easy eliminations.

Today we go after something harder. Something woven so deeply into how organizations operate that most leaders have never thought to question it. The work that exists because someone, at some point, wanted visibility, and we’ve been producing these reports like clock work ever since.

Status reports.

I’m going to make a stronger claim than I made in earlier parts of this series. Most status reports in your organization should not exist. Notice, I didn’t say they “should be shorter” or that they “should be AI-assisted.” They just shouldn’t exist.

If that sounds extreme, stay with me. By the end of this piece I think you’ll agree, and you’ll know how to start cutting

.

The Math Doesn’t Math

Walk through your organization’s recurring written reports, project status updates, weekly team summaries, monthly progress reports, quarterly reviews, executive dashboards, field reports, department updates, and annual reports. Do we really need all of these reports?

For each one of these (and other) reports, you should ask two questions.

1. How many person-hours go into producing it each cycle?

2. How many seconds does the average reader actually spend with it?

The arithmetic is brutal. A weekly status report that takes an hour to produce, distributed to ten people who skim it for thirty seconds, is paying sixty minutes of production time for five minutes of total reading attention. The producer gets one minute of attention back for every twelve they invested. Not a great dividend.

Now multiply that by every recurring report in your organization. Multiply by every team. Multiply by a year. Yikes. This is really bad stewardship of time.

The reason this persists isn’t that anyone has done the math and decided it’s worth it. The reason it persists is that nobody has done the math at all. The status report is just there. It has always been there and cancelling it feels risky. Producing it feels safe… so it continues. This is like so many of the things we do. We’ve just always done it and no one asks what the desired out come is.

It’s time someone put status reports to the needle test. What does your organization exist to do? Does this status report move the needle on the important things that your organization does? If not… why are you doing it?

What a Status Report Is Actually For

To eliminate status reports intelligently, we have to be honest about what they were originally supposed to do.

The original purpose of a status report was simple. Leaders had no other way to know what was happening across their organization. The work was distributed, the people were distributed, and the only way to maintain visibility was to ask everyone to write down what they did and send it up the chain.

This made sense in 1985. It made sense in 2005, mostly. By 2015, it was already starting to creak.

Today, in 2026, the original premise of the status report has collapsed. Your leaders, today, can ask AI to summarize what happened this week from the actual artifacts of work. You can look at Teams messages, the document edits, the ticket movements, the calendar items, the deployments, the deliverables. The AI can compile that summary in seconds. It’s more accurate than what the human author would have written, because the human author has incentives to perform the week back rather than describe it. The AI has no such incentive. It just reports what happened.

If the original purpose of the status report was visibility, that purpose is now better served by other means. Cheaper, faster, more accurate, and available on demand instead of on a calendar.

So why are we still writing them? A few reasons and none of them are very flattering. We write them because we’ve always written them, and the routine has become identity.

We write them because the act of writing the report has become the way the writer feels they’ve done the week’s work, even when the report is the only thing they actually finished.

We write them because the recipient asked for them once, three years ago, and nobody has ever asked whether the recipient still wants them.

We write them because nobody is willing to be the person who stops, and nobody is willing to be the person who tells someone else to stop.

None of these reasons survive the needle test. None of these reasons connect to actual progress on what matters!

The Three Categories

Walk through the recurring written reports in your organization. Each one falls into one of three categories.

Category 1: Reports that drive a decision.
A budget update that determines whether to approve next quarter’s spend. A sales pipeline review that changes how resources are deployed. A risk register that surfaces something requiring intervention. These reports earn their hours. They produce decisions. The needle moves because of them.

These should be kept. Probably sharpened, probably shortened, definitely AI-assisted in their production. But kept.

Category 2: Reports that are read, but don’t drive a decision.
The weekly team summary that goes to the manager, who skims it but never acts on it. The monthly project update that the steering committee reads but never modifies the project based on. The quarterly review that produces follow-up questions but rarely follow-up action. These are reports that have an audience but no consequence.

These are the hardest to eliminate, because someone is technically reading them. The reading feels like it justifies the writing. It doesn’t. If the reading doesn’t produce a decision or a change, the report is consuming attention without producing leverage. The right move is to convert these to on-demand. The recipient asks for the summary when they actually need it. The producer isn’t on a calendar.

Category 3: Reports that are barely read.
The status email that gets a 12% open rate. The project update that nobody opens. The dashboard that has not been viewed in three months. The annual report that took six weeks to produce and got read for ten minutes total across the entire intended audience.

These should not exist. Cut them. The recipients won’t notice. If they do notice, that’s the moment to ask whether they actually need the report, in which case you can replace it with something better. Most of the time, they won’t notice.

The Needle Test for Reports

Here’s the test, applied to every recurring report you produce or receive.

If this report were not produced this cycle, what would happen? Would any decision change? Would any action be different? Would the work move slower or faster?

Stop and sit with that. For most of the reports you produce, the honest answer is “nothing would change.” The report wouldn’t arrive. Nobody would notice. The work would proceed exactly as it does now.

That is a report that isn’t moving the needle. That is a report that should not exist.

The objection most leaders raise here is “but what if leadership wants to know?” The answer is to give leadership a better way to know. Not the recurring report. The on-demand AI summary. Generated when leadership actually wants it. Pulled from the underlying work artifacts in real time. More accurate than the human-authored version. Available in seconds.

Good leadership doesn’t actually want a status report. Good leadership wants visibility. Those are not the same thing.

What to Replace Them With

Three replacement patterns work, and each one connects to a different reader need.

For decisions:
Keep the report. Sharpen it. Use AI to draft the first version from real artifacts, then have the human author edit and add judgment. The shape of the report stays. The hours invested drop. The decisions still get made.

For visibility without decisions:
On-demand AI summaries. Configure your AI tools to compile weekly summaries from your team’s actual work. Look at the commits, documents, deliverables, and constituent communications. Look at the things that move the needle on your core purpose. Make these available in a shared space. Anyone who wants the summary can pull it. Nobody is forced to write or read on a schedule.

For ritual or habit:
Stop. The report exists for no reason that survives the needle test. Cancel it. Tell the audience you’re cancelling it. Offer a replacement only if anyone actually objects. Most won’t.

The third category is the one most leaders find hardest to act on. Cancelling a report feels like loss. Adding a report feels like investment. The political math always favors the addition. AI changes the equation only because it lowers the cost of the alternative, but only the leader can make the choice.

The Deeper Problem

The reason status reports proliferate in organizations is that we’ve gotten lazy about what we measure. We measure activity instead of outcomes. We measure effort instead of progress. We measure inputs instead of things that actually move needles.

A weekly status report is, in its most common form, a list of activities. I had three meetings. I sent two emails. I reviewed four documents. The activities feel like proof of work, but they aren’t proof of progress. None of them tell you whether anything moved.

Compare that to a report that connects to a needle. Project X moved from 60% to 75% completion this week. Customer churn dropped from 4% to 3.2%. Two of our three quarterly milestones are now on target. These are reports about outcomes. They tell you whether the needle is moving. They are short, because outcomes can be expressed in numbers and numbers don’t need many words.

The activity-based status report is a symptom of unclear goals. The needle-based status report is a symptom of clear ones. If your team’s status reports read like activity logs, the problem isn’t the reports. The problem is that the team hasn’t been told clearly enough what they’re supposed to move, so they fall back on what they did.

This is the deeper structural critique. The status report problem and the goals problem are the same problem. A team with a clear needle doesn’t need to write activity logs, because the needle itself is the report. Did it move? By how much? What’s the bottleneck? That’s the entire status update.

If your organization is producing mounds of activity-based status reports, you don’t have a reporting problem. You have a goals problem. The reports are working as designed. They’re showing you that nobody has clear enough goals for outcomes to be the natural unit of measurement.

The elimination of the report is the easy part. The elimination of the goal-haziness underneath it is the work that actually matters.

The Annual Report Is the Boss of This Category

I want to call out one specific case before I close, because it’s the largest single instance of this problem in many organizations.

The annual report. The year-in-review document. The “what we accomplished this year” packet. Many organizations spend weeks producing one. The graphic design, the photo selection, the carefully framed statistics, the donor-facing narrative, the leadership letters, the financial summaries.

For most organizations, the annual report is the most expensive single piece of recurring written work the organization produces. And for most organizations, the annual report is read by a tiny fraction of its intended audience for a fraction of the time the producers spent on it.

The annual report should be on the chopping block. Not because the work it describes doesn’t matter. The work matters. But the report, as a 60-page document produced over six weeks and read for ten minutes by 12% of the intended audience. That report isn’t the right artifact for the job.

The replacement isn’t “a shorter annual report.” The replacement is to think about what the annual report is actually supposed to do, and design directly for that outcome.

If it’s donor cultivation, the right artifact is probably a series of personal videos and individual donor calls, not a glossy document.

If it’s institutional credibility for grants and partnerships, the right artifact is probably a tight one-pager and a website with on-demand depth, not the printed pack.

If it’s internal celebration of the year’s wins, the right artifact is probably an in-person event with a story arc, not a document.

The annual report exists in many organizations because nobody has asked the question. It’s one of the highest-leverage eliminations available to most leaders, and it’s one of the rarest to actually attempt. I get it. You “have to have” an annual report. Just think about whether or not you are actually communicating on the needle movers.

One Take-away

Today, I want to ask you to spend half an hour thinking about all of the reports you have.

Make a list of every recurring written report you produce or are responsible for. Status updates. Project reports. Weekly summaries. Monthly briefings. Quarterly reviews. The annual report.

For each one, ask the test question.

If this report were not produced this cycle, what decision would change? What action would differ?

If the answer is nothing, that report is on the chopping block. Send a message to the recipient. I’m pausing this report for the next month. If you find you actually need it, let me know and we’ll figure out a better way to give you what you need.

Then wait. See who reaches out. The number will be smaller than you expect.

Tell me what you cancelled. Tell me whether anyone noticed.

The status report has been silently consuming hours of your team’s most senior people for years. Many of those reports were never serving the purpose anyone intended. The AI era makes the alternative cheap. The needle test makes the choice clear.

Justify everything you keep.

Eliminate everything else.

Especially the reports nobody actually reads.

Thanks for reading AI, Innovation, and Faith! This is Part 5 of The Elimination Project. Part 1 (the manifesto), Part 2 (meetings), Part 3 (subscription decay), and Part 4 (email) build the foundation. Part 6 takes on approval chains. The politically hardest elimination in the series.

If you haven’t read Part 1, start there. The whole series rests on knowing what your work is supposed to move.

Yesterday’s piece on Unsubscribing from email lists was the warm-up for today. Sort your deleted folder. Spend 30 minutes unsubscribing. The result was an immediate volume drop and a quieter inbox. If you did the work, your email is already cleaner today than it was yesterday.

Today we are going to do a bigger audit of email.

Subscriptions handled the easy category. It wiped out the automated mail you never asked for (or you don’t remembe raskign for). What remains is the harder stuff. The actual messages from actual humans about actual work. The inbox you can’t unsubscribe from because the senders are your colleagues, your customers, your team, your supporters, etc... The mass of email that’s technically real and yet, when you look at it honestly, is mostly not advancing anything you really care about. It’s not moving the needle.

Time to apply the needle test to these emails as well.

The Volume Trap

The average knowledge worker sends and receives over 100 emails a day. A leader at scale receives more. The hours spent in the inbox, every week, are real. For most of us, those hours are second only to meetings as a consumer of the work week.

When most of us feel the email crush, our instinct is to find tools that help us go faster. AI drafts. AI triage. AI summarization. AI sorting. AI auto-replies. The marketing copy for these tools sounds great. The promise is “do email faster, get more time back.”

That promise is a trap if you haven’t done the needle work first. Faster email in service of unmeasured activity is just faster activity. Let’s say you actually hit inbox zero by 9:30 AM. You likely still have no idea whether you moved anything that matters.

Speed isn’t the answer to the email problem. Discrimination is. Most of your email isn’t actually trying to advance your needle. It’s trying to advance someone else’s, or trying to perform responsiveness, or trying to keep itself alive through email ritual. The work is to see which is which and stop processing the rest.

Five Categories

Open your inbox right now and walk through the last 30 days of messages. Every email you’ve received fits into one of five categories.

Category 1: Email that advances the needle directly.
A message from a colleague making a real decision on the work that matters. A donor reply on a strategic conversation. A field worker raising a problem only I can solve. These are rare. They deserve full attention and a thoughtful response. AI is mostly the wrong tool for this category, because the response is where the real work happens.

Category 2: Email that surfaces information I need to do the needle work.
Updates from the team, intelligence from partners, signal from the field. AI is excellent at compressing this category. Summarize the digest. Surface what’s new. Flag what needs my attention. This is where AI saves real time.

Category 3: Email that exists because someone wanted to feel responsive.
Status updates I didn’t ask for. Long replies to my one-line questions. CCs on threads I didn’t need to be on. Replies that say “thanks!” on a thread that was already closed. This is where the elimination work happens. Not “respond faster.” Stop being on the threads at all.

Category 4: Email that exists because the sender wanted my approval.
Some of this is real, but if we’re honest, much of it is rubber-stamp. AI can pre-screen the rubber-stamps and just send them through, leaving you only the ones that actually require judgment. Better still, redesign the workflow so the rubber-stamps don’t reach you in the first place. Build a better process!

Category 5: Email that exists because of automation, subscription, or accident.
Newsletters, vendor updates, spam notifications, “in case you missed it” digests. If you did the Part 3 audit, this category should already be dramatically smaller. If you skipped Part 3, go back and do it now. That 30 minutes is the prerequisite for everything in this piece.

For most of you, the breakdown after the Part 3 cleanup will be roughly: Category 1 around 5%, Category 2 maybe 15%, and the rest some mix of 3 and 4 (with Category 5 mostly already cut). Sixty to eighty percent of what remains isn’t advancing your needle. It’s consuming the attention you would have used to advance it.

Once you see that breakdown, the elimination strategy writes itself.

What to Eliminate, in Order

Here’s the order I would attack what’s left.

Start with Category 4 — the rubber-stamp approvals.
Walk through your standing approval chains and ask which ones you’re actually evaluating versus which ones you’re rubber-stamping. The rubber-stamps either need to be redesigned (so they no longer come to you) or automated (so they go through without your touch). If you’re not adding judgment, you’re not adding value. You’re adding latency. Approvals are getting their own deeper treatment in Parts 6 and 7. For now, just notice how much of your inbox is approval theater.

Then attack Category 3. I call this the responsiveness theater.
This is the hardest one because it’s cultural. The “thanks!” reply, the “just looping you in,” the CC-everyone reflex. You can’t fix this for your whole organization in a week. You can fix it for yourself by being explicit. Tell your team you don’t need to be CCd on threads where you’re not making a decision. Tell them you don’t need acknowledgment replies. Model the behavior by sending fewer of them yourself.

For Category 2, the information surfacing. deploy AI.
This is where the AI tools earn their keep. Have an AI agent triage your inbox each morning. Surface the threads that need attention. Summarize the long ones. Flag the ones connected to your current priorities. Most modern email tools (Gmail with Gemini or Outlook with Copilot) can do meaningful versions of this today.

For Category 1, the email that actually matters. be a human. Read the message carefully. Write a thoughtful response. This is the work AI cannot do for you, and you’ll know which messages they are because they’re the ones connected to your needle.

The order matters. Start with Category 2 and deploy AI to triage and you’ll get faster at consuming the noise. but you won’t have eliminated any of it. Cut the volume first. Then the remaining email is the email that actually deserves your attention, and the AI tools become useful in service of the work that matters.

A Standing Rule for Your Inbox

Once you’ve done the audit, install a standing rule.

Before I respond to any email, I ask whether responding moves the needle. If the answer is no, I don’t respond.

This rule will feel rude the first week. It will feel disciplined the second week. By week four it’ll feel like the only way you’ve ever worked.

A few corollaries that follow from the rule.

You stop sending acknowledgment replies. The “thanks, got it” reply is itself a Category 3 email — except now you’re the sender. Just because you received an email doesn’t obligate you to write back.

You stop responding to threads where the only available action is to clarify a misunderstanding the original sender created. Send the question back to them. Don’t do their thinking for them.

You stop staying on CC chains where you have no decision to make. Reply once, asking to be removed. Better yet, don’t reply. Just filter the thread out of your inbox.

You stop writing 200-word emails when 20 will do. Some people write 2000-word emails… you know who they are. The recipients are also leaders trying to protect their attention. A short email respects them. A long email doesn’t.

These aren’t rudeness. They’re the disciplines of someone who actually has a needle and is trying to move it.

The Email That Is Really a Meeting in Slow Motion

One specific category breaks more leaders than any of the others, and it’s worth flagging before we close.

You know the threads I mean. The 14-message back-and-forth where four people reply-all over three days to settle something a one-page document would have closed in an hour. The thread that started as a simple question and became a slow-motion meeting nobody scheduled. The chain that ends with someone saying, “we should probably get on a call about this.”

Yes. You probably should have. Or better yet, you should have written the decision document first.

This is the inverse of the meeting elimination from Part 2. Sometimes the elimination move isn’t less synchronous communication, but more. A short, focused conversation can replace a week of slow email. Asynchronous isn’t always the answer. The answer is whichever format moves the needle fastest.

Stop and ask yourself, when an email thread is dragging: would 15 minutes of synchronous conversation, or one well-written decision document, end this faster? If yes, do that instead. Stop the email thread. Replace it.

One Take-away

Today, do this one thing.

Run the inbox audit. Walk through the last month of email. Sort it into the five categories. Note the percentages. The numbers will surprise you.

Then write down, in a single sentence, the standing rule you’re going to apply going forward. Mine is I do not respond unless responding moves the needle. Yours might be different. The point is to make the rule explicit, to yourself, in writing.

Tell me what your audit revealed. Tell me what percentage of your inbox was actually connected to your needle. If the answer surprises you, you’re not alone. That surprise is the start of the work.

Justify everything you keep.

Eliminate everything else.

The needle decides what stays in your inbox, and what does not.

Thanks for reading AI, Innovation, and Faith! This is Part 4 of The Elimination Project. Part 1 (the manifesto), Part 2 (meetings), and Part 3 (subscription decay) build the foundation. Part 5 takes on status reports — the most expensive form of organizational theater.

If you haven’t read Part 1, start there. The whole series rests on knowing what your work is supposed to move.

Yesterday’s piece took on meetings. Meetings are the most expensive thing on your calendar. Today we go after the smallest, easiest, and most underestimated elimination in the series. Before we get to the bigger inbox audit later, we’re going to do the warm-up that makes the bigger audit possible.

For the last two or three years, I’ve made it a personal practice to unsubscribe from things every few months. I thought I was being disciplined and my inbox was reasonably clean.

Last week I tried something new and discovered I was wrong.

The Sorted Deleted Box

Here’s the assignment for today.

Open your email program. Go to your Deleted Items folder. Sort by sender.

Just that. Sort by sender, descending by count. The senders sending you the most email rise to the top.

What you’re about to see may surprise you. The senders that come up at the top of that list aren’t your colleagues. Not your boss. Not your customers or partners. Not the people whose work you actually care about.

The top of the list is automated mail. Subscriptions. Notifications. Newsletters. Credit card alerts. SaaS update emails. “In case you missed it” digests from platforms you joined years ago and barely use. Daily summaries from services you signed up for once and forgot about. You know, the things you immediately delete when you see them in your inbox.

In my case, the top of the list had senders I didn’t even recognize. Names I’d never noticed. Domains I had no memory of subscribing to. Some of them had been sending me daily email. I’d been deleting their messages. I’d never once stopped and asked the obvious question.

Why am I receiving this in the first place?

I unsubscribed from almost a hundred of them in one sitting. Less than 10 minutes of work. The result was immediate and substantial. Remember, I clean my subscriptions ever few months. I can’t believe it had gotten so bad all over again.

The Hidden Cost

You might think the cost of these emails is zero. They arrive. You delete them. Total time per email? Maybe a second. What’s the big deal?

The big deal is that “delete it in a second” is a lie we tell ourselves about email volume.

Every email you delete is a microscopic context switch. You glance at the sender. Your brain decides “junk.” Your hand moves to delete. The whole process takes a second or two. Then you scan the next one. Across 50 of these in a morning, you’ve done 50 micro-evaluations. You’ve moved your eyes 50 times. You’ve spent two or three minutes deleting nothing of value.

That’s the obvious cost. There’s a less obvious one.

The hidden cost is that those messages are there. They’re sitting in your inbox alongside the messages that actually matter. Every time you open the inbox to find the real email, you’re scanning past the noise. The noise isn’t free. It’s the visual and cognitive equivalent of working in a room where 80% of the surfaces are covered in junk mail you have to look past to find the one thing you came in to do.

It isn’t the deletion that costs you. It’s the perpetual scanning past. Multiply that by the dozens of times you open your inbox each day. Multiply that by every workday for a year.

The arithmetic is depressing, but the elimination is easy.

The One-Month Rule

Here’s the pattern I use now.

If I haven’t read anything from a sender in the last month, I’m not going to read anything from them in the next month either. Unsubscribe.

That’s it. That’s the whole rule.

The voice in your head will object. But what if there’s something important in there? It wasn’t important enough that you read it last month. It won’t be important enough that you read it next month either.

But what if I might need that information someday? If you need it, you know where to find it. The internet didn’t disappear. The website still exists. The newsletter archive is searchable. The product update page is one Google query away. You aren’t losing access to the information. You’re losing the obligation to scan past it daily on the off chance you might need it.

But the company asked me to subscribe. Companies ask for many things. You owe them nothing. Your inbox isn’t a free advertising surface. Unsubscribe.

The one-month rule cuts through the hesitation. If you can’t remember opening anything from a sender in the last 30 days, the relationship is dead. You’re doing both yourself and the sender a favor by ending it. Their open rate goes up when uninterested subscribers leave. Your inbox goes quiet and stop subscribing to things you don’t read.

The Categories You Will Find

When I sorted my own deleted box, the senders fell into a handful of patterns. Yours will too.

Subscriptions you signed up for during a project that ended. That research newsletter you subscribed to when you were writing the white paper in 2022. That industry digest you joined when you were doing research. That platform you set up for the conference you attended in 2023. The conference is over, but not the emails.

Marketing from companies you bought from once. You ordered something. They added you to their list. They’ve been sending you weekly promotions ever since. Even if you genuinely liked the product, you didn’t sign up for a daily relationship. Some of these were daily!

Notifications from SaaS tools you barely use. Project management platforms, collaboration tools, productivity apps. They send you a notification every time anyone does anything. Most of these can be turned off in settings. Many can’t, in which case the unsubscribe is the only path.

News and content digests. The morning news email. The weekly industry roundup. The “what you missed this week” digest. These feel valuable in the abstract — you’re staying informed. In practice, you’re deleting them. Stay informed in a way that fits how you actually consume information. RSS readers, set times of day, deliberate visits to specific sources. Not push notifications you delete every morning.

Notifications from financial accounts. Credit cards, banks, investment platforms. Some of these matter (fraud alerts, statements). Most don’t (rewards updates, marketing offers). Most banks now let you turn off the marketing notifications without losing the security ones. Go into your account settings and do it.

Email from people you no longer work with. Personal newsletters from a colleague three jobs ago. Updates from a company you stopped using. Mailing lists from an organization you left. The relationship is over. The list doesn’t know that. Unsubscribe.

The aggregate of these is shocking when you actually count it. You’ll find dozens. Probably more. Almost none of them are actively harmful. All of them are quietly consuming attention.

How to Actually Do It

There are a few ways to attack this. Pick one.

The 30-minute manual sweep. Sort your deleted folder by sender. Scroll down the top 50 senders. For each one you don’t recognize or don’t actually read, click into a recent message, scroll to the bottom, and click unsubscribe. Repeat until done. This is what I did last week. It works. It’s also the version that scales to “almost a hundred subscriptions” in one sitting.

The unsubscribe service. Tools like Unroll.me, Clean Email, or Leave Me Alone aggregate your subscriptions and let you bulk-unsubscribe with checkboxes. Some of these tools have privacy concerns worth checking. Read the terms before authorizing access to your inbox. Personally, I would not use them and just do my own unsubscribing.

The Copilot or AI agent approach. If you use Microsoft 365 with Copilot, Gmail with Gemini, or another AI-integrated mail platform, ask it directly. “Identify senders I have received more than five emails from in the last 90 days but have not opened any of them. Show me the list.” The AI does the analysis you would have done by hand. You make the final decisions and unsubscribe.

The standing rule. Once you’ve done the initial cleanup, install a standing rule for yourself. Every time you delete an email without reading it, ask whether you want to unsubscribe right then. The unsubscribe link takes 10 seconds. Doing it once permanently saves you from doing the deletion forever. This is the rule that prevents subscription decay from rebuilding over time.

I run the manual sweep on the first of every month. It takes about ten minutes. The result lasts a couple of months before subscription gravity slowly rebuilds the noise. It’s one of the few maintenance routines I’ve found genuinely worth it.

The Wider Application

I gave this one its own piece, rather than tucking it into tomorrow’s broader email post, because the principle generalizes beyond email.

You probably have:

• Calendar invites for recurring meetings you stopped attending months ago that are still on your calendar.

• Slack and Teams channels you joined for a project that ended that still send you notifications.

• App notifications on your phone from apps you no longer use.

• Subscriptions to services you stopped using but still pay for.

• RSS feeds, podcast subscriptions, YouTube channels you joined years ago and never check.

• Group memberships, mailing lists, and online communities you belong to but no longer participate in.

Every one of these is a subscription decay problem. Every one of these is consuming a small slice of attention every day, week, or month. None of them are catastrophic. All of them are removable.

The same one-month rule applies. Haven’t engaged in the last month? You’re not going to engage in the next month either. Cancel. Unsubscribe. Mute. Leave.

The aggregate effect of cleaning all of these up isn’t a 10x productivity gain. It’s something quieter and more important. Your attention stops being defined by the leftover commitments of decisions you made years ago. The inbox, the calendar, the notifications, and the apps… all of them start reflecting what you actually care about now, not what you happened to sign up for in 2021.

This is what justify everything you keep looks like at the maintenance level. Not a one-time elimination. A recurring practice. Every couple of months, you go through and ask, of every standing subscription, does this still serve a needle I am trying to move? If yes, keep. If no, cut.

Most things, when you ask honestly, get cut.

What to Do with the Recovered Quiet

You aren’t going to recover hours by unsubscribing from a hundred mailing lists. You’re going to recover something subtler.

The inbox will be quieter. The morning will be calmer. The deletion routine won’t be fifty rapid taps. It’ll be five emails that actually warrant attention.

That quiet is the dividend. Don’t fill it. Don’t subscribe to a new daily digest because you “miss being informed.” Don’t opt back into the marketing emails because you “might want to know about the sale.” Don’t start the new app that promises to help you keep up with the industry. The quiet isn’t a vacuum to be filled. It’s the cleared space your needle work needs.

The people I watch making real progress on AI, on innovation, on actual organizational change all share one trait. Their inboxes are unusually quiet. Their calendars are unusually open. Their notifications are unusually rare. They’ve done the elimination work, and they protect the result.

You can do the same thing. It starts this afternoon, with one sorted Deleted Items folder.

The Take-away

This week, do exactly one thing.

Open your email program. Go to your Deleted Items folder. Sort by sender, descending.

Spend 10 minutes unsubscribing. Aim for 50 unsubscribes. You’ll find more.

Then come back and tell me how many you cut. My guess is that the median reader of this article unsubscribes from somewhere between 40 and 120 senders in their first sweep.

The number will surprise you. The quiet that follows will surprise you more.

Justify everything you keep.

Eliminate everything else.

The senders crowding your inbox have been making that decision for you. Take it back.

Thanks for reading AI, Innovation, and Faith! This is Part 3 of The Elimination Project. Part 1 (the manifesto) and Part 2 (meetings) build the foundation. Tomorrow, Part 4 takes on the wider email problem — five categories of inbox content and the order to attack them.