On this page
The newsletter that worked and then didn't
Eighteen months ago, a respected executive director made what looked like a reasonable call. His newsletter had a small, serious readership. Two thousand names, mostly senior practitioners in his field, several of them longtime donors. The list grew slowly. It was expensive, in staff hours, to write the weekly note he personally signed. A consultant suggested that AI-assisted drafting could let him publish three times a week instead of once, at roughly the same cost. The case was clean. More surface area. More touchpoints. More of his voice in the feed of the people who mattered.
By the end of the first quarter, output had tripled. Open rates held. Click-through held. The dashboard reported, cleanly, a three-fold increase in engaged impressions. He mentioned the success to his board, carefully, as one data point among many.
Two quarters in, the dashboard still looked fine. The interior of the list did not. Roughly forty percent of the names he most cared about — the long-tenured senior practitioners, the ones whose private replies had once made the newsletter worth writing — had silently stopped opening. A handful had unsubscribed. Two had sent a short message he still has not answered. The shorter, more frequent pieces tested well inside the list and tested poorly among the people he had actually written the newsletter for in the first place.
His output had gone up. His impact, as he had originally meant the word, had gone down. He had not noticed the trade, because the instrument he was measuring with could not see it.
This is the single most common trade happening across mission-driven organizations right now, and it is almost always invisible on the dashboards that were supposed to tell leaders whether things were working.
What integrity actually means here
Integrity is the wrong word if it only means moral performance. Most leaders hear integrity and reach for a values conversation. That is not what this piece means by the term, and the confusion is part of the trap.
Integrity, in the load-bearing sense, is structural coherence. It is the alignment between what your organization says, what your organization ships, how your organization ships it, and who your organization is. Four layers: message, mission, medium, maker. When those four are coherent, a serious reader or donor can recognize the organization across any surface. A board deck reads like the annual report, which reads like the sermon, which reads like the donor letter, which reads like the consulting memo. Different forms. Recognizably the same organization.
When the coherence breaks, the organization becomes illegible to the people it was built for. The cost of that illegibility is not emotional. It is structural. Donors who can no longer recognize the organization begin giving elsewhere, quietly, without ever articulating the reason. Staff who can no longer recognize the organization lose the internal compass that made their work efficient, and start asking in meetings what they used to answer alone. Partners who can no longer recognize the organization stop introducing it to their own networks. None of these are catastrophic events. All of them are compounding losses, and they tend to show up on the books about two years after the coherence first started to slip.
This is the definition that matters for the rest of this piece. Integrity is coherence across message, mission, medium, and maker — the thing that makes an organization recognizable to itself and to the people who care about it — and it is the thing AI puts under specific pressure, in specific ways, that most scorecards were not designed to detect.
What impact actually means here
Impact is also the wrong word if it only means reach. Most scorecards track reach: open rates, impressions, attendance numbers, subscriber counts, social followers, conference registrations. These are real numbers. They are also, individually, poor proxies for the thing the organization was founded to produce.
Impact, in the load-bearing sense, is the rate at which the organization's actual work reshapes the lives and systems it was called to reshape. Not the number of people who saw the video. The number of people whose decisions changed because of it. Not the size of the cohort. The number of cohort members whose vocational trajectory shifted. Not the attendance at the conference. The number of attendees who lead differently a year later as a direct result of what happened in the room.
Impact, measured this way, is expensive to observe and easy to miss. Most organizations track its proxies rather than the thing itself, because the proxies are legible weekly and the thing itself is legible in years. The problem is that the proxies and the thing can diverge. A newsletter can produce more impressions and less decision change. A cohort can grow larger and shift fewer trajectories. A conference can sell out and change nothing. When the proxies rise and the thing itself falls, the organization is louder and less useful — and the leadership team, reading the proxies, will often celebrate the first without noticing the second.
This is the structural pairing that matters. Integrity is coherence across surfaces. Impact is the rate at which the work reshapes the reality it was aimed at. Both are real. Neither is the other. Most organizations quietly optimize for proxies of impact at the expense of integrity, and they do it without ever holding a meeting in which the trade is named out loud.
AI has repriced them both
What AI has done to this pair is the core of the tension. It has not made either one impossible. It has repriced them sharply, in opposite directions, and the pricing asymmetry is the reason the trade is now happening so fast.
AI has made the proxies of impact radically cheaper to produce. Output can multiply in weeks rather than years. A communications team that once shipped one newsletter a week can ship three. A research function that once produced two memos a month can produce ten. A discipleship team that once wrote one curriculum in a year can ship three variants in a quarter. The cost curve of producing the surface of the work has dropped, and the dashboards that track surface will reliably show the drop as a win.
AI has made the coherence that produces integrity significantly harder to maintain. Not because AI is incapable of coherent output — that is a separate conversation — but because the tempo that AI makes available outruns the tempo at which a human leadership team can verify coherence. A team shipping three pieces a week at the level they used to ship one cannot also read each piece with the attention that produced the coherence in the first place. The organization's voice, which used to be held by a small number of people who could still read everything, begins to drift — not because anyone decided to drift, but because no one had the hours to hold the line.
The asymmetry sharpens over quarters. Each quarter of faster output with the same leadership review capacity produces a little more drift in voice, a little more mismatch between message and maker, a little more fuzz in the donor's mental model. The scorecard shows growth. The interior shows loss. The trade is real, and it is running in one direction, and almost no organization is calibrated to see it.
This is the load-bearing sentence of the piece. You can ten-x the output of a discipleship team in a month. You cannot ten-x the discipleship. The first number is a production number. The second number is a formation number. They are priced differently, and AI has only moved one of them.
Two failure modes, one thinning
Two kinds of organization will not make it through the next decade well.
The first kind will hold integrity by refusing AI. Their output stays modest. Their voice stays intact. Their best staff, who know the terrain has shifted, will either leave for organizations that are using the new capability or stay and quietly burn out carrying a workload that has not been re-scoped. Their donors, whose attention is being shaped by every adjacent organization's increased presence, will drift toward noisier peers whose output volume makes them seem more alive. The integrity will remain coherent. The organization will shrink. In five years it will be mentioned less often in the rooms that used to mention it, not because it has betrayed its mission but because it has become invisible to the people it was trying to serve.
The second kind will chase impact by adopting AI without protecting integrity. Their output will rise. Their scorecards will look healthy. Their voice will slowly converge on a sector-wide median that reads smooth and forgettable. The serious readers will unsubscribe silently. The serious donors will take longer to return calls. The staff will produce more and believe in it less. In five years this organization will still exist, at larger scale, producing more content than ever, and it will no longer be the kind of organization whose work shapes decisions. It will be a content operation with a mission statement attached.
Both organizations are losing. One looks healthier on the way down because the volume is rising. The other looks less embarrassed on the way down because the craft is intact. Neither is the example this book is written for.
The trade is only short-term
The argument this piece is making is not that integrity and impact are unreconcilable. It is that they are only tradable in the short term. In the long run, the organizations that compound are the ones that built AI practice on top of their integrity rather than in place of it. That sequence is slower at the start. It is much faster once the foundation holds.
The leaders who get this right do three things the two failure modes do not. They refuse to trade integrity for impact even when the short-term metrics make the trade look rational. They refuse to abandon impact in the name of integrity, because they know that integrity without impact is another word for irrelevance. They build a structure that lets both grow together, on a timeline the quarterly dashboard will not reward.
That structure is the subject of the rest of this book. It has a shape. It proceeds in an order. Its first step is not a tool. It is a frame, and it is the frame most organizations are currently skipping — partly because no one in the sector is organized to notice it, and partly because the obvious default move, put the most technical person in charge of AI adoption, is precisely the move that guarantees the tension in this piece will be badly handled.
The next piece names that default move and explains why it is a category error. AI adoption is not a tools problem. It is a leadership, formation, and human problem. Anyone who treats it as a tools problem has already, without meaning to, chosen one of the two failure modes in this essay. That is where we go next.
Read next: This Is Not a Tools Problem — why "get the tech team involved" is the wrong first move, and what the right first move actually is.