Alternative titles considered: "Guides, Not Gurus: What Trust Looks Like in the Age of Generative AI" · "When No One Grew Up With These Tools: A Discernment Guide for Mission Organizations"
Finding AI Guidance Worth Trusting
A discernment guide for mission-driven organizations choosing help in a moment no one has mastered yet. Revision 1. A cluster spoke of There is no such thing as AI safety.
The real question is not which tool
A nonprofit executive director I respect told me recently that she had been pitched by four consultants in a month, each claiming to be "the person" on AI for mission-driven work. Two had slide decks that looked nearly identical. One had pedigrees she couldn't have afforded five years ago and could apparently afford now. One was somebody her board member had forwarded because they had seen him on a podcast. She had the budget to hire one of them. She did not have the frame to decide which.
Her question was not really about tools. She already had tools. Her question was not really about strategy, either. She had read enough strategy decks to recognize when one had been bolted together from LinkedIn posts and conference summaries. Her question was the one mission-driven leaders keep arriving at, usually in private, often after a few expensive months they don't want to talk about publicly:
Who can I actually trust to walk with us through this?
I should tell you now that I am one of the four.
Not in her month, and I am not going to say which organization or which conversation. But I sell exactly this. I build AI systems for organizations whose work runs on trust and I sell the training that goes with them, and when a director like that one is deciding between four people, I am the kind of person she is deciding between. Everything below is a set of criteria for choosing someone like me, written by someone like me, and I have an obvious interest in where the criteria land.
So read the rest adversarially. When a test I propose happens to be one I would pass, that is exactly when you should push on it, and further down I have marked the place where I think you should push hardest.
That question is harder than it sounds. The market is not short on expertise claims. It is not short on confidence, credentials, or carefully staged authority. What it is short on is people who can admit, out loud, that this is a new moment for everyone. and who have still done the patient work of being useful inside the newness. The inflation of AI expertise has made credibility paradoxically harder to read, not easier. Every vendor, every consultant, every solo operator has the same aesthetic package now. The confidence is distributed. The competence is not.
I want to say as plainly as I can: the choice mission-driven organizations are making right now is not primarily a tools choice or a vendor choice. It is a trust choice. You are choosing who to entrust yourself to in a period where no one. including me. has decades of experience in the stack you are being asked to govern. The quality of that choice will shape what your people become, not just what they ship.
This piece is about how to make that choice well.
We are early, and that matters
It helps to say the boring fact first. Modern consumer-facing generative assistants, at anything like today's scale, are recent. ChatGPT's public release in November 2022 is a useful time marker. not because the calendar proves anything by itself, but because the human consequence is real: outside a handful of specialized labs, none of us. including the most visible voices in the field. has more than a few years of lived practice in this stack.
There is no elder class for this. There is no thirty-year veteran of generative collaboration the way there is a thirty-year veteran of accounting software or database design. That is not romantic. It is descriptive.
Two things follow.
The first is that humility is mandatory. Anyone selling you the tone of having figured this out is, at minimum, ahead of what the evidence supports. That does not make them dishonest. It often makes them performative in a way the moment will quietly correct for them later.
The second is more useful: experience still exists, but the experience that matters here is experience of grappling, not experience of having arrived. The people worth trusting are the ones who have built, broken, revised, and stayed in relationship while doing so. Their maps are drawn in pencil. They revise them publicly. They tell you what they tried last year that they would not try this year, and why.
Credentials from 2015 are not disqualifying. They are also not predictive of help with the particular work your organization is actually trying to do in 2026. The question to hold is not does this person have a resume? but does this person have a current practice I can examine?
Why default consulting logic fails here
Most organizations, when they meet AI as a leadership problem, reach for the pattern that has worked for every technology adoption of the last twenty years. The pattern goes roughly: form a committee, hire a consultant, pilot a tool, run a training, write a policy. It has delivered cloud migrations. It has delivered CRM rollouts. It has delivered email marketing infrastructure. The leaders running it are not naive; they are doing what has worked.
AI breaks the pattern, and the consulting economy that is rushing to meet the moment is, in many cases, breaking it further.
Here is what I mean. The part of the economy that sells AI has every incentive to start the conversation at Solutions. at the tool, the pilot, the deployment. Demos are vivid. Pilots fit fiscal years. Screenshots of outputs circulate on boards. The quieter work. governance boundaries that actually bind, structured experiments that produce shared language, formation that forms staff judgment rather than staff speed. does not screenshot well. Nobody takes a photo of a disagreement that was resolved before a donor letter went out.
So the default consulting logic tends to sell you the visible end of the sequence while skipping the parts of it that do the moral and organizational work. You end up with a tool portfolio. You do not end up with a capability. A year in, the licenses are active, the staff have been trained, the dashboards look right, and the organization's distinctive voice. the voice it took a decade to build. has quietly repriced itself toward the industry mean. Nobody can point to the meeting where that happened. That is because the problem was never in a meeting. The problem was upstream of the meeting.
The people who can help you here are people who will resist starting at Solutions. They will ask, early and without embarrassment, about what you are trying to protect, what your leadership actually owns, and what formation your staff is undergoing whether you named it or not. If the first two conversations are about tools, the third will also be about tools, and so will the twelfth. That is not a minor cultural preference. It is the shape of the engagement.
This is not a tools problem. A consultant who cannot say that sentence and mean it is a consultant whose incentives are pointed elsewhere.
Expert versus guide
This is the distinction that, more than any other, I want mission-driven leaders to carry into their decision.
An expert, in the popular imagination, is someone who has arrived. They have answers. They have confidence. They have a slide for every fear you have about AI, and several for fears you did not know to have. An expert speaks from the position of I have done this. In a stable domain, that posture is earned and useful. In this domain, in 2026, it is often a performance.
A guide is someone who is walking the territory still. Their maps are in pencil because the ground is moving faster than ink can dry. They do not have a slide for every fear, because they know which fears are real and which ones are vendor-manufactured. They speak from the position of I am still inside this, and so are you; here is what I have learned; here is what I do not yet know. They are translators more than oracles. Their first instinct is to ask what your room looks like before telling you what the room should do.
If the AI moment were stable, you would want an expert. Because it is not, you want a guide.
The difference is behavioral, not cosmetic.
Experts, in the sense I am using the word, arrive. They present. They explain. They close. They are very hard to interrupt productively, because the architecture of their authority depends on continuity. the deck has to run. When something goes wrong with their advice, they tend to reframe it as an implementation problem. When a client's mission shifts under them, they tend to restate the framework.
Guides walk. They ask what it is like to lead inside your particular sector, inside your board culture, inside your theological or programmatic commitments. They tell you, early, about things they got wrong somewhere else and would not repeat here. They decline to answer questions they do not have grounds to answer, and point you toward someone else who does. They are interruptible on purpose; the authority is in the walking, not in the speech.
Mission-driven organizations. nonprofits, churches, seminaries, institutions carrying inherited work. need guides more than they need experts right now, because the failure modes of AI adoption in mission-driven contexts are not failures of technique. They are failures of judgment, pace, and formation. Those cannot be imported in a deck. They have to be grown in place, with accompaniment.
The test is not whether the person on the call is smart. Most of them are smart. The test is whether the person on the call will still be useful to you after the engagement ends, when a model changes, a staff member uses a tool badly, a donor asks a pointed question, and you have to respond in your own voice without the consultant in the room.
What I actually look for
This is not a rubric. It is a pattern I have learned to read, usually in the first two or three conversations with someone who is proposing to help an organization I care about. Take what is useful; discard what is not.
Evidence of grappling. Fluency held next to honest wrestling. Real wins you can name, and real revisions the person does not hide. People who can say "I used to think X; after the last eighteen months of building this, I now think Y, and here is why" are showing you the thing that matters. People whose opinions have not visibly moved since 2023 are showing you the absence of the thing that matters.
Evidence of building. Not slides about building. Something shipped. Something maintained. Something reviewed by others who can be named. Ask to see it. Ask who else has touched it. Ask what the failure modes were. If the answer is polished in a way that makes failure sound exciting and educational rather than costly and specific, you are probably looking at marketing.
Resistance to totalizing answers. "It depends" is a trustworthy sentence when it is followed by a clear list of what it depends on. "It depends" as a rhetorical hedge, without the dependencies named, is a different and less useful move. Guides can articulate the shape of a decision even when they cannot prescribe the decision. Experts, in the performative sense, tend to skip the shape and go straight to the prescription.
Stewardship of pace. The trustworthy people I know treat learning and publishing as distinct activities. They do not push every experiment out as a press release. They know that early learnings, when staged as expertise, borrow credibility they have not yet earned and spend credibility the client has not yet given. Someone who treats every experiment as a LinkedIn post is not helping your credibility; they are underwriting their own platform on the back of yours.
Accountability to named relationships. Who corrects this person? Whose feedback have they incorporated publicly? Who would catch them if they drifted? In a field changing this fast, isolated authority is not authority; it is fragility wearing a brand. Someone whose peer group is visible, whose intellectual lineage is traceable, and whose mistakes get surfaced by people they respect is more trustworthy than someone whose public image is managed as a single-author product.
Fit to mission, not just sector. A consultant who understands nonprofit operations is not the same as a consultant who understands your organization's specific work, its theology or public commitments, its donor covenant, its history of promises kept. Ask how they would know when a draft betrayed your mission. If they cannot answer that in your language, they will not recognize the drift when it starts.
Willingness to disappear. The best guides I know plan their own exit from the start. They are working toward the moment when the organization can propose, sandbox, deploy, and retire an AI use case without them in the room. Consultants whose engagements require their continued presence to remain stable are, functionally, renting you capability rather than building it.
None of these, by itself, is a rule. Taken together, they describe a person you can walk with. That is different from a person you can buy.
The network test
One of the most reliable signals of credibility in a fast-moving field is not individual. it is relational.
Brian Eno's word for this is scenius: the genius that emerges from a scene, the collective intelligence of a community of practice whose members sharpen, correct, and cite each other over time. Scenius is not a platform trick. It is a slow social fact. It is what makes a group of people collectively wiser than any of them could be alone, and it is what makes a field self-correcting when any single member gets something wrong.
The reason this matters for your trust choice is simple. When models are changing every few months and no one has thirty years of practice in the stack, isolated authority is fragile. The person who is brilliant alone will get something important wrong and may not notice until you notice, and by then your organization has spent real credibility on it. The person who is embedded in a network of aligned peers. people who will call them out, build on them, disagree with them, cite them, revise them. has a different kind of resilience. Their mistakes get caught. Their language sharpens in public. Their blind spots are covered by people whose blind spots are different.
This is not the same as follower counts. Follower counts are performance metrics; they measure reach, not correction. Scenius measures correction. The question to ask is not "how many people follow this person?" but something more like: "who engages with them in ways that would be embarrassing to fake? Who publicly disagrees with them, gets incorporated, and comes back? Who do they name when they say they have changed their mind?"
For mission-driven leaders, there is a second reason the network test matters. Your organization is also a node in a trust graph. The people who vouch for the consultant you hire will end up, functionally, vouching for the consultant to your staff, your board, your donors. You are not just buying an individual engagement; you are importing a trust network. Make sure you would be comfortable being introduced to the rest of it.
Ask, directly: who corrects you? whose work sharpens yours? who would tell me if you had missed something important for my kind of organization? The answer to that question, more than any biography, tells you what you are actually choosing.
And here is where you should push hardest on me.† I build a product whose entire premise is networked credibility. A test that says choose the person embedded in a network of peers is one I designed my company to pass, and I am the one proposing it as a criterion. That does not make it wrong. It does mean you should ask whether a solo practitioner with no visible network could still be the right person for you, and the honest answer is that they sometimes are, particularly if they have long relationships in your sector that never took a public form. The network test is a signal, not a filter. I would not hire on it alone and neither should you.
The hard test
Most of what I have written so far could be extracted into a list of things to look for. The list would be useful. It would also, on its own, flatten the thing I most want mission-driven leaders to hear.
So here is the test I would not let go of, even if you forgot everything else.
Who is willing to lose a little reach to protect your formation?
That is the question that separates guides from the commodity market of AI advisors. Almost every operator you talk to has reach as an implicit objective. their reach, your reach, the reach of the tools they are selling. Reach is the default metric. Reach is how most of us, honestly, measure whether a thing is working. Reach, in the AI era, is also the metric most easily inflated and most easily hollowed out.
Formation is different. Formation is what your staff become by using these tools. It is what your leaders become by making these decisions. It is what your organization's voice becomes after a year of drafts in which the model was invited to the table. Formation does not screenshot. It does not raise money. It does not appear in any dashboard vendors sell. It is, however, the actual thing that determines whether your organization is still itself in five years.
A guide worth trusting will, at some point, tell you to slow down in a way that costs them. They will recommend against a tool they could have sold you. They will suggest your board have a conversation before the procurement. They will decline a piece of work because they do not yet have the evidence to do it well. They will name the category of staff formation you are avoiding because it is expensive and not visible. They will, in effect, take less reach. and less money. in order to protect something in your organization that only you can see.
If you cannot identify a moment in the conversation where the person across from you chose your formation over their platform, you do not yet know whether you are looking at a guide. You are looking at someone very capable who has not yet been tested in the specific way that matters.
This is not a sentimentality. It is the hardest piece of leverage you have in this market. Use it.
Where my own work fits. briefly, and without overclaiming
I said at the top that I am one of the people you would be choosing between. Here is the shape of what I sell, so you can apply the tests above to it. What I can say with some confidence, because I have been accountable to leaders whose life's work can be harmed by my mistakes, is that the questions above are the ones I try to be on the receiving end of.
The frame we work inside is the AI Stewardship Sequence: Safety → Sandbox → Skills → Solutions. Safety first means the organization's leadership knows, in plain language, what is forbidden and what is allowed before a vendor contract puts pressure on the answer. Sandbox second means the organization accumulates evidence before it accumulates certainty. structured experimentation on non-critical work, with dated notes and observed failure modes. Skills third means staff judgment is formed before scale arrives, not trained after the tools are already embedded. Solutions last means deployment happens into an organization that can course-correct, because the upstream work is in place.
Most organizations meet this sequence backwards. The market has every incentive to sell them Solutions first. The inversion feels productive. It produces artifacts, meetings, the sensation of keeping pace. It almost never survives the first serious review.
I offer the sequence not as a sacrament but as a case study. The principles. amplification over fabrication, voice preservation, paced experimentation, boundaries on what we automate, networked credibility as a design aspiration rather than a magic spell. should be testable outside any one project. If they are not, I would rather you tested them somewhere else than took them on my word.
What I am not offering is certainty, or a single stack that solves your ethics for you, or immunity from error. I have been wrong, sometimes expensively, sometimes publicly. I expect to be wrong again. The honest version of this work is that you do not have to trust me yet. You only have to be willing to test the ideas in your actual room.
A short discernment layer (questions, flags, signals)
I said earlier that a checklist is a derivative of the essay, not a substitute for it. What follows is meant to be used that way. not as a rubric you grade people against in a single meeting, but as the rough shape of a discernment process you can carry into a few conversations and a few references.
Questions worth asking out loud
- What did you believe about AI and our kind of organization eighteen months ago that you no longer believe, and why?
- Describe something you have built or helped build in this space that is still running. Who else touched it? Who can I talk to?
- What would you refuse to do for us, even if we asked for it and offered to pay?
- Who corrects your thinking in this field? Whose public disagreement have you incorporated in the last year?
- What does formation look like in this engagement, and how would we know it is happening?
- If our senior leader had to describe our relationship to AI in one paragraph six months from now, what would the paragraph sound like, and who would have written it. them, or you?
- When should we stop working with you, and how will we know?
Red flags
- Solutions-first from the opening conversation. Tools, pilots, platforms on the table before governance, staff formation, or mission fit have been seriously discussed.
- A single-author public image. Nobody publicly disagrees with them, sharpens them, or corrects them. Their authority looks managed rather than lived.
- Every experiment is already content. Learning and publishing are not distinguished; every client engagement seems to produce a case study or a post before the ink is dry.
- Unmoved opinions. Views on AI in your sector have not visibly updated despite eighteen months of new evidence. Confidence unchanged since 2023 is a signal, not a reassurance.
- No named relationships. They cannot or will not tell you who else they are accountable to, who they would send you to for a different kind of problem, or who would catch them if they drifted.
Green flags
- They name things they got wrong without being asked, and name them specifically enough that you believe them.
- They push back on your framing before they sell their solution. and the pushback is in service of your mission, not their engagement.
- They recommend against things you came in asking for, including, sometimes, the engagement itself.
- They distinguish pace from speed. They can explain why slower in the first quarter is often faster in the third.
- They plan their own exit. The engagement is designed to make your staff capable of running the work without them in twelve to eighteen months.
None of these, individually, decides the question. Taken together, they describe the person I would want walking with an organization I cared about through this particular moment. Not a savior. Not a celebrity. A guide, whose maps are in pencil, who is still inside the work, and who is willing to lose a little reach to protect your formation.
A closing word
You do not have to trust me. You do not have to trust any particular voice in the room, including the most famous ones. What you do have to do is choose carefully, because the choice you are making in this moment is not really about tools or strategies. It is about whose judgment will be quietly shaping your staff's judgment, whose posture will be shaping your leadership's posture, whose voice will be showing up, a little bit, in every draft your organization sends into the world for the next several years.
That is a bigger choice than the invoice makes it look.
Choose a guide. Choose someone who will walk with you, who is accountable to people you can name, and who will protect the parts of your organization that do not show up on a dashboard. because those are the parts that, five years from now, will decide whether your mission was amplified by AI or quietly replaced by it.
We are early. Nobody has this figured out. That is the honest ground. Standing on it clearly is itself a form of authority. Walking on it with you is the only kind of help worth buying.
† No source. This is a judgment, not a fact, and it is a judgment against my own interest, which does not make it more reliable. Weigh it as a judgment.
Before this publishes
Two decisions. Confirm the executive director is comfortable with the opening scene as told, even unattributed, since she would recognize herself. And decide whether the AI Stewardship Sequence should be named here at all, or whether naming the product's own framework inside a discernment guide is one step past what the disclosure covers.
One tightening note. At roughly four thousand words this is long for a cluster spoke. The discernment layer at the end could stand alone as the free diagnostic tool, and the essay would be better at three thousand.
Related reading: There is no such thing as AI safety, which is the argument underneath this one · What changed in your insurance while you were not looking, which is the dated version of why this decision got more expensive this year · Why five layers, and what I am adding