How to talk to your team about AI

The three fears your team has and isn't saying out loud. The five conversations that handle them. And the one mistake that quietly costs you a quarter of trust.

There is a meeting every CEO rolling AI into their company eventually has to run, and it's the one most of them run badly.

Picture it. You've been quietly getting leverage out of AI for a few months. Your leadership team mostly knows. The rest of the company mostly doesn't. You've got the all-hands on the calendar, and you've got a slide, maybe a single slide with three bullet points, and you've got a knot in your stomach the size of a fist. The knot is correct. Your body knows something your slide doesn't.

Here's what it knows. Three things are true in that room at the same time, and they're pulling against each other. One: this is the most important operational shift your company has made in years, and you need everyone to lean into it. Two: every person listening is, underneath the professional face, running a private calculation about whether their job is on the chopping block, whether their work will still feel like their work, and whether the craft they've spent years building still matters. Three: the trust you've banked with these people over a long time can be drawn down in the next thirty minutes faster than you built it, and trust is the one account that takes far longer to refill than to drain.

I run AI across more than one company, and I've thought about this conversation more than almost any other part of the rollout, because I'm convinced the rollout doesn't live or die on the technology. It lives or dies here, in the room, on the day you name it. Get the technology wrong and you lose some weeks. Get this conversation wrong and you lose the people who were going to make the technology worth anything.

So this is the piece I'd hand a CEO the week before that meeting. The three fears, named honestly. The five conversations that handle them. And the specific mistakes that look like kindness and are actually cowardice.

The three fears your team has and isn't saying out loud

Start by naming what's actually in the room, because everything downstream depends on getting this part right. There are three fears, they arrive together, and they're tangled. You cannot pull on one without the other two moving.

The layoff fear. Every person in that meeting is doing two pieces of math at once. First: is my job next? Second, and quieter, and somehow worse: if not me, then who, and what does it say about us that we're doing this? This is the loudest fear and the one CEOs handle worst, because the instinct is to skate past it. Don't skate past it. The fear is in the room whether or not you name it, and the only thing your silence accomplishes is to confirm that the topic is too dangerous to say out loud. A fear you won't name is a fear you've validated.

The quality fear. Your most skilled people are afraid AI will drag the work down. The designer has seen the soulless layout. The writer has read the flat, hedged, say-nothing paragraph. The person who answers your customers has seen the support reply that technically responds and actually helps no one. They have all watched bad AI output, probably more of it than you have, and the fear underneath their fear is that you haven't seen enough of it to know the difference, that you're about to trade the thing they're proud of for something faster and worse. This fear is a compliment disguised as resistance. The people who have it are the people who care about the work.

The skill-obsolescence fear. This one is the most personal and the least spoken. "Will the thing I'm good at still matter?" It splits by seniority in a way that's easy to misread. Junior people fear the ladder they were climbing is losing its rungs, that the entry-level work AI now does was the work they were going to learn on. Senior people fear something sharper: that twenty years of hard-won craft is about to become a commodity anyone can rent by the month. And here's the part CEOs get backwards: the senior fear is usually bigger, not smaller, because the senior person has more identity tied up in the craft and more to lose if it's devalued. Do not assume your veterans have it figured out. They're just better at hiding it.

The reason to name all three, in the same meeting, is that the three fears are intertwined, and addressing one while ignoring the other two is worse than addressing none of them. Reassure people about layoffs but say nothing about quality and you've told your best people you don't care about the work. Talk about quality guardrails but dodge the layoff question and the room hears you confirming the layoffs. They move together. You handle them together or you don't really handle them.

What this conversation looks like done badly

The badly-run version is seductive because every wrong move in it feels like a kind one in the moment.

It opens warm and stays vague. The CEO talks about "exciting tools" and "augmenting, not replacing" and "everyone has nothing to worry about," and the warmer and vaguer it gets, the more the room's anxiety climbs, because people can hear the shape of what isn't being said. The layoff word never comes up, which everyone notices, which means the one topic on everyone's mind is also the one topic the CEO apparently can't say. Vagueness reads as evasion. Reassurance without specifics reads as a setup.

The tempting move is to not name the layoff fear at all, because not naming it feels like protecting people from a scary word. It isn't protection. It's cowardice, and a room full of adults can smell it. The single most trust-destroying sentence available to you in that meeting is "no one's job is at risk" when it isn't true, because the moment it turns out not to be true, every other thing you said that day gets retroactively reclassified as a lie. You don't just lose the layoff conversation. You lose the whole speech, backward.

The badly-run version also tends to land in writing first. An email, a Slack post, a deck dropped in a channel before anyone's heard your voice. Writing-first is efficient and it's a disaster, because it forces every person to react alone, at their desk, with no one to check their worst interpretation against. By the time you get to the live conversation, reactions have hardened into positions, and positions are far harder to move than fresh fear.

So the done-well version inverts every one of those. It's specific where the bad one is vague. It names the hardest thing in the first few minutes instead of hoping no one notices the omission. It happens live, in voice, in a room or on a call where people can read your face, before a single written word goes out. The written follow-up comes second, as confirmation of what you already said out loud, not as the first anyone hears of it.

None of that is complicated. It's just uncomfortable, and the discomfort is the reason most CEOs reach for the warm, vague, written version instead. The whole craft here is choosing the uncomfortable, specific, spoken version on purpose.

The five conversations every CEO needs to have

A rollout isn't one announcement. It's a sequence of conversations, and skipping any of them is where the trust leaks out. Here are the five, in order.

1. The leadership pre-conversation (a month or two ahead). Your leadership team is rowing the boat, not riding in it, and they cannot find that out at the all-hands alongside everyone else. Before the company hears anything, your leaders need the shape of the rollout, your actual posture on layoffs, and the talking points, and they need to have practiced saying them. The all-hands must not be the first time your VP of whatever hears the plan, because if it is, the room will watch your leadership react in real time and conclude, correctly, that leadership wasn't aligned. Brief them early. Give them the words. Make them rehearse.

2. The all-hands announcement. Name all three fears in the first five minutes, out loud, in plain language. Commit to what you can actually commit to: the specific layoff posture, the specific quality guardrails, the specific reskilling support. Be concrete or be silent; there is no useful middle. And don't oversell the upside. The upside doesn't need your help, and piling it on while people are scared reads as a man selling them something. State the hard parts plainly, state your commitments specifically, then get out of the way.

3. The individual conversations (the first week or two after). Anyone whose role is genuinely changing by this deserves a real one-on-one, with you or with their manager if you can't scale to all of them, where the specifics of their situation get walked through. Not a hand raised in a town hall. A door closed, a real conversation, the actual answer to "what does this mean for me." This is the conversation that costs you the most time and buys you the most trust, because it's the one that proves the all-hands wasn't theater.

4. The return (around the ninety-day mark). At the all-hands, tell people exactly what you'll revisit and when, and then actually revisit it. The act of coming back is itself the trust-builder, almost regardless of what you report, because most CEOs announce and vanish, and a team that's been through a few of those has learned that "we'll revisit this" means "this is the last you'll hear of it." Be the CEO who comes back. The contrast alone does most of the work.

5. The role-redesign conversation (a few months in). By month four or five you can see how roles have actually shifted, which means you can do the thing you couldn't honestly do on day one: rewrite the role descriptions, adjust comp where the work genuinely changed, and show the team that the rollout produced real decisions about how work is done now, not just a speech. This is where people find out whether the whole thing was leverage for them or leverage against them, and the answer needs to be visible, not implied.

One line ties the five together. Skipping any of them is the version where you save thirty minutes today and pay it back as thirty hours of trust repair over the next quarter (and trust repair is the most expensive work there is, because it's slow and you can't delegate it). Run all five. The sequence is the product.

Common mistakes

Past the big ones, here are the quieter failure modes, the ones that look fine from the CEO's chair and rot the rollout from underneath.

Announcing in writing first. Covered above, and worth repeating because it's the most common and the most avoidable. Live first. Written second. Always.

Refusing to name layoffs at all. If you genuinely will never lay anyone off over this, say so, plainly, and mean it, and then live up to it. If you might, say what would trigger it. The thing that destroys trust isn't bad news; it's the discovery that you managed the message instead of telling the truth. People can carry hard, honest news. They cannot carry the feeling that you were handling them.

Letting leadership freelance the message. Without explicit, rehearsed talking points, your leaders will each say something slightly different, your team will compare notes within the week, and the small inconsistencies will read as a cover-up even when they're just improvisation. Brief them. Give them the exact words. Have them practice out loud, because a talking point you've only read is a talking point you'll mangle under a real question.

Treating it as one event. The announcement is the first conversation, not the last. The CEO who never returns to the topic looks, in hindsight, like a CEO who hoped everyone would forget. Returning is the whole game.

Underestimating the senior-talent fear. It's worth its own line because it's the one most often missed. The instinct is to worry about junior anxiety and assume the veterans are fine. They're not. They have the most identity in the craft and the most to lose if it's commoditized, and they're the best in the building at hiding it behind competence. Talk to your seniors specifically. Don't wait for them to raise it, because they won't.

Where this fits in your operating system

This piece is the human half of the rollout. The other half is the actual stack you're rolling out, and the two halves only work together.

If you want the technical foundation under all of this, the case for running AI in the terminal instead of chat tabs is the pillar I'd start with, and the CLAUDE.md explainer and setup walk-through are where a real install begins. Skills are the unit that turns a one-time workflow into something repeatable; the explainer and the build-your-first-one walk cover that. The meeting-capture pillar is the highest-ROI place most CEOs start. And once the rollout is running, the strategic frame for what compounding actually looks like is in the autonomous, self-improving company.

But start here, with the conversation, because it comes first in time and matters most. Run the leadership pre-conversation this week. Tell the people rowing the boat what's coming, before anyone else hears a word of it. The thirty minutes you spend on that one conversation is the highest-leverage thirty minutes of the entire rollout, and it's the one nobody puts on the calendar.

Run it. Then tell me how it went. I read every reply, and the rollout stories, the ones that went sideways and the ones that landed, are the ones I learn the most from.

Andrew

Related reading

• Why CEOs should use Claude Code in the terminal · the technical foundation under the rollout
• What is a CLAUDE.md file? · where a real install begins
• How to set up your CLAUDE.md file · the thirty-minute setup
• What are skills in Claude Code? · the unit of repeatable leverage
• How to make a skill in Claude Code · build your first one
• Granola for CEOs: the highest-ROI AI install of 2026 · the best place to start the technical half
• The autonomous, self-improving company · the strategic frame for what comes after the install
• How to get your team to actually use AI · the sequel: once you've announced it, how to drive real adoption