What it is: The discipline of using AI yourself, visibly, on real work — including the parts that don’t go well — instead of mandating it from a distance.
Why it matters: Stanford research found 69% of CEOs use AI less than an hour a week, and most of them know their teams have noticed. BCG found companies with deeply AI-engaged leadership are 12x more likely to be top-5% performers.
How to build it: Four small, concrete habits — sharing a real AI output today, running a monthly demo, giving up an old habit in public, and naming what’s off-limits — take about thirty minutes a day combined.
About a year and a half ago I had a problem with my engineers. We tried to speed up code review with LLMs, and they didn’t want to. Not “we tried it and it’s not ready” — a quiet, deep, professional resistance. Because they’re engineers. They understood exactly how the model could be wrong, and what a confidently wrong code review could cost if anyone trusted it.
I could have announced a mandate, run a workshop, sent an encouraging Loom. Instead I started running some code reviews myself through GPT-3.5 and other LLMs, before our engineers got to them, and dumped the output into the team chat. Raw. Whatever the model said.
Some of it was garbage. Some was funny in ways we’d reference for weeks. Occasionally it caught something real — a sloppy null check, a contract mismatch. The point wasn’t that the LLM was good at code review back then. It wasn’t. The point was that the boss was using the thing, in public, including the parts where it embarrassed itself.
We track AI usage — token consumption against licensed seats, nothing creepy. Three months later, our engineers were using AI roughly three times as much. Nobody had been told to, they’d just got curious.
That’s the entire article in one paragraph. If you want your team to use AI, the highest-leverage move is not a training programme or a shiny Miro roadmap. It’s using AI yourself, visibly, including the parts that don’t go well.
The data backs it, loudly#
BCG found companies whose C-suite is deeply engaged with AI are 12x more likely to be top-5% AI performers. McKinsey tested 25 management attributes — CEO oversight of AI came out most correlated with EBIT impact. Atlassian measured a single visible leader demo increasing usage by ~90%. And Gartner found 37% of employees who don’t use AI have one reason: their coworkers aren’t. In a 30-person company, the most visible coworker is you.
Now the say != do gap. Stanford’s Nick Bloom found 69% of CEOs use AI less than an hour a week; 28% not at all. About 53% actively conceal how little they use it — “AI shame.” Most leaders endorse AI without using it. Teams know. They’ve always known (sometimes secretly using AI themselves, not always beneficially for the company).
So, endorsement isn’t information. Your team can’t see your prompt history. They can see whether your decks got better, whether you stopped manually summarising meetings, whether you ever drop an LLM output into chat without making a big deal of it. None of those signals come from a speech.
Champions, Curious, Skeptics, Resistors#
Rough industry split: 20% champions, 40% curious, 25% sceptics, 15% resistors. In tech teams there are more skeptics, and they’re more articulate — engineers know how the rose-tinted glasses are made. Their resistance isn’t anti-progress, it’s pattern-matching from years of cleaning up after someone else’s enthusiasm.
The temptation is to surround yourself with champions — please, don’t. A team of only champions is chaos — endless experiments, every shiny tool “adopted” before the previous one was understood. Skeptics are the immune system, not the internal enemy of adoption to be defeated. The curious 40% are the leverage point, and they’re watching you specifically.
What to do this week#
Use AI on something real today and share the output. A client email, a Slack thread summary, a contract review. Drop the result in a team channel with one sentence: what worked, what didn’t. Mediocre output is fine. The signal isn’t quality — it’s that the founder is publicly trying. Apply to your own visible workflow.
Run a monthly demo session. Thirty minutes, anyone shows one workflow they tried. One rule: you must say what didn’t work.
Make a visible sacrifice. Don’t add AI on top — replace something. Stop writing weekly updates by hand. Nothing convinces a skeptic like the boss giving up an old habit.
Define what’s not OK. In a small company, the leader is the governance framework. Don’t paste production configs into any model. Don’t dump the codebase into ChatGPT. Don’t share client data with free-tier tools. Three sentences everyone can recall under pressure — not a 30-page policy.
Four behaviours, maybe thirty minutes a day combined. That’s the whole skill.
This is part of the AILS Framework — 9 AI Leadership Skills for founders and leaders of growing companies. The full framework with research and exercises is publishing weekly at andrewbush.org/ails.
If this resonated, Skill 3: Process Redesign, Not Process Patching covers what to do once your team is actually engaged. For the team side of the same problem, Your Tech Team Has AI Tools. Now What? covers adoption from the engineer’s seat.