How to get your team to actually use AI

You gave the speech. You bought the seats. Three weeks later nobody's using it. The gap isn't access. It's adoption, and adoption follows the person at the top.

You did the hard part already. You ran the all-hands, you named the layoff fear out loud, you signed the order for the licenses. If you haven't had that conversation yet, go run it first: how to talk to your team about AI is the prequel to this piece.

Then the speech ends, the seats get provisioned, and three weeks later you check the usage dashboard. A handful of power users. Everyone else logged in once and never came back. The tools are sitting there. Nobody's using them.

That's the moment this piece is for.

The honest answer

Access is not usage, and you just paid for access.

IBM's 2026 CEO study of 2,000 leaders found the gap in one stat: 86% of CEOs believe their people already have the skills to work with AI, while only 25% of the workforce uses it regularly on the job. The same study found 83% of CEOs say success depends more on adoption than on the technology itself.

Buying the tool moves a line item. It does not move behavior. The work that closes the gap starts the day after the purchase order clears, and almost nobody schedules it.

Why it's harder than it looks

The instinct is to assume your people are refusing AI. They're mostly not. They're using it in secret and afraid to admit it.

Microsoft's 2026 Work Trend Index found most knowledge workers now use AI, but around 78% bring their own tools to work ("shadow AI"), shadow-AI surveys put unauthorized-tool use around two-thirds of employees. The tell is in the fear numbers: roughly 52% hide that they used AI on their most important work, and about 53% worry it makes them look replaceable. That is not a refusal problem. That is a permission problem, and permission comes from you.

Then there's the money trap. MIT reported in 2025 that around 95% of enterprise generative-AI pilots showed no measurable P&L impact, and the root cause was a learning and integration gap, not weak models. The models were fine. The org never did the work to put them inside the actual job.

There's a clean way to hold all of this. Call it the 10-20-70 split: success is roughly 10% the algorithm, 20% the tech and data, and 70% people and process. Most companies invest in the exact inverse, all budget on the 30%, nothing on the 70%, then act surprised when nothing changes.

The specific reasons adoption stalls are boring and fixable:

• No concrete use case tied to the person's actual job. "Use AI" is not a task.
• Workflow friction. The AI lives in a separate tab, not in the system they already work in. A separate tab is a separate decision every time.
• No leadership role-modeling. If the CEO endorses it but never uses it in front of anyone, the team reads the real signal.
• The one-shot quit. They tried it once, it got something wrong, and they decided it doesn't work. (That failure is usually a setup problem, not a model problem: why AI agents fall apart in real work covers it.)
• Fear. The 52% and 53% above. Nobody learns in public when they think the learning makes them look disposable.
• No protected time. You can't pick up a new tool in the cracks between meetings, and neither can they.

What to do this week

You don't need a mandate. You need to make AI the path of least resistance for one real task, and you need to go first. Seven moves, in order of payoff:

• Go first, in public. Don't endorse AI. Use it where people can see, and share your own prompts and your own wins. The team copies the CEO's behavior, not the CEO's memo.
• Pick ONE high-frequency task per role and make AI the default path for it. Not a new tool to adopt. The default way that one task already gets done. Pull the task from the menu in ten things they can start this week and assign one per team.
• Use champions, not webinars. One person per team who's genuinely into it, given a couple of hours a month to teach peers. Peer teaching beats a top-down training session every time, because it happens in the actual workflow.
• Protect the learning time. A recurring hour. An "AI Friday." A lunch-and-learn. If it isn't on the calendar, it doesn't happen.
• Measure outcomes and real usage, not license counts. Seats bought is a vanity number. Track the task that got faster, the hours that came back, and celebrate concrete wins by name and out loud.
• Make it safe to share failures. This is the direct counter to the 52% who hide it. When you publicly show a prompt that returned garbage, you give everyone permission to be a beginner.
• Bake it into rituals you already run. Onboarding, standups, weekly reviews. The moment AI is a separate initiative, it competes with the day job and loses. Folded into existing rituals, it's just how the work gets done.

The laggard problem

There will be holdouts. Some 2026 companies are answering with mandates: Shopify's 2025 memo made "reflexive AI use" a baseline expectation, Duolingo went AI-first, and Starbucks tied a slice of tech bonuses to AI-adoption goals (reported May 2026). A mandate can be honest, and for some teams it's the right forcing function.

But a mandate forces a number, and a number invites theater. Amazon scrapped an internal AI-usage leaderboard in 2026 after employees gamed it, assigning agents pointless tasks to juice their scores (nicknamed "tokenmaxxing") and spiking compute costs in the process. An Amazon exec's line was basically: don't use AI just for the sake of using AI. As one operator put it on X: "A lot of 'AI adoption' isn't adoption. It's workplace theater. Employees must sound excited, managers must report gains."

The durable fix for laggards isn't a quota. It's making AI so useful to their actual workflow that NOT using it becomes the harder choice. You don't have to push someone who's already saving an hour a day.

Want to maximize your AI leverage? Upgrade to Pro.

This week, do the thing you're asking them to do. Pick one task you personally run every week, do it through AI in front of your team, and share the prompt and the result, including where it got things wrong. Then ask each of your leaders to pick ONE task their team will make AI the default for. Adoption follows the person at the top, not the mandate. Then tell me what changed. I read every reply.

Andrew

Related

• How to talk to your team about AI
• Ten ways CEOs can use AI this week
• What is an AI agent?
• Why AI agents fall apart in real work