Guide

How I Roll Out AI Coding Tools to a Skeptical Team

The mandate from above is "adopt AI." Your team has watched tool fads come and go. This is the rollout I actually run, built on evidence and volunteers instead of a mandate and a leaderboard.

Most AI rollouts I've watched up close didn't fail on the tooling. They failed on the plan around it. Somebody announced a mandate, bought seats, and waited for a productivity chart to bend. Three months later the seats were half-used, the skeptics felt vindicated, and nobody could say whether it had helped. Here is how I try to avoid that, written for the manager who has to make it work rather than the vendor selling it.

Name what a rollout is not

"Everyone should use AI" fails as a directive for a boring reason: it isn't falsifiable. If nobody can be doing it wrong, nobody can be helped, and the mandate quietly decays into box-checking or silent non-adoption. So the first thing I do is refuse to run it as a mandate. The goal is not "usage went up." The goal is that specific work gets specific leverage, and that I can see where it did and where it didn't.

Find the volunteers, not the average

Every team has two or three engineers who are already curious about the tools on their own time. Start there. Give them real seats, real time, and a standing invitation to show the rest of us what worked. A volunteer who demos a genuinely faster afternoon does more for adoption than any all-hands slide I could build. The skeptics don't need me to sell them. They need a peer they respect saying "this actually saved me an hour, here's exactly how."

The trap to avoid is measuring the average too early. The average includes people who haven't started, and it will tell you the tool is useless right when your early adopters are learning the most. Watch the leading edge first.

Pick one habit that leaves an artifact

"Use AI more" gives you nothing to see or coach. Replace it with one concrete habit that produces evidence you can actually inspect. The one I lean on: before writing code, draft an implementation plan with AI (approach, edge cases, what could go wrong) and attach that plan to the ticket. It does double duty. The file either exists on the ticket or it doesn't, so the habit is observable. And the plan itself is the engineer's real thinking on the page, which is something I can give feedback on.

One habit is not a whole practice. Reviewing with AI, test design, and debugging would each want their own inspectable artifact later. But one good habit that everyone can see beats five vague ones nobody can.

Make the feedback loop something people want to run

Periodically I pull in the plans and have AI evaluate each one on the quality of the thinking, not on whether a plan exists. Every engineer gets their own specific feedback directly. That framing matters more than the mechanics. "My manager is grading me" is a loop people avoid. "Here is a loop I can run on my own work whenever I want" is one they use. Keep the feedback genuinely useful and you also solve the gaming problem, because a loop people want to run is not one they try to trick.

Give the skeptics an honest exit

Some of your most senior people will push back, and some of that pushback is correct. AI is worse at some of your work than your engineers are, and pretending otherwise burns your credibility with exactly the people you most need. So I say it out loud: if the tool makes a task slower or worse, that is a finding, not a failure to comply. Bring it to me. The skeptic who documents where the tool falls down is doing the rollout a favor, and treating them that way is how you keep them in the conversation instead of quietly opted out.

Handle the tools and the money in the background

Two operational things will bite you if you ignore them. First, seat and usage visibility: on most plans you cannot easily see who is actually using what, and you will want that picture before renewal. Second, cost: AI spend is migrating from flat per-seat fees to metered usage, and metered usage can climb quietly. Put a meter on it early so the finance conversation later is boring. Neither of these should be visible to the team as surveillance. They are your governance job, not a scoreboard.

What I'd tell you to skip

  • The kickoff mandate. It produces nothing you can coach and it spends credibility you'll want later.
  • The leaderboard. The moment usage becomes a scoreboard, people optimize the number, and the number was never the point.
  • Waiting for the perfect metric before you start. Begin with one observable habit and a feedback loop. The measurement gets better as you go.

Take what helps

A rollout is not a purchase and an announcement. It's a habit, a way to see the habit, and a feedback loop people actually want to use, introduced to the volunteers first and extended outward as the evidence accumulates. Your team and your tools will differ. Take what helps, skip what doesn't.

If you want the longer version with the numbers attached, the measuring-adoption playbook is the same story from inside one real team.

Get new guides by email

One email when a new playbook or guide goes up, nothing else. Unsubscribe anytime.