Leading engineers through the AI transition

About three weeks into our rollout, one of my strongest engineers asked me in a one-on-one whether he was training his replacement. He was not the only one thinking it. He was the only one who said it out loud.

Say the quiet part first

I told the whole organization the same thing the next day: nobody on this team will be replaced by AI, and here is the roadmap that proves it. The roadmap had grown by a third since the pilot. That growth was only possible because of the capacity we had unlocked, and every item on it needed a person who understood our customers, our systems, and our failure modes.

You cannot skip this step. Engineers will not adopt a tool they believe is a layoff in disguise, and they will not tell you that is why.

Coach the judgment, not the prompt

Prompting is easy to teach and quickly obsolete. Judgment is the durable skill: knowing when the output is wrong, when the approach is wrong, and when the task should not be automated at all. Our office hours shifted from "how do I get it to do X" to "here is a diff it produced, what is wrong with it." That format produced the best conversations I have had with the team in years.

Redesign the ladder

Our career ladder rewarded volume of code shipped in a way that no longer made sense. We rewrote it around outcomes, review quality, and the ability to direct work, including AI-assisted work, toward the right result. Promotions since then have gone to people who make the whole team faster, which is what the ladder should have said all along.

Measure engagement alongside velocity

We tracked engagement survey scores against the adoption analytics. They moved together. The teams that adopted fastest reported the highest engagement, and the free-text comments explained why: less boilerplate, more time on the problems they joined to solve.

Capability gained, not chaos. That is the standard, and it is a leadership standard, not a tooling one.