AI on the team: from novelty to plumbing
The interesting question is no longer whether engineers use AI tools, but where they belong in the workflow — and where they quietly make things worse.
The first wave of AI adoption in engineering teams looked like everyone trying the same demo. The second wave — the one that matters — is quieter: the tools disappearing into the workflow until they're plumbing rather than novelty.
Where it earns its place
The durable wins cluster around work that is high-effort and low-stakes to verify:
- Scaffolding — tests, fixtures, migrations, the boilerplate nobody enjoys.
- First-draft comprehension — orienting in an unfamiliar service before the deep read.
- Review triage — surfacing the obvious before a human spends attention on the subtle.
In each case a person still owns the outcome. The tool compresses the distance to a first draft; it doesn't sign off on the result.
Where it quietly costs you
The risk isn't the code AI writes. It's the understanding it lets you skip.
Generated code that works is the most dangerous kind, because it removes the pressure to understand why it works. On systems where the cost of being wrong is measured in customer money, that pressure is the point. The teams getting real leverage treat AI output as a confident junior's pull request: useful, fast, and reviewed exactly as carefully.
Adoption, in the end, is a leadership question, not a tooling one. The job is to make the fast path and the responsible path the same path — and to be honest about the places where they still diverge.