A year ago, if you'd asked me what an engineer does, I would have said something about writing code. I would not say that now.
What changed is not which tools we use. It is what we expect an engineer to be. A year ago we still thought of AI as a tool an engineer reached for, the way you reach for a debugger. By the end of the year we did not think of it that way at all. AI is not in the toolbelt. AI is in the work. The engineer who is fastest is not the engineer with the best AI tooling on the side. It is the engineer whose entire workflow assumes the AI is there.
That changed what our team actually spends its time on. Writing code is, at this point, probably the smallest part of the job. The job is understanding the space we are working in, identifying which gap is the biggest one to close, deciding what the MVP looks like, and figuring out how fast we can iterate on it. The implementation is the easy part. The judgment about what to implement is the entire job.
Take a concrete case. We used to start most mornings triaging Sentry alerts — reading each one, finding the code path, querying the database to see how many records it affected, deciding whether it was urgent. That used to be the first hour of the day for whoever was on call. Now an AI runs that loop overnight and leaves a Slack summary by the time the team logs in. The morning starts with the ten-minute decision about which fixes are worth shipping today, not the hour of context-loading that used to come before the decision. That hour is gone. Nobody misses it.
We used to think being "specialized in a language" was a meaningful identity for an engineer. It isn't, not anymore. If you have the fundamentals and a solid foundation, you can move at speeds that would have been absurd a year ago — in any stack, on any problem. The people who built their identity around being the best Ruby developer, or the best React developer, are about to discover that nobody is paying for that distinction the way they used to. The value has moved upstream, into the question of what to build and why.
This is what ties our internal AI work to everything we are building next. The new tools we ship — for our customers, for our team, for ourselves — should have AI at the core, not bolted to the side. With human experts at the touchpoints that matter. That belief is why the last year of internal experiments mattered. We were learning how to build that way ourselves before we built it for anybody else.
The shift is uncomfortable for engineers who built their identity around a craft. That is fair. The craft was real. But the model can do most of it now — the syntax, the boilerplate, the search-replace, the "translate this from one stack to another." What it cannot do is decide what is worth doing. That is what to build an identity around now.