What an AI-first engineer looks like in the room
There is a moment in an interview when it becomes obvious. Here is what is actually happening in it — and how to practice it.
The clearest yes I ever gave a junior candidate came unprompted. I had not asked about AI at all. Partway through a normal problem, they started narrating how they manage context across a long task — what they keep in the window and what they deliberately drop, when they reach for one model over another, how they sometimes run two models against each other and use the disagreement to find the weak spot in their own reasoning.
They were not using the tools. They were engineering with them. The difference was visible in about ninety seconds, and it was the whole interview.
The good news is that what I was reacting to was not talent or charisma. It was a posture — a set of observable habits — and posture is learnable. Here are the three moves underneath it.
1. Treat context as a resource you manage
An AI-first engineer does not dump everything into a prompt and hope. They think about the context window the way a good engineer thinks about memory: as a finite resource with a cost. What goes in, what stays out, what gets summarized, when to start fresh. When they talk about a task, they talk about these decisions out loud — and that is what tells you they understand the tool as a system rather than a slot machine.
2. Choose models deliberately — and against each other
There is rarely one right model for a job. The tell is hearing someone reason about trade-offs: this one for speed, that one for careful reasoning, a third to check the first two. Running models adversarially — having one critique another’s output, or generating two answers and interrogating the gap — is not a trick. It is the same instinct that makes a good engineer write a test that tries to break their own code.
Seniority was never really about years. It is the visible quality of your decisions.
3. Narrate your reasoning
The first two moves are invisible if you keep them in your head. The candidates who land the yes are the ones who think out loud: here is what I tried, here is why it failed, here is what I changed and what I would check next. This is not performance. It is the single fastest way to let an interviewer see the quality of your decisions instead of guessing at it from a finished answer.
How to practice before the interview
You do not need a job to build this. Keep a short decisions log for a week: every time you use a model on real work, write one line on why you chose it and what you would have done if it had failed you. Then practice narrating a debugging session aloud, to a rubber duck or a friend, until thinking out loud stops feeling awkward. Two weeks of that changes how you come across more than another fifty applications will.
None of this is a guarantee — a search has too many moving parts for anyone to promise an outcome. But it is the difference between being sorted as a generic junior and being recognized as what the headline says does not exist anymore. The species is real. This is how you make yourself legible as one.