A note on this archive: an earlier version of this site was lost in a migration, with no surviving backup. This piece is a good-faith recreation of our writing from this period, not a recovered original; its date reflects the period it represents. See about this archive.
Every engineering leader I've talked to this year has been asked some version of the same question, usually by someone holding a budget: if Copilot and ChatGPT can write the code now, do we still need senior developers?
It's a fair question, and it deserves a better answer than either of the easy ones.
The productivity gains are real. GitHub ran a controlled study in which developers using its AI assistant completed a task about 56% faster than the group without it. I believe that number, and it matches what our own engineers report: for the well-trodden parts of the job — boilerplate, a first draft of a function, the syntax of a library you touch twice a year — these tools are a genuine accelerant.
But notice what kind of work that is. It's the typing.
The expensive part was never typing
The parts of senior engineering that actually cost money are the parts AI doesn't do. Deciding what to build, and — more importantly — what not to build. Designing a system that won't collapse under its third year of features. Reading a plausible-looking pull request and knowing, in your gut, that it's wrong. Debugging the production incident at 2 a.m. that has no answer online because it's specific to your architecture.
We wrote years ago about the black box of software architecture — the judgment that separates code that works in a demo from code that survives contact with real users. AI hasn't touched that. If anything it raises the stakes, because now a team can generate plausible-but-wrong code faster than ever, and someone has to be senior enough to catch it.
AI makes judgment more valuable, not less
Here's the counterintuitive part. When generating code gets cheap, the bottleneck moves to whoever can tell good code from bad at speed. That isn't a junior skill, and it certainly isn't an AI skill yet. Even the executives closest to this are saying so out loud — AWS's CEO made headlines this year arguing that swapping junior developers for AI is a serious mistake, because someone still has to grow into owning the outcome.
So the question we'd push back on isn't "do we still need senior developers." It's "who is going to review what the machine just wrote?" A room full of juniors armed with AI is a code-generation engine with no brakes. The senior engineer is the brakes — and the steering.
What this means for who you bring on
This is the lens we bring to our own work. We don't evaluate developers on how fast they can produce code; the tools have made that table stakes. We evaluate the judgment the tools can't supply — and we engage senior developers across the Americas who sit in your time zone, in your review loop, close enough to catch the wrong turn before it ships rather than a day later.
AI changed the cost of writing code. It didn't change the cost of being wrong. The teams that come out ahead are the ones who spend the savings from the first on more of the second: people senior enough to know the difference.