Spoiler: it's not about who codes faster.
For two years now, the conversation about AI and engineering has been circling the same question.
Will AI replace engineers? Everyone has an opinion. Most of them miss the point entirely.
The answer is probably no — not in any meaningful timeframe, anyway. What’s interesting is what the question reveals about how companies think about what engineers are actually for. Because most of them have it wrong. And the market is starting to feel the consequences of that.
Not a coder. That distinction matters more than it sounds.
An engineer is someone trained to think in systems. To take a messy, ambiguous problem and decompose it into something that can actually be solved. To anticipate where things break before they break. To make decisions under uncertainty and own the outcome when things don’t go the way the spec said they would.
The programming language was always just the tool. The thinking was always the job.
What’s been happening — quietly, for years, and faster recently — is that a generation of developers entered the profession with the tools but not quite the thinking. Stack Overflow first. Then Copilot autocompleting before they finished the thought. The habit of reaching for the answer before sitting with the problem. The muscle that makes you slow down and ask “wait, are we even solving the right thing?” — the one that takes years of hard, uncomfortable experience to develop — never really got trained.
AI didn’t cause this. But it’s making it a lot more expensive.
For most of the last decade, a great engineer was the one who wrote more, faster, with fewer bugs. Output was the job. Speed was how you got promoted.
AI writes code now. Engineers using GitHub Copilot completed tasks 55% faster in a 2025 Microsoft study. That number keeps going up. The execution layer — boilerplate, tests, scaffolding, documentation — AI handles that now, or will soon enough.1
So if that’s your definition of a valuable engineer, you have a problem. Because the thing you were optimizing for just got automated.
And yet. Most job descriptions for senior engineers in 2026 still read like they were written in 2019. Years of experience in X technology. Proficiency in Y framework. Knowledge of Z tools. All of which measure execution speed. All of which AI is absorbing.
The hiring criteria didn’t move. The job did.
Joe Bertolami led engineering at Microsoft, Google, and Snap for two decades. His read: the next generation of engineers will spend less time typing code and more time supervising AI.2
Supervising sounds passive. It isn’t.
Real supervision isn’t sitting back while the model runs. It’s thinking — hard, before the model starts — about whether the problem is correctly defined, whether the approach makes any sense, whether you’re asking the right question at all. Because AI is extremely good at answering questions. What it can’t do is tell you if you’re asking the wrong one.
AI produces plausible, confident output. It optimizes for whatever question it was given. If that question was wrong — if the problem was misdiagnosed, the requirement was vague, the constraint was never stated — AI will give you a perfect answer to the wrong problem. Fast. At scale. With no hesitation.
The engineer who catches that isn’t the one reviewing the output at the end. It’s the one who thought carefully before the first prompt. Who asked what we’re actually trying to solve before anyone started building.
And sometimes — a senior engineer looks at a system and knows it’s going to break, before running a single test. Years of watching things fail, compressed into instinct. That’s not a skill you can prompt.
Critical thinking. The oldest skill in engineering. Somehow the most underdeveloped one in 2026.
The market figured this out before most hiring managers did. What’s growing isn’t demand for faster coders. It’s demand for engineers with systems-level judgment — people who manage complexity, make architectural decisions, and take real accountability for what ships.
The engineer doing well right now isn’t the fastest at generating code. It’s the one who asks why before they ask how. Who can sit in a room with a founder or a CTO and translate — genuinely translate, not just nod along — between what the business needs and what the system can actually do. Who knows when to let AI run and when to stop because something looks right but leads nowhere.
That profile has always been rare. AI just made it the only one that compounds into real value.
Not “how many engineers do we need?” and not “which stack do they know?”
Bertolami closes his piece with something that’s easy to read past:
“The question isn’t whether we’ll need engineers. It’s whether we’ll have enough.”
Not enough coders. There’ll be plenty of those, or AI fills in. Enough engineers who think before the model runs. Who understand the business well enough to know when the output is wrong. Who can look at something technically correct and say — that’s not what we need, and here’s why.
That engineer is getting scarcer. The demand is going up. And most companies are still writing job descriptions for someone else entirely.
