I noticed it in a code review last week. The PR was flawless — clean, typed, tested, generated in about four minutes. And I sat there for twenty minutes on one question the AI never asked itself:
Should this feature exist at all?
That gap is the whole story of where this job is going.
The thing that actually changed
For most of my career, being a good developer meant producing — turning an idea into working code, fast, correctly. Typing speed, syntax recall, knowing the framework cold. That was the moat.
The moat is gone. 29% of code is now AI-generated. A junior with a good agent out-produces a mid-level dev from three years ago. The production of code has been commoditized almost overnight.
What hasn't been commoditized — what actually got more valuable — is everything around the code:
- Knowing which of the three "working" solutions won't rot in six months
- Smelling the requirement that's going to cause a data-migration nightmare
- Deciding the feature the AI happily built shouldn't ship, because it doubles your support load
- Reading a system and knowing where it will break under load before it does
None of that is typing. All of it is judgment. And judgment is the one thing the model can't hand you, because it requires having owned the consequences of being wrong.
The trap nobody warns juniors about
Here's the uncomfortable mechanics of it. Entry-level roles are down ~25% year over year. The market stayed busy — ~1.5M openings — but it got selective: mid and senior engineers benefit, junior-friendly roles dried up.
Why? Because the tasks juniors used to cut their teeth on — the CRUD endpoint, the form, the glue code — are exactly what AI does best. The old apprenticeship ladder had its bottom rungs sawn off.
And there's a nastier second-order effect the burnout research is starting to flag: deskilling. When the AI writes it and you just skim-approve, your own situational awareness quietly erodes. You stop building the intuition that becomes judgment. You get faster and shallower at the same time, and you don't notice until the day the AI is confidently wrong and you can't tell.
The danger of 2026 isn't that AI replaces developers. It's that it makes you feel productive while quietly hollowing out the exact skills that were going to make you senior.
How I'm trying to not get hollowed out
No grand framework, just what's actually working for me:
- Read the code the AI writes like you'll be quizzed on it. Not to nitpick — to keep building the mental model. The moment you're rubber-stamping, you're deskilling.
- Do the hard part yourself sometimes. Deliberately hand-write the gnarly concurrency bug, the tricky query, the thing you'd normally delegate. That's your reps. You lose the muscle you stop using.
- Spend your saved time on judgment, not more output. AI gave you back hours. Don't refill them with more generated features. Refill them with design, with reading systems, with asking "should this exist."
- Own outcomes loudly. Put your name on decisions, not just commits. "I chose this tradeoff and here's why" is the skill that isn't going anywhere.
The developers who thrive from here won't be the ones who prompt fastest. They'll be the ones who can look at a screen full of confident, working, AI-generated code and say "no — not like this," and be right.
That's a harder skill than typing ever was. And nobody can generate it for you.
Real question for everyone shipping with AI daily: do you feel sharper or shallower than you did two years ago? I go back and forth, and I want to hear the honest version. 👇
I write about AI, building real things, and the honest ways they break. Follow me here if that's your lane. 👋
Top comments (0)