The thing that every dev I know is quietly worrying about is not AI taking their job. It's losing the skill to do their job.
We're not talking about it out loud because it feels like admitting weakness, and because these tools are genuinely powerful. You can roll out a feature in an afternoon that previously took you a week. Autocomplete guesses what you're thinking. A chat agent lets you restructure that flaky module you've been avoiding for months. But then in that in-between moment, the third or fourth prompt, the third or fourth tab-tab-tab-accept cycle, that quiet question pops up: could I still do this from scratch?
Now don't get me wrong, I'm not saying that devs couldn't write a for loop if they wanted to. I'm saying that this part is what's fading. The deep work of wrestling with a problem, thinking it through, fumbling around in the dark until the solution crystallizes. Because instead of spending fifteen minutes thinking through the problem, the model gives you the answer in fifteen seconds, so what you're doing is no longer building something. You're reviewing it, which is a far less taxing and less creative task.
The research is starting to catch up with this concern. In a Microsoft study presented at CHI 2025, 319 knowledge workers showed a clear trend: the higher the trust in AI-generated content, the lower the level of critical thinking. The change in workflow is not about problem solving, but about checking and integrating. In another study, this time from MIT Media Lab about essay writing, people who used LLMs were the ones with the worst brain connectivity. They called it cognitive debt. Even productivity is more doubtful than the hype says: in a 2025 study by METR, experienced open source developers using AI tools turned out to be slower than those who didn't, but convinced themselves of the opposite.
All the devs I know have felt this, not as some abstract fear, but as a concrete experience of reaching for an answer and realizing the shelf is empty. I don't need a study to tell me that the knowledge I used to have is slipping away. That the syntax I once produced on autopilot is now a few keystrokes and a prompt away. That the kind of instinct only a developer who's been through dozens of bad architectural decisions can have is now being outsourced to a model that has seen a million architectures but never lived inside one.
The scary part is that it happens under the radar. You're still writing good code, because the model writes good code. You're still getting good performance reviews. You're still shipping. Until you're not, and that's when the tools don't work: when a weird bug shows up, when a novel constraint hits, when an edge case the training data never saw appears. Only then do you realize if there's actually a developer under the hood, or if it's just a workflow.
No one wants to go back to the old days of writing boilerplate the hard way. But the devs who are quietly preparing are doing it in the not-so-cool way: writing some code without AI, wrestling through the confusion, keeping those weak muscles working. The scary part is not that AI will replace you. It's that one day you'll wake up and you'll have replaced yourself, one accepted suggestion at a time.
Top comments (6)
This got me thinkingπ€. I wonder if there's a difference between reviewing AI's work and actually thinking through it.
If we just accept whatever AI gives us, then yeah, our problem-solving skills might get rusty. But sometimes reviewing means questioning a design choice, trying an edge case nobody thought of, or figuring out why the AI got something wrong.
I've been working with AI-driven testing, and one thing I've learned is that getting AI to do the work is only half the job. You still need to know how to challenge what it gives you.
Maybe the real question isn't whether we can still build everything from scratch. It's whether we can still recognize when AI gets it wrong.π
fair point, scrutinising the output is half the job. building from scratch is how you learn what "wrong" looks like in the first place....π
Haha, that's a good point! You need to have made a few mistakes yourself before you can spot them in someone else's work. Maybe that's why experience still matters, even when AI can do so much of the actual work. π
The reviewing vs building split is the part that stings. Reviewing a 90% right answer feels like work but you skip the fumbling where understanding forms. The skill I'd worry about losing fastest is debugging when the model gets stuck, because that's when you find out what's still under the hood. Curious if you think reviewing compounds into anything over time or just plateaus
Spot on about debugging, that's when you actually find out what you know.
i think reviewing compounds for spotting bugs and bad patterns, kind of like reviewing PRs all day. it just plateaus when you have to solve something from scratch. reading a solution doesn't teach you the same way getting stuck does.
I build Aladdin, a job agent. Your debugging point is the one I keep coming back to. The agent reads the actual form for each application, and when it cannot read a field it asks the user instead of guessing. That only works if someone still knows what a correct answer looks like. Reviewing does build one skill, spotting a wrong field or a made-up detail. It does not build the other one, knowing what to do when the agent is stuck. Your METR point fits that. The model is fast on the familiar part and slow where judgment lives.