Vibe Coding Is Frying My Brain
A 2025 study from METR: sixteen experienced open-source developers used AI coding tools and ended up 19% slower — while believing they'd gotten 24% faster. An IEEE paper, "No Vibe Without Comprehension," put EEGs on 26 programmers reading AI-generated code. As complexity rose, cognitive load hit saturation.
My first reaction was "so what?" Not wrong, but not what I'm dealing with. My problem isn't the load of reviewing one line of code. It's that AI generates output faster than my brain can process it. My head hurts right now, writing this.
I tried to sort it out in a self-interview.
What's your current state?
Three to four projects running at once. Each one has more than one AI session open. While I was typing answers for this interview, three sessions sat waiting for my approval. I should have gone back to check them. I didn't.
Not a metaphor. Actual headache.
Do you review the code yourself?
No. Codex writes code. Claude reviews it. I skip diffs entirely — to shrink the information I touch.
Still exhausting. Code review is gone, but what remains is directing and QA. Splitting work, setting direction, checking output against intent. Reading code was one kind of load. Judgment and context-switching replaced it.
How is the fatigue different from when you used to code?
Back then, fatigue came from designing logic and structure in my head. That was the bottleneck. Thinking took time. One project was hard enough, and many days ended with nothing to show for hours of deliberation.
Now it's reversed. AI is too fast. Results pile up faster than I can review and respond. Keeping up means constant context-switching — check A's output, jump to B for direction, handle C's approval. Repeat without pause. That's what frying your brain feels like.
This interview proved the point. Each answer I typed drew three new questions within five seconds. No pause to think. When AI responds faster than you can process, the gap turns straight into stress.
Could you juggle this many projects before vibe coding?
Not even one. Vibe coding changed that. Progress moves fast, and I can reach areas I never could before.
The trouble is success breeds greed. One project rolling turns into "maybe this too." More projects, more sessions, more context. Addiction fits. This week was too much.
Have you tried using a management agent?
Tried it. Set up an agent to manage projects at a higher level. Didn't help much. I'm still the one supervising. Steering toward what I want, giving instructions, checking results — none of that delegates.
Offloading review, skipping diffs — all ways to shrink what I have to touch. The middle of the pipeline automates. The endpoints — direction and final judgment — need the human.
Would others experience this too?
I've always multitasked heavily. Ran 3–4 mobile games at once, handled questions for three projects simultaneously at work. Built up some tolerance.
If someone with that tolerance is getting fried, people less used to multitasking will hit the wall sooner. Vibe coding's barrier to entry keeps dropping. Everyone gets to taste fast results. The cognitive price of that speed doesn't get much airtime.
So what are you going to do?
Self-restraint. That's it. I've tuned the harness, delegated code review, tried management agents. Done all of it. Brain still fries. The middle of the pipeline optimizes fine. The human at both ends does not.
AI throughput scales. The cognitive capacity of the person supervising it doesn't. Whether that's a structural limit of vibe coding or incomplete adaptation — no idea yet. Interview's over. Going to rest.