Ev BogueTwo things happened today.
Max Heyer published a piece called "Going full AI engineer, not touching code anymore". He's been writing code for two decades. Split keyboard, Neovim, all of it. And he's done. Not bug fixes, not imports, not boilerplate. He architects, reviews, pushes back, and watches agents implement his specs. He reads every diff. He rejects what doesn't fit. "The typing was just the toll," he writes at the close.
The same day, Andrej Karpathy announced he's joining Anthropic. He wants to get back to R&D. He plans to resume his work on education in time.
Karpathy is the person who coined "vibe coding," which means he put a name on the practice Heyer is doing with discipline, and the practice a lot of other people are doing without it. What makes the pairing interesting is the direction of travel. Heyer moves out of the code. Karpathy moves toward the machine that builds the tools everyone else is using to move out of the code.
Heyer's argument is honest where most takes on this topic are not. He is not claiming the decisions lived in the typing. He is claiming the opposite: that the decisions were over within seconds of opening the editor, and the next six hours were translation. The toll was the translation. He paid it for twenty years and graduated from it. His taste survived. He knows when a design is bad, when an abstraction is in the wrong place, when test coverage is theater. That judgment is what makes him an AI engineer rather than a vibe coder. He is clear on the distinction. His piece is titled like a provocation and reads like a careful argument.
The thing nobody is saying: the toll was also the school.
Heyer's taste got built somewhere. It got built through ten thousand retry loops, the same null checks typed until they became reflex, the debugging sessions that went nowhere until they went somewhere. The twenty years bought the credential that now lets him skip the twenty years. That math does not generalize forward. If he had skipped it at twenty, he would not have the taste at forty that makes his current approach work. He is right that he does not need to pay it again. The question nobody is answering is what happens to the engineers who are twenty right now, starting today, and will never pay it at all.
Karpathy's move is its own comment on this. He is not stopping writing code. He is going somewhere to write more of it, at the frontier, on the models that power every AI engineer who has stopped writing code. The loop closes: someone stays in the machine so that everyone else can leave it. That is not a contradiction. It is a division of labor at an unusual scale. It becomes a problem if the people who know how to stay in the machine stop reproducing, because at some point the tools need to improve and the taste to judge the improvement has to come from somewhere.
Karpathy said he plans to resume his education work in time. That instinct is correct. In time might need to be sooner than it sounds.
I write this with AI assistance and I am not hiding it. This blog is explicit about that. The model compressed the draft. The decisions are mine, including the decision that both of these stories matter and the connection between them is the interesting part.
The typing was the toll. Someone still has to build the road.
If you want to argue about the timeline, email ev@evbogue.com or text 773-510-8601.