DHH says a language agents write badly is in trouble
Rauch agrees and says the old rule of picking what humans enjoy writing no longer holds.
DHH says he had agents implement and optimize the Campfire web app in Elixir, Go and Rust. Ruby was the slowest, which he says he never minded when the payoff was productivity and developer joy. His question is what happens to that trade when you are no longer reading the code.
When people pushed back that frontier agents are not good at writing Elixir or Go, he said that argument cuts the other way. A language agents cannot write well out of the box, in his telling, has a hard future. He is careful to add the obvious caveat that not every app needs Rust levels of performance, far from it. But if agents are genuinely good at Rust, and you accept that agent acceleration means nobody reviews every generated line, then the thing that used to be a tiebreak starts looking like the main argument.
Any programming language that agents can't write well out of the box is going to have a hard time in the future.
He is not claiming Rust is free. He names slow compile times as a real drawback and says speed is not the only consideration for a codebase. The claim is narrower than the headline version going around, which is that when agents write the code and validate the output, the old scoring system needs adjusting.
Rauch brings receipts
Guillermo Rauch sided with him and added a case study. Vercel has been moving to Rust for a while, and one of the first projects was Turborepo, ported from Go to Rust. The migration completed. The return on investment, he says, was controversial internally, because the human cost of the move was substantial even though the team believed Rust was the better choice for low level OS access in a build system. Go, he notes, is very fast and beautifully designed and easy to iterate on.
What's "best for humans" is no longer necessarily "best for business".
His conclusion is the part worth sitting with. The old calculus, picking what humans enjoy and can hire for, no longer automatically matches what is best for the business. He also doubts Rust is the endpoint, since Rust itself was designed before agents arrived.
The business read. This is two people with large codebases arguing from their own migrations, not a benchmark. But if it holds, what keeps a language alive stops being its community and starts being how much of it sat in the training data.