DHH had agents rebuild Campfire in Elixir, Go and Rust
Ruby came out slowest, and Guillermo Rauch says the old reason to ignore that result just stopped applying.
DHH says he had agents implement and optimize the Campfire web app in Elixir, Go and Rust. Ruby was the slowest. He is upfront that he never cared about that, because the payoff was productivity and developer joy, then adds the line that makes the whole thing land. But if you're no longer reading the code?
Yes, Ruby is the slowest. I didn't care about that when the pay-off was huge in terms of productivity and developer joy. But if you're no longer reading the code?
His point is not that Rust wins. He notes Rust has its own drawbacks, slow compile times among them. The argument is narrower and harder to dodge. When agents write the code and validate the output, the part of a language choice that was about human comfort stops carrying the weight it used to.
Vercel already ran the experiment the expensive way
Guillermo Rauch replied with Vercel's own version. The company has been moving to Rust for a while, and one of the first projects migrated was Turborepo, from Go to Rust. The migration completed. Rauch says the return on investment was actually quite controversial internally.
The reasoning is worth the detail. Rust was better in their eyes for low level OS access, which matters for a build system. But the human migration costs were heavy, and Go is fast, well designed and easy to iterate on. Rauch says they were conflicted precisely because humans were writing the code, even though they believed Rust was the better choice.
That is the pivot. The thing that made the port a close call was never the destination, it was the people doing the typing.
Rauch also throws cold water on the idea that Rust is where this lands. He says it is quite unlikely to be the end state, because Rust itself was designed before agents arrived at scale.
The business read
Two founders who have shipped a lot of code are describing the same shift from opposite ends. DHH ran three rewrites of a real app as a test rather than an argument, which is only possible because agents made the rewrite cheap. Rauch has the receipts from doing it the old way and paying for it.
Neither of them is telling you to port anything. What changed is that the strongest objection to a rewrite, your team's learning curve and the months lost to it, is the specific cost agents attack first. If you have been sitting on a language decision you could not justify, the justification math is different now. Nobody in either post has published numbers on what the Campfire ports actually cost to produce, so treat this as two experienced operators reading the same trend, not as a benchmark.
What's "best for humans" is no longer necessarily "best for business