Theo answers a skeptic with 24 days of Opus output

A commenter, @barthazian, asked Theo what he had actually built, saying the only thing he saw from the account was "a lot of loud takes". Theo answered with a list covering the 24 days since Opus dropped.

Theo - t3.gg

@theo

X

Is all of this real stuff that will ship to users? Absolutely not. But a lot of it is.

The list, in his words: ts-rust, a complete rewrite of the TypeScript compiler in Rust that he says runs 2-3x faster than the official version, which he shipped earlier this month. 245 PRs into T3 Code, the AI coding app he says has over half a million users, with more than 60 of those PRs in the last 48 hours alone. 46 PRs into Lakebed, his cloud for building apps with agents. Over 1,200 commits into lust-lang, which he describes as a subset of Rust that compiles 100x faster. Plus Slopalytics, a model bench site built on data from Artificial Analysis and T3 Code, Slophouse, an internal ClickHouse database for anonymized T3 Code usage data, and fckbench and t3os, which he calls started rather than finished.

The number to argue with

Theo puts the total at "rough estimate is 22 years of work, in 21 days". That is his own estimate and he does not show how he got there, so it is the weakest claim on the page. The PR counts and the commit count are the opposite. They are public, they are countable, and anyone who wants to dispute the post can go count them.

He also pre-empts the obvious objection himself, conceding that not all of it is real stuff that will ship to users, but insisting a lot of it is. In a follow-up he said that mapping it out this way made the value he gets from his Claude subscriptions look "kind of insane".

The post ends with a jab at local models, asking who can get this done with a 27b model on their GPUs at home, which puts it squarely in the same argument he has been running about whether local AI makes economic sense.

For working engineers, the useful reading is not the year-count. It is that a list like this is now the format a skeptic gets answered in, and that the only part of it a reader can verify is the commit log.