Thorsten Ball's team pushes to main, no code review
Thorsten Ball says his team ships with no pull requests and no code review, with agents reading production logs and checking their own work. Meanwhile Gemini 4 Argon got a perfect Skatebench-private score from Theo and the worst false claim rate he has measured.
The big disagreement
Theo says local AI only makes sense as a hobby, replies push back
Responding to a post about a 10-person startup running a $18,000 128GB-VRAM box, Theo argued the economics are bad: eight years of $200 Claude subs buys you quantized DeepSeek v4 Flash at 30tps. He says enjoying it is fine, framing it as liberation is not, and that the $200 plans will get cheaper, not vanish.
Claude max sub will not cost $200 for long.
Receipts
Thorsten Ball: no pull requests, no code review, agents push to main
Thorsten Ball described how his team ships: no pull requests, no code review, straight to main. He says their agents read production logs, watch deploys, grant feature flags and check their own work end to end, and he posted a walkthrough and a video of the setup.
No pull requests, no code review. We push to main.
Model and agent watch
Gemini 4 Argon aces Theo's bench, then fails his other one
Theo says Gemini 4 Argon is the first model ever to get a perfect score on Skatebench-private, then ran it on his codebase analysis bench and found it the only model with a greater than 50% false claim rate. dax, separately, calls it a frontier model and says it was the second highest used model by his team last week.
It’s still a Gemini model though. Does weird shit regularly and I don’t feel like I can trust it at all.