Marc Lou: OpenAI models are the overthinking guy
Marc Lou says he fed his VO2 max lab results to GPT-6 Astra and got 200 sentences of hedging, while Opus 5.5 gave him the gist and the steps. DHH, meanwhile, is having Codex and Claude review each other.
Real money
Marc Lou says the gap between Opus 5.5 and OpenAI keeps widening
Marc Lou says the more he uses Opus 5.5, the further behind OpenAI's models look, and not only on coding. He says he sent GPT-6 Astra his VO2 max lab test results and got around 200 hedging sentences, while Opus 5.5 gave him the gist plus actionable steps.
OpenAI models right now are like that overthinking guy who can't make a decision.
Stolen playbooks
DHH's adversarial review: tell Codex to review it with Claude
DHH published his technique for adversarial code review: when he is driving from Codex he says "Review this with claude", and when he is in Claude he says "Review this with codex". He says the models already know how to kick off a review through the CLI, take turns, and settle an argument. His words: no magic.
Founder fights
Jack Ellis and Rauch answer the "software engineering is dead" crowd
Jack Ellis says he just spent 15 hours refining a big new feature, asking questions the frontier models had not even considered, and that engineering is not going anywhere. Guillermo Rauch made the structural version of the same point: every program can be hardened and optimized basically forever, and agents will happily drill no matter what.
Even in an infinite token world, you need to know when to stop and what tradeoffs to accept.