
Building Ahead of Model Capability

Why OpenAI is merging Codex and ChatGPT and the future of knowledge work | Andrew Ambrosino
Topic 10 of 17
Published Jun 28, 2026 · Clip 35:18–39:21 (4:04)
Andrew confirms they build artifacts ahead of model capability, but stresses being clear about what stage of the design process something is in — written code means you have an artifact to test against future models, not something to ship. He traces a thread through Operator, Atlas, Codex, and ChatGPT: fundamentally the same feature re-released with different intelligence, totally changing outcomes. The original Codex web release was 'too AGI-pilled' for its moment — delegate a task and it comes back finished — while Claude Code's local, question-asking form factor matched where models were. You might need to release a thing six times before it works with the shape unchanged. On ambition: once a feature exists people optimize paper cuts, which is why OpenAI keeps a bottoms-up exploration culture — the Codex app disrupted ChatGPT and will itself be disrupted, and one team can't be great at both disruption and reliability, so you design a process for both.