
Planning When Models Keep Changing

Why OpenAI is merging Codex and ChatGPT and the future of knowledge work | Andrew Ambrosino
Topic 9 of 17
Published Jun 28, 2026 · Clip 31:39–34:03 (2:24)
Andrew says they do nothing revolutionary: the shorter term something is, the more detail it needs, and nine-month plans must stay hazy because any precision added now is false precision. On the applied side, anything planned in November may have been true in December but isn't what happened — teams need to know what models will be able to do on what timeline. At his last company, the product process broke when models drove features, so they listed everything they might build over a year or two, prototyped all of it, shipped what was ready, and retried the rest each time models leapt, since viability depended on model smartness rather than feature shape. He's confident the February Codex app would have absolutely failed in November — the only difference was the models.