Ex-OpenAI researcher: iterate on the plan document with multiple frontier models before writing code

doodlestein · x · 2026-09-20

A thread on getting good work out of frontier models: lateinteraction complains he's giving 'Ultra'-tier models more detailed context and feedback than he's ever given a human, yet still extracting only OK output. doodlestein (ex-OpenAI) diagnoses insufficient upfront planning: iterate many times on the plan document with input from multiple frontier models — problems are far easier to navigate in plan space before they're code.

Related event: Researcher's rant on brittle frontier models sparks resonance(5 posts)→

Original post →

More from coding & agent

coding & agent channel →