OpenAI Codex lead on the new shape of product work

Andrew Ambrosino, product and engineering lead for the Codex desktop app at OpenAI, discusses how AI is reshaping product work: implementation has become cheap while taste and curation matter most, prototypes versus documents, the blurring of design, engineering and PM roles, planning under fast-moving models, and Codex expanding into general knowledge work.

01Key learnings

• Implementation is no longer the expensive part of building; the scarce skill is taste, meaning deciding what to build and curating among many attempts.
• Pick the medium deliberately: use a document to clarify a vague product area, and a prototype to stress-test an interaction pattern.
• Prototypes can look production-ready and anchor teams too early, so label whether something is an exploration or a shipping candidate.
• Role boundaries are blurring, but specialties with real best practices still matter; don't abandon product discipline just because everyone can code.
• Plan short-term in detail and keep long-term plans hazy, since nine-month precision is false precision when model capabilities shift every few months.
• Features may fail not because the idea is wrong but because model intelligence isn't yet good enough; the same product can succeed months later with better models.
• Use your own product daily as a dogfooding loop, and set up personal automations like daily Slack briefs to discover what should become first-class features.
• Good hires are high-agency, high-taste people who can take an idea from start to finish and steer AI agents rather than typing code character by character.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.