AI’s third era: the rise of persistent AI coworkers

Tara Seshan leads product for Codex and ChatGPT Work at OpenAI, after six years at Stripe and a stint at Sutter Hill. The conversation covers how PM work is shifting toward empirical, fast hypothesis testing, the rise of persistent AI coworkers, the need for human ambition and steering, and how OpenAI's ChatGPT Work and Codex products are converging.

01Key learnings

• Be prolific and empirical rather than theoretical: get to something testable with users as fast as possible instead of writing long reasoning docs.
• Sharpen the single most important hypothesis (the 'eigenquestion') and design fast tests to resolve it; that loop has always been the core PM job.
• Build for where models will be in two to three months: building for current capabilities or for distant future capabilities are both wrong.
• Expect the human role to shift toward steering agents rather than rowing; opinionated judgment and direction-setting remain uniquely human.
• Use AI to expand the set of things you can personally prototype (design, pricing models, financial scenarios), and elevate team ambitions by asking whether something could be more ambitious or faster.
• Keep writing-as-thinking human-authored, while automating writing-as-reporting; get docs to about 70% then iterate with people to reach 100%.
• Prefer prototypes and experiment results over long docs as the persuasive artifact when communicating direction.
• Product-marketing fit can be tested before building the product; refining the pitch and narrative early can de-risk what you build.

02Frameworks mentioned

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