OpenAI Product Lead: The 50-Page Document Is Losing Its Value

Summary

In an 81-minute Lenny’s Podcast interview, Tara Seshan, identified in the article as the product lead for OpenAI Codex and ChatGPT Work, argues that AI is rapidly reducing the value of long documents as proof of thoughtful work. She describes a progression from chat, to co-pilots for specific tasks, to persistent agents that retain context and execute complex workflows. In this model, agents “row” while people steer: humans set direction, take responsibility for outcomes, judge quality, add perspective and expression, and coordinate and motivate teams. Seshan says OpenAI abandoned rigid one-year product roadmaps because products built around current model weaknesses can become obsolete after an upgrade, while products built around assumptions about models a year away amount to blind testing. Instead, teams plan around model progress expected over the next two or three months and maintain frequent communication with researchers about which capabilities will improve and when. She also says long documents are now mainly for personal thinking; prototypes, A/B-test data and user feedback should carry more weight in product decisions. Her preferred process is to bring a roughly 70%-complete proposal to decision-makers so they can help shape the remaining work. Routine reporting and formatting can be delegated to AI, but writing that establishes logic, chooses a direction or resolves disagreement should remain human because the writing process itself helps clarify thought. As AI lowers the cost of design, analysis, coding and prototyping, she identifies ambition and the ability to validate sharper hypotheses quickly as increasingly important differentiators. For knowledge work, ChatGPT Work is described as exposing sources, inputs and unfinished work so people can inspect the process rather than trust a polished final output alone.