AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
with Tara Seshan
30 Aug 20264 min read1h 20m
TL;DR
Tara Seshan, OpenAI's product lead for Codex and ChatGPT Work, argues that the only viable planning horizon in AI product development is 2-3 months — building for now is wrong, building for a year out is equally wrong. She says the PM role is shifting from grand strategy and long reasoning docs to rapid empirical testing, and that the next frontier is 'multiplayer' work where teammates steer their own persistent AI coworkers together in real time.
Key Moments
Tara Seshan
“I came into the company expecting that there was a treasure trove of like open AI secret strategy and actually open AAI is open.”
Tara describes her biggest surprise about working at OpenAI — that strategy becomes public almost immediately.
“Rather than writing out some like long reasoning doc almost like a PhD thesis of what I think should be the plan for the next like end amount of time instead it's like how do I get to something I can try out and test with users as fast as possible.”
Tara contrasts the theoretical PM style she practiced at Stripe with the empirical approach required at OpenAI.
“I do think that increasingly the future of work will look more like steering than rowing in the sense that there will be agents that you'll you'll be able to work with that do a lot of the rowing and your role increasingly becomes steering the ship in the right direction and pointing it in the right direction.”
Tara describes how she sees knowledge work evolving as agents take on more tactical execution.
“Ideally, work feels like a multiplayer game where all of us together are getting stuff done, steering our agents as our agents continue to take care of more and more of those like rowing tactical tasks.”
Tara describes where she thinks collaborative AI-assisted work is heading — beyond one person and their agent.
“You fail if you build for where the models are now. You fail if you build for where you think the models will be in a year. Like both outcomes are equally wrong.”
Tara explains OpenAI's internal mantra about the only viable product planning horizon being 2-3 months.
Tara Seshan leads product for both Codex and ChatGPT Work at OpenAI. Before OpenAI, she spent six years at Stripe, joining as one of the first five PMs and being named one of the top three Stripes across the entire organization for many of those years. She also led product at Watershed, was a founder, a Thiel Fellow, and one of Lenny's newsletter fellows.
Takeaways
1
Build for 2-3 months, not now or next year Tara's core OpenAI mantra: building for current model capabilities makes your product stale on launch; building for one year out makes it irrelevant today. The only viable window is 2-3 months, staying tightly synced with the research roadmap to know where model capabilities are actually heading.
2
Empirical beats theoretical in fast-moving AI markets In stable markets like payments, rigorous long-form strategy docs are a competitive advantage. In AI, that approach is a liability. Tara switched from writing PhD-thesis-style planning docs at Stripe to asking: what is the single sharpest hypothesis I can test with users as fast as possible? The Shashir Mehrotra 'eigen question' framework — identifying the one most critical thing to test — becomes the primary PM deliverable.
3
The next AI work shift: multiplayer agent steering Most current AI work is one person with one agent, isolated from colleagues. Tara says the next frontier is multiplayer — teammates steering their own persistent AI coworkers collaboratively, with shared visibility into each other's agent threads. OpenAI internally was already sharing Codex thread screenshots on Slack as an early signal of this need.
4
The PM's new job is elevating team ambition Because AI tools have dramatically expanded what's executable, the ceiling of possibility has shifted upward faster than most people's mental models. Tara argues PMs must now actively push teammates to raise their ambitions — asking 'couldn't we do this 10x bigger or faster?' — rather than simply coordinating execution on pre-set goals.
5
Cloud agents need data access to be useful — boring but critical Intelligence improvements get the headlines, but Tara flags that cloud infrastructure and data access are equally limiting factors. An agent that can't connect to your Google Docs, Slack, and internal databases is like a new hire locked in a room with no tools. Solving this 'meat and potatoes' plumbing is as important as model capability improvements for real-world agent effectiveness.
6
OpenAI has almost no top-down product direction Tara expected a 'payments bible' equivalent — a secret strategic playbook — and found the opposite. OpenAI operates founders-led, meaning individuals are essentially founders of their product area with very limited top-down direction. Strategy becomes public almost immediately because it surfaces directly in the product and messaging cycle.