AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
What this video is
⚡ a 82-minute video, readable in 60 secondsThis is a Lenny's Podcast interview with Tara Seshan, who leads product for both ChatGPT and Codex at OpenAI, on how AI products are evolving from chat to agents to a possible third era of a persistent co-worker. She argues teams must build for where models will be in 2 to 3 months, not today or a year out, and that as AI levels the playing field on raw capability, ambition and human taste become the real differentiators. The conversation covers how ChatGPT's work mode and Codex are converging into one experience, her writing-as-thinking discipline, and why she now sees product marketing fit as being as important as product market fit.
[00:11] Host frames a possible third era of AI products beyond chat and agents: a persistent co-worker that gets things done with you.
Key takeaways
- [00:23] Tara Seshan: building for where models are now, or for where you think they'll be in a year, are both equally wrong.
- [00:32] Tara Seshan: the only viable target is building for 2 to 3 months out.
- [00:38] Being prolific and empirical matters more than being academic or theoretical, favoring quick testable prototypes over long reasoning docs.
- [01:18] Tara Seshan leads product for both Codex and ChatGPT at OpenAI, working alongside engineering manager Andrew Amberino.
+ 24 more takeaways
- [02:54] Unlike her prior founder-led companies, at OpenAI everyone in their area is essentially a founder, with very limited top-down direction.
- [08:52] She references Shashir Rahul's phrase 'the itigen question,' the single most important thing to test, above any grand strategy.
- [11:22] The future of work will look more like steering than rowing, with agents doing the rowing while humans point direction.
- [13:20] If everyone has access to the same AI tools, the human becomes the differentiator, an almost unfair advantage.
- [14:27] Software is compared to filmmaking rather than real estate: putting money in doesn't guarantee value out, it requires artistry and opinionation.
- [20:27] The easy stuff is now super easy and even the hard stuff is easy, so what separates people and companies is how ambitious they can be.
- [23:03] Patrick Collison's list at patrickcollison.com/fast catalogs unreasonably ambitious projects executed in very short time periods.
- [24:30] A big part of the PM role now is elevating others' ambitions by asking whether the possibility ceiling is meaningfully higher.
- [25:31] Internal OpenAI meme: 'are you mainlining it yet?', meaning using the product all day every day to get things done.
- [26:48] 'Feeling the AGI,' staying conscious that AGI is coming, is a core cultural touchstone tied to believing AGI can be beneficial.
- [28:54] She recalls Kevin Weil's line from a prior podcast: 'this is the worst the models will ever be.'
- [31:30] The corporate finance team uses ChatGPT's work mode to generate complex financial models, turning a previously manual, single-expert task into something the whole team can execute.
- [33:01] The stated north star is to merge ChatGPT's work mode and Codex so users don't have to decide between them.
- [35:21] The product challenge is to decomplexify agent use so a billion consumers don't have to understand concepts like harnesses.
- [36:53] When you have conviction something is transformative, shipping it is way better than making it perfect.
- [46:20] Humans remain the 'entity of accountability': an AI agent is like a report, but a human ultimately owns whether the output is high quality.
- [53:50] She never automates 'writing as thinking,' believing outlining, drafting, cutting, and editing is one of the most important steps for clarifying her ideas.
- [57:30] A previous manager told her the right approach is to write a doc to 70% completion, then bring in the people whose buy-in you need to get it from 70% to 100%.
- [01:03:05] She says she underrated 'product marketing fit' before realizing it can precede building the product itself, calling it as important as product market fit.
- [01:04:18] Demo: an AI website builder generated a personal portfolio site for Tara in 11 minutes 25 seconds after a 'make it more awesome' prompt.
- [01:05:14] She says coding and knowledge work differ fundamentally: coding output can be verified via tests, while knowledge work output like a deck can't simply be checked by its final numbers.
- [01:05:55] ChatGPT product work must adapt to knowledge work by making the model more of a collaborator, showing in-progress work, citations, and inputs so users can trust the output.
- [01:20:46] A cloud feature lets users toggle work on the mobile app so the AI keeps working remotely even when not running locally, then chat with it once done.
- that they want and and you get $100,000
Shown on screen — grab and go
QUOTEToni Morrison quote card on meaningful work
1. Whatever the work is, do it well not for the boss but for yourself. 2. You make the job; it doesn't make you. 3. Your real life is with your family. 4. You are not the work you do; you are the person you are. TONI MORRISON, The Work You Do, the Person You Are (2017)
Transcribed from the on-screen quote card at 1:15:48; word-spacing restored and the year, OCR-read as '2o17', normalized to 2017. shown at 1:15:48
QUOTELenny Rachitsky testimonial on the sponsor card
Every startup that I'm an investor in that starts to expand upmarket ends up working with Workos and that's because they are the best.
Quote text read verbatim from the WorkOS testimonial card OCR at 6:06; attribution to Lenny Rachitsky is per the frame's visual description. shown at 6:06
Their links, sorted & clickable
🛠️ Tools they use1
WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more💼 Sponsorship & business1
*Recommended books:*🔗 Other links21
Mercury—Radically different banking, now with Command*Episode transcript:*• Newsletter• Newsletter• Codex• ChatGPT Work• Stripe• Watershed• Thiel Fellowship• Meet your Lenny’s Newsletter Fellows• The rituals of great teams | Shishir Mehrotra of Coda, YouTube, Microsoft• The nature of product | Marty Cagan, Silicon Valley Product Group• Product management theater | Marty Cagan (Silicon Valley Product Group)• Patrick Collison’s examples of fast projects• Inside ChatGPT: The fastest-growing product in history | Nick Turley (Head of ChatGPT at OpenAI)• Barbarian Days: A Surfing Life• Anna Karenina• The Power Broker• War and Peace• Wolf Hall_Production and marketing byWas this brief useful?