Aug 30, 2026 · 1h 21m · lennys-podcast
AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
OpenAI Product Lead Tara Seshan discusses the transition into AI's third era—collaborating with persistent AI coworkers—and explains how product teams must discard static strategy documents in favor of empirical experimentation and high-velocity iteration. She shares inside perspectives on OpenAI's decentralized culture, the essential role of human taste and cognitive discipline, and the infrastructural shifts shaping future AI tools.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Lenny holds 28.8% of the talking time here. How this is scored →
Lenny as informed peer 4.1 Guest teaching 4.7 Guest disagreement 1.3 Lenny pushing back 1.1
speaking balance: gold is Lenny, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 53:15 Rejecting blanket AI writing adoption
Tara firmly dismisses broad-brush attitudes toward AI writing, insisting that delegating writing-as-thinking to LLMs corrupts genuine intellect.
Hardest push from Lenny ▶ 29:25 Challenging confusing ChatGPT mode togglesLenny openly confronts Tara on the confusing user experience of having separate Chat, Codex, and Work mode options inside the product.
Biggest teaching moment ▶ 1:03:00 Product marketing fit precedes product market fitTara educates Lenny on Sutter Hill's counter-intuitive incubation strategy where crafting the sales narrative and positioning comes before building the software.
Lenny holds their own ▶ 1:10:20 Lenny sharing deep dive analysis of The Power BrokerLenny demonstrates his literary expertise by breaking down the multi-part analytical structure and companion podcasts for Robert Caro's masterwork.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Lenny as informed peer | Guest teaching | Guest disagreement | Lenny pushing back | Why |
|---|---|---|---|---|---|---|
| Inside OpenAI: Experiencing a 'Founders-Led' Culture | 3 | 4 | 1 | 1 | Lenny opens by inquiring about the realities of working inside OpenAI versus external perceptions. Tara educates the audience on OpenAI's 'founders-led' decentralized culture and the absence of secret master strategies. | |
| Sponsor Message: WorkOS | 4 | 5 | 2 | 1 | After the WorkOS sponsor read, Lenny asks what PMs lose in this new fast-moving environment. Tara reframes PM strategy from academic PhD-style thesis writing to empirical, fast hypothesis testing. | |
| Steering vs. Rowing: Human Taste and Intuition in Agentic Workflows | 5 | 5 | 1 | 2 | Lenny probes how loops expand to general knowledge work. Tara introduces the steering versus rowing mental model, arguing that human taste, determinism, and auteur-like vision remain the central differentiator. | |
| The Next Paradigm: Persistent Multi-Agent Coworkers and Cloud Infrastructure | 4 | 6 | 2 | 2 | Lenny suggests the industry might be in a slow takeoff scenario. Tara politely pushes back, pointing out that unlocking persistent agents is less about raw model intelligence and more about mundane cloud infrastructure and data access. | |
| Elevating Ambition in the Age of AI Capabilities | 4 | 4 | 1 | 1 | Lenny and Tara explore the necessity of elevating ambition when AI lowers the barrier to multidisciplinary work. Tara cites Patrick Collison and Tyler Cowen on pushing the boundary of what can be accomplished rapidly. | |
| OpenAI Internal Memes and Building for Two to Three Months Out | 5 | 5 | 2 | 1 | Lenny references internal cultural memes like 'maximally accelerated' and quotes Kevin Weil. Tara explains the counter-intuitive rule that building for today or building for one year out both fail, requiring teams to build strictly for two to three months out. | |
| Demystifying ChatGPT Architecture: Chat, Codex, and Work Modes | 4 | 6 | 1 | 2 | Lenny pulls up the ChatGPT interface and asks Tara to clarify the confusing distinctions between Chat, Codex, and Work modes. Tara explains the underlying unified Codex engine and OpenAI's north star of seamless automatic harness selection. | |
| Shipping Fast vs. Perfection: Bringing Agents to a Billion Users | 4 | 5 | 1 | 1 | Lenny asks about balancing radical speed with a billion active users. Tara explains how conviction in transformative capabilities prioritizes rapid deployment over legacy pixel-perfect polish. | |
| Sponsor Message: Mercury | 4 | 4 | 1 | 1 | Following the Mercury sponsorship read, Lenny inquires about the observable vibe shift from Claude Code to Codex on tech Twitter. Tara attributes it to continuous internal dogfooding rather than sudden marketing changes. | |
| Role Fluidity and the Evolution of Engineering and PM Craft | 4 | 4 | 1 | 1 | Lenny brings up the blurring lines across PM, design, and engineering roles. Tara discusses how startup accountability transcends rigid boundaries while acknowledging nostalgia for traditional craft. | |
| The Enduring Value of the Human Brain in the AI Era | 4 | 4 | 1 | 1 | Lenny asks where human intellect remains indispensable. Tara outlines three pillars: ultimate accountability, expressive auteur craftsmanship, and interpersonal care and enthusiasm. | |
| Actionable AI Workflows: Generating Instant Sites and Data Visualizations | 3 | 6 | 1 | 1 | Tara teaches Lenny about underutilized features in Codex, demonstrating instant dynamic 'Sites' and the '/visualize' command as replacements for static slide decks. | |
| Writing as Thinking vs. Writing as Reporting | 4 | 5 | 2 | 1 | Lenny asks about Tara's renowned brief-writing process. Tara firmly distinguishes between writing as reporting (which should be fully automated) and writing as thinking (which must remain strictly human). | |
| Protecting Cognitive Sharpness Against AI Brain Rot | 4 | 4 | 1 | 1 | Lenny asks how to protect against cognitive decline and AI over-reliance. Tara details her personal rule: start human, end human, and invest prep time equal to the audience's reading time. | |
| Lessons from Sutter Hill Ventures: Product Marketing Fit Precedes Product Market Fit | 4 | 6 | 2 | 1 | Lenny asks about Tara's time at Sutter Hill Ventures. Tara shares the contrarian insight that Product Marketing Fit (narrative and positioning) should precede Product Market Fit and code development. | |
| Live Product Demonstration: Generating a Website for Tara | 5 | 5 | 1 | 1 | Lenny shares the live site generated by Codex during their interview. Tara contrasts coding agents (evaluated by test suites) with knowledge work agents (which require transparent reasoning and citations). | |
| Lightning Round: Essential Literature and Reading Rituals | 5 | 4 | 1 | 1 | Lenny kicks off the lightning round with book recommendations. Tara suggests Barbarian Days and Anna Karenina, while Lenny shares his experience reading The Power Broker. | |
| Lightning Round: Cinematic Influences and Creativity Under Constraints | 4 | 4 | 1 | 1 | Tara shares her reflections on Christopher Nolan's film and Kurosawa's Rashomon, drawing parallels between artistic constraints and raising personal ambition with modern digital tools. | |
| Lightning Round: Cozy Software and Favorite AI Experiments | 4 | 4 | 1 | 1 | Tara highlights the 'cozy software' movement and personal tools built by friends, while quoting Toni Morrison's four rules of work. | |
| Lightning Round: The Thiel Fellowship and Fostering Early Ambition | 5 | 4 | 1 | 1 | Lenny and Tara discuss her experience in the Thiel Fellowship cohort alongside Ari Weinstein and Dylan Field, framing it as an early inflection point in elevating personal ambition. | |
| Plugs and Final Thoughts: Embracing Mobile Cloud Workflows | 4 | 5 | 2 | 1 | Lenny assumes persistent cloud execution is a mobile-only feature. Tara corrects him, emphasizing that cloud-based agent execution operates across desktop, web, and mobile. |
Statements from this episode (33)
Seshan: OpenAI operates with extremely limited top-down management
“OpenAI is actually founders-led which is that everyone inside the company, especially in their area, is in essence kind of a founder to some extent. The level of like top-down direction at OpenAI is extremely limited relative to places I've worked for prior.”
Seshan: OpenAI ships internal developments to users faster than peers
“So much of what OpenAI does immediately becomes something that users can touch and feel in the product. And that cycle is faster than anywhere else I've seen.”
Seshan: Empirical experimentation matters far more than theoretical strategy in AI
“And so actually being prolific and being more like empirical is way more important than being like maybe more academic or theoretical.”
Seshan: Defining core hypotheses matters more than grand strategy in AI
“And what that means is the thinking you need to do is being as pointed as possible about what your core hypothesis is. And that hypothesis definition is the most important thing. Like, what is actually to use the Shashir Raholtra phrase, like, the eigenquestio…”
Seshan: Future of work shifts to humans steering while AI rows
“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 rig…”
Seshan: Software resembles filmmaking, requiring auteur taste over mere capital
“Software is not like real estate. You don't like put money in and get value out. It is a little bit more like filmmaking, where you can put a lot of money into a film, but that doesn't guarantee that the film is successful or good. There is some like auteur st…”
Seshan: AI agents will evolve into persistent coworkers needing periodic syncs
“People are increasingly thinking about agents that are persistent, that feel like teammates, that feel like coworkers. Where you can work with them the way I might work with someone on my team, which is they do a whole bunch of work. I provide input and then t…”
Seshan: Cloud agent effectiveness depends on infrastructure alongside intelligence
“A huge part of making these agents useful and achieving some of these futures are on the intelligence side, certainly, but a lot of it is also just really tactical, like data access, like cloud infrastructure and reliability pieces that feel yeah, they feel mu…”
Seshan: Top AI users expand capability scope rather than automate tasks
“I think the people that we see who are most effective at using AI tools don't simply use it to automate rote tasks, but use it to expand the set of things that they are capable of doing.”
Seshan: Core PM responsibility in AI era is elevating team ambition
“I think elevating others ambitions or reminding them of what's possible here is a huge part of the product management role. Like when folks say, hey, I think we can get this done in this way, or we can get this done by this timeline, or maybe this is the first…”
Seshan: AI products fail unless built for models 2-3 months out
“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, and I'm sure many people have talked about this, but both outcomes are really equally wrong. If y…”
Seshan: OpenAI's goal is eliminating manual model toggles in ChatGPT
“Our north star here is that users do not need to make decisions between picking between all these different options. Ideally, there is no toggle here that you go to the box, you type in your tasks. Like I would like to build a really awesome app that I don't k…”
Seshan: ChatGPT Work mode runs on Codex under the hood
“In work mode, that's where under the covers, this is codex. We've removed some of like the coding UI, like you don't, you're not going to see a work tree pop up all of a sudden in, in work mode, but it is the same power to get things done”
Seshan: ChatGPT Work mode brings agents to one billion users
“One of the goals of launching work in ChatGPT web and launching it in the desktop app and bringing these things together was to look at those billion people who are using ChatGPT and bring them more and more of the agent's power.”
Seshan: AI products have entered a second era defined by agents
“The second era of these products is clearly working with agents and, you know, primarily has been coding agents. We'd like to bring it to more, more domains, certainly like knowledge work.”
Seshan: In the AI era, shipping urgently beats waiting for polish
“At previous companies, like, polish was king. Getting every UI interaction or getting every little thing completely right was way more important than shipping something early, because time didn't make as much of a difference in terms of the outcome. And so, as…”
Seshan: Codex development process remains unchanged despite shifting market perception
“And actually with the Codex app the team who initially got it up and running and we're working on it we're super, again, user focused, tight iteration loop, really dog food at the thing, like mainline to the app as much as possible to get get everything right …”
Seshan: Stripe maintained no role boundaries between engineers, PMs, and designers
“Actually like Stripe was very, very much this way where there are no boundaries around what a engineer could do versus a product manager could do versus a designer could do. Everyone could do anything.”
Seshan: AI models abstract specialized tasks, forcing craft to become broader
“Some pieces of our craft are actually getting abstracted by models being able to do it really effectively, maybe better than individuals can. And your craft moves from being able to do that very specific task you did in the past to Now applying it to some othe…”
Seshan: AI-generated sites realize Alan Kay's vision of malleable software
“Sites kind of realized the dream of like malleable personal software that Alan Kay, you know, flagged in the sixties of like the true personal computer is one that has personal software. In some ways sites are like the tangible way to make that possible.”
Seshan: Prompt-built web apps replace docs and sheets in PM workflows
“The easy reach of building a site all the time has changed what my day-to-day looks like, which often in previous worlds used to look like creating lots of artifacts, like dachshunds, cheats, and whatever it might be. Now I just make sites all the time.”
Seshan: Automate writing for reporting with AI, never writing for thinking
“I think you should use the models as much as possible for writing as reporting, and in as much as you think with writing, as I really do, and I think a lot of people do, you should not use it. You know, you shouldn't replace your thinking with it.”
Seshan: Long docs no longer signal rigor; prototypes and mocks win
“A long dock is not a signal that you thought through something because you can easily produce a long dock That indicates that you haven't, and so actually the point, maybe one of the biggest changes I've experienced personally in my day-to-day, which has been …”
Seshan: Great collaborators prefer rough drafts over fully polished documents
“Very few great people want to interact with like a perfectly polished, finished idea, like a perfectly polished idea. Their new ideas just like bounce off of it versus something that has more crags and more rough edges that they too can polish with you togethe…”
Seshan: Meeting prep time should match total collective attendee time
“I have this like personal belief that if I'm going to make someone read my document, I have to at least read it first that number of times. Or I think about this in meetings, too, that if I'm gonna call a meeting with a set of people, I need to have prepped th…”
Seshan: Starting and finishing writing yourself prevents cognitive deterioration from AI
“Start yourself and yourself with a piece of writing and that doesn't deteriorate your thinking.”
Seshan: Sutter Hill uses internal recruiting tool Reticle to map networks
“They have a secret tool called reticle where they have a map of, you know, everyone in that they've interacted with and the 10 best people that those people have interacted with that helps them be so, so effective at this.”
Seshan: Product-marketing fit should precede product-market fit and building
“The idea that the way you talk about the product and the way you market it can proceed actually even building the product. It should probably come from some sort of bringing together of understanding the technology deeply and then understanding, like, the ente…”
Seshan: Coding AI allows output verification; knowledge work requires process auditing
“Coding is so output oriented that when you ask it to do a coding task, you can verify whether it did the task correctly or well via tests. You can try it out and see if it works. Like there is a way to validate it based on the output. But knowledge work is dif…”
Seshan: Knowledge work AI requires showing intermediate citations and in-progress work
“A lot of work that we have done and have to continue to do is continue to adapt the product to knowledge work, which means way more focus on Making ChatGPT your collaborator, allowing you to see all the in-progress work, see its citations and inputs, help you …”
Seshan: The future of software is bespoke tools for friend groups
“It is exactly what I think the future should be, which is people should make software that exactly meets their and their friends needs.”
Seshan: Ari Weinstein leads computer use initiatives at OpenAI
“Ari leads a lot of our computer use stuff at OpenAI and he's like shipped a whole bunch of great Things for computer use”
Seshan: ChatGPT Work executes tasks in the cloud while offline
“Even more transformatively, go try it on mobile, then take like a long subway ride or something like that, or a muni ride. And when you pop out after having no service, the thing is done for you. That's the part that feels super duper magical. You're not like …”