https://www.youtube.com/watch?v=zMvBMfj4cSQ
Tara Seshan, who leads product for Codex and ChatGPT Work at OpenAI, walks through how the product manager's job has changed inside a frontier lab: what she stopped doing, the three internal memes she says drive product development, and why she will hand a status report to a model but never a brief. Lenny Rachitsky builds a website about her live during the recording, presses her on AI brain rot, and takes her through the lightning round.
👤 Guest: Tara Seshan, product lead for Codex and ChatGPT Work at OpenAI
🎙️ Host: Lenny Rachitsky, who writes Lenny's Newsletter
📰 Published: 30 August 2026
🔴 YouTube | 🟢 Spotify | 🟣 Apple Podcasts | 🔗 Show notes | ⏱️ 82 min | ✅ Time saved: 55 min
Key Takeaways
There is no secret strategy document at OpenAI, and top-down direction is unusually thin
"I came into the company expecting that there was a treasure trove of like OpenAI secret strategy"
"And actually OpenAI is open."
The two-to-three-month rule: building for today's models and building for next year's are equally wrong
"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."
Long strategy documents lost to fast tests, because this market cannot be reasoned through in advance
"being prolific and being more like empirical is way more important than being like maybe more academic or theoretical"
The job is steering, not rowing, and the steering keeps moving up a level
From pressing tab on a line of code, to directing something comprehensive, to the goal level
Ambition is the constraint now, and raising other people's ambition is the PM's job
"elevating others ambitions or reminding them of what's possible here is a huge part of the product management role"
Work mode in ChatGPT is Codex underneath, and the goal is that users never pick
"our north star here is that users do not need to make decisions between picking between all these different options"
Polish stopped being king; shipping something transformative unfinished beats holding it
"done is better than perfect and we have so much more to do"
Writing as reporting gets automated, writing as thinking never does
"But writing is thinking is something I never will automate."
She writes hundreds of docs, but for herself — the shareable artifact is now a prototype or an A/B result
Knowledge work cannot be verified the way code can, so the product has to show its work
Citations, inputs and in-progress work matter in a way that passing tests makes unnecessary in coding
Sutter Hill taught her that product marketing fit can precede the product
Pitch 100 people, refine the narrative, and only then commit to a product shape
Human brains stay valuable for accountability, authorship and each other
"who ultimately owns the outcome here?"
The Codex turnaround on social media was the market catching up, not a change in how the team worked
OpenAI Is Run Like a Company of Founders, and There Is No Secret Strategy Bible
Seshan has been at OpenAI for about a year. Asked what surprised her most, she said the familiar parts were the ones she expected — smart colleagues, high urgency, hyperscaling. What was different was the absence of instruction from above. "The level of like top-down direction at OpenAI is extremely limited relative to places I've worked for prior." Every company she had worked at before was founder-led; this one, she said, is run as though everyone in their own area is a founder.
The distance between her and the market is "very, very thin" — the insulation from users she associates with a larger company is not there
The bigger surprise was the opposite of what she went in expecting: "I came into the company expecting that there was a treasure trove of like OpenAI secret strategy" — the equivalent of the payments bible she had at previous companies "And actually OpenAI is open." Almost every internal view about how the world should look, how product should be built or how the model should behave turns up quickly in the public product or the public messaging Rachitsky asked whether that meant there was no secret room with the AGI in it holding the master plan. "or at least I'm not in that room for sure", she said
What she finds inspiring is the speed of the cycle: what OpenAI does becomes something users can touch faster than anywhere else she has seen
In This Market the Grand Strategy Document Loses to the Fastest Test
Asked what a product manager loses in this new world, Seshan named the analytical set piece. In a slower or more established market — payments, her example — a team can reason from first principles through what a competitor will do next, and the market rewards whoever thinks most rigorously. Careless decisions there are ones that could have been predicted.
This market does not work that way. It is emergent, fast-changing and tied to research, so "being prolific and being more like empirical is way more important than being like maybe more academic or theoretical".
The switch felt jarring at first. Instead of a long reasoning doc "almost like a PhD thesis", the question became how to get to something testable with users as fast as possible She said she wondered whether she was failing to do her due diligence, and concluded that the thinking still has to happen — it just has to be pointed
What the thinking is for now: defining the hypothesis. She used the Shishir Mehrotra phrase, the eigenquestion — "What is like that specific most important thing to test and everything else like any other grand strategy you concoct is not relevant."
The Core of the Job Never Changed, and Now Everyone Else Does It Too
The trappings of product management — running execution to a date, the specs, the presentations — were never the job, Seshan said. The job is the essential question about the product: what determines whether it works, how to test that, how to read the result, and how to feed it back into a sharper hypothesis.
That means understanding users, the market and the technology, then pulling the three together into the sharpest hypothesis available
What has changed is that the whole company now works this way: engineering managers, engineers, data scientists and designers have all moved to the same problem-definition and testing loop
She counts that as good news for product managers, because it is the part of the job they were always focused on, and many of the other trappings have fallen away
The Future of Work Is Steering, Not Rowing
On the question of loops expanding from software engineering into all knowledge work, Seshan said "the future of work will look more like steering than rowing", with agents doing the rowing.
The steering keeps moving up a level: from writing a line of code and pressing tab, to directing something more comprehensive, to the goal level, and probably higher still
What stays with the person is the direction — "where do I want to take this thing next" — and that is not only a reading of the data
The part she thinks is underrated is intuition and the wish to see a particular future: picturing how the product should look, she said, "not because the converse is not an equally viable strategy, but because I would like the world to look like the direction that I would I'm pushing it in"
Running agents in larger and larger loops is great, "but right now like you really still need to steer"
She also expects steering to become a group activity — several people steering a shared set of agents rather than each person working alone with their own
Software Is Less Like Real Estate Than Like Filmmaking
Rachitsky put it to her that if everyone has the same tools, the only remaining advantage is the human. Seshan reached for clothes: functional clothing does the job for everyone, but what a person wears is a statement that works by contrast with what everyone else is doing.
She said products feel similarly opinionated, and cited a line she attributed to Patrick Collison, or possibly to John Collison — "software is not like real estate. You don't like put money in and get value out. It is a little bit more like film making where you can put a lot of money into a film but that doesn't guarantee that the film is successful or good."
That artistry, she said, relies on the team having something interesting to say about the product
Rachitsky brought in Marty Cagan's point that the first idea for a product is rarely the thing that ships, and that the process of finding out is the work Seshan agreed, and said the loops are moving faster and faster, so what matters is how quickly a person can form the intuition, get the information behind it, and put it into action with people and agents
The Next Shift Is a Persistent Coworker You Share With Your Colleagues
Asked what the next three to six months of knowledge work look like, Seshan named two changes. The first is agents working at higher levels of abstraction — "let the agent cook" — moving toward agents that are persistent and feel like teammates: "they do a whole bunch of work, I provide input, and then they do work again".
The second is that agent work is still solitary. Her work with her agent is separate from what her colleagues are doing with theirs.
The internal workaround shows the gap: for a while everyone was sending each other screenshots of their Codex threads on Slack to explain how they got to a number "But that's also not quite the most natural way for someone to collaborate together." How to make agent work collaborative is one of the things the team is working on
Rachitsky pictured it as a chat: his agent checking Seshan's agent's analysis against how he thinks about the world
"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."
What Makes Agents Useful Is Prosaic: Data Access, Cloud Infrastructure and Reliability
Rachitsky observed that the trust has grown gradually, and that the world looks more like a slow takeoff than the fast one people feared — not a 300 IQ AI arriving at once.
Seshan agreed the models are extremely smart, but said the things that unlocked working with agents together are mundane.
Local agents are convenient because they can reach everything on the machine. A cloud agent needs a large amount of infrastructure built before it can do the same work
The colleague analogy she used: "just like a colleague who you hire who you like lock into a room never give them access to like Google Docs and Slack and I don't know the company database would not be that useful to you"
Intelligence matters, and so does how long a model can stay on a long-running task, but data access, cloud infrastructure and reliability matter "in some ways just as much for end effectiveness"
Ambition Is the Constraint Now, and Raising Other People's Is the PM's Job
Rachitsky said the tools have made ambition compulsory rather than optional: "The easy stuff is super easy. The hard stuff is easy." What separates people and companies now, he said, is how ambitious they can be.
Seshan said the most effective users of AI tools do not automate rote tasks with them — they expand the set of things they are capable of doing at all. The old unicorn hire was the product thinker who could also engineer and maybe design, because that person collapsed the translation layers between functions.
"We kind of all have that superpower now" — she can spin up designs, build a prototype, work out a pricing model and model the scenarios herself
On what actually limits people: "Your ambitions are no longer limited by like what you're capable of executing yourself, what you're capable of communicating." The hard part is expanding the thinking
She pointed to Patrick Collison's list of unreasonably fast projects at patrickcollison.com/fast, and made the argument that the list is now dated: every one of those projects was executed before these tools existed, so the number of them should be rising exponentially
The Tyler Cowen line she keeps coming back to is that people underrate the impact of asking someone what the more ambitious version of their plan is, or whether it could be done faster, or at ten times the scale "elevating others ambitions or reminding them of what's possible here is a huge part of the product management role" — when someone proposes a first version or a timeline, the job is to ask whether the ceiling is meaningfully higher
Three Memes: Maximally Accelerated, Mainlining It, and Feeling the AGI
Rachitsky recalled that Nick Turley, who he thinks had a version of Seshan's role before her, had an internal meme: is this maximally accelerated? Seshan confirmed it, said there is an emoji for it in Slack, and added the second one she and her engineering manager ask the team — "are you mainlining it yet which is like are you using this product all day every day to get your thing done".
The third is cultural rather than operational: feeling the AGI, or staying conscious of AGI coming, and believing in the mission of making it beneficial
The three together are scope, speed and use: are we being as ambitious as possible, are we moving as fast as possible, and are you using the thing yourself and bringing your taste to bear on whether it works
Rachitsky called mainlining the successor to dogfooding, and said the team's obsession with the product is visible in how they post about it publicly
The Two-to-Three-Month Rule
The refrain Seshan says she keeps in the back of her mind while building is whether the product is right for where the models will be in two to three months.
"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." Both, she said, are equally wrong
Building around a past model's capabilities is entirely wrong; being too early is also wrong. "The only way to build is two to three months"
What that requires of the product: treating model capability as the center of the product and getting out of the model's way in the constructs built around it
How she knows what two or three months looks like: communicating tightly with research. The roadmap is not a black box — there are focused efforts to make the model better at particular things, coding or writing, and product development has to track the research agenda
Rachitsky recalled Kevin Weil, then chief product officer, saying on the show that "this is the worst the models will ever be", and said it is almost a cliché now but still absurd. Seshan agreed: "it's absurd. It's absolutely absurd."
The North Star Is a ChatGPT Where the User Never Picks
Rachitsky opened the app and asked her to explain the dropdown between ChatGPT and Codex and the chat-and-work toggle inside it. Seshan said the toggles are a waypoint, not the destination: "our north star here is that users do not need to make decisions between picking between all these different options" — a user should type the task and the product should pick the harness and the model.
Choosing between ChatGPT and Codex is really a choice of interface: a development-oriented UI, or the same power inside ChatGPT. "if you're a Codex user, keep using Codex. You're not missing out on anything"
Chat mode is for conversations and search. "But in work mode that's where under the covers this is Codex." The coding interface is stripped out — no work tree appears — but the capability is the same
The example she gave of work mode in use is a complex financial model, and she said the corporate finance team uses it for things that were previously manual or needed one specialist
Asked whether work mode is anything more than Codex with a different interface, she said the difference is at the UI level only: how much of the chain of thought and technical detail is exposed
The separation exists to meet people where they are, she said, because working with agents has transformed how every developer works and the same should happen for knowledge work
Bringing Agents to a Billion ChatGPT Users, and Why Done Beat Polished
Rachitsky asked how she balances a product with a billion users against a fast-moving new one. Seshan said launching work in ChatGPT on the web and in the desktop app was aimed exactly at those users: "if you think about the first era of AI products as chat, the second era of these products is clearly working with agents and primarily has been coding agents", and the third era, she said, might be a persistent coworker that gets things done with you, possibly alongside other people.
The product problem is not distribution but naturalness: how to make adopting agents easy and not an explicit decision, and how to "decomplexify" concepts like harnesses that make no sense to a billion consumers
The habit she had to unlearn: at previous companies polish was king, and shipping before every interaction was right was not worth doing, because time made little difference to the outcome Here, conviction that something is transformative beats finishing it. "done is better than perfect and we have so much more to do"
Rachitsky noted the confusion when the app first launched and how fast the team iterated in response. Seshan said feedback before launch is better where you can get it, and that "iterating as quickly as possible and listening to the right signals" is the point either way
The Codex Turnaround Was the Market Catching Up, Not a Change in How the Team Worked
Rachitsky described a vibe shift on social media from Claude Code toward Codex over the past few months, and asked what changed internally. Seshan reached for the Zen line about chopping wood and carrying water before and after enlightenment, and said the honest answer is that nothing changed.
The team that got the Codex app running was user-focused, iterated tightly and used the app constantly. Outside attention arrived later
"it was merely the— and to some extent like the market catching up — that was the change"
The operating mode has not moved: everyone building it uses it, fixes their own problems, and listens to problems from elsewhere in the company and from users
Rachitsky said what stood out in the answer was that it was human — a team obsessed with the customer and the product, not AI, made the difference
"the desktop team especially like asks like founders and cares about every piece and every detail", she said, and when they see something that should be better they build it, test it internally, and ship it only if people find it useful
Role Boundaries Are Dissolving, and the Craft Question Has No Answer Yet
Asked about roles overlapping, and people asking what a designer or a marketer is now responsible for, Seshan said the thing she has always liked about startups is that "everything and nothing is your responsibility, ultimately you're accountable for success". Stripe was like that too — no boundaries between what an engineer, a product manager or a designer could do.
What she does insist on is a single owner for whether the product is used, wanted, high quality and effective. Whoever that is, someone is the DRI, and the rest of the work gets picked up by affinity and capability
The counterweight is craft, and she said she does not have an answer. She loves the craft of product management, and named Shreyas Doshi, Marty Cagan and Shishir Mehrotra as people who have advanced it Some of that craft is being abstracted away by models doing it well, and the craft moves to another part of the discipline "how do I balance my desire to be part of a team and use these tools and feel so compelled by how effective one can be now with all these products"
Rachitsky said the engineering role is unrecognizable — writing code all day was the job and no longer is
"people mourn like the flow state of writing code", she said, calling it a tough transition
Where Human Brains Stay Valuable: Accountability, Authorship and Each Other
Asked where human brains remain most valuable over the next couple of years, Seshan gave three answers.
Accountability. "who ultimately owns the outcome here?" She said an agent can be thought of as your report, and the person still owns whether the end product was high quality and did what it was meant to — particularly in regulated industries and anywhere a human interface is required
Expression. She returned to the filmmaking analogy: "the greatest films are not the ones with the biggest budgets." What a person chooses to build, and how it feels, stays a human question
Each other. The part of her work that involves talking to her team, getting collectively enthusiastic about an area, learning together and elevating each other's ambitions has become more important, not less
She Builds Sites Instead of Decks, and Uses Slash-Visualize for Charts
Asked how her own work has changed, Seshan said she builds sites constantly — presentational artifacts, but also anything else.
She built a site as a game for her team to play together, because sites have a database, and a site for a recent backpacking trip that tracked the elevation of the route and let everyone input their food
"But with a site, it is literally a prompt. I literally with a prompt say like build me this exact tool that I need to get this thing done and it just does it." They are shareable, can auto-update, and can pull internal data into a dashboard
She framed sites as the realization of the malleable personal software idea Alan Kay raised in the 1960s, which people have tried to approximate with configurable blocks in tools like Notion
"rather than like painstakingly laboring over some sort of like slide deck, a site is just a way more dynamic surface for presentation" — building sites has replaced the docs and sheets that used to fill her day
Sites can be public, shared with a team or private, and can be made from work mode, Codex, the web or mobile
Her second tip was slash-visualize in Codex. Asking it to visualize her ChatGPT usage pulls in everything she has done and builds a visualization of it. She said presenting charts and data in a way that carries the story has been an endless problem and this makes it simple
Rachitsky kicked off a site about her live during the recording, which prompted a short exchange about her name — "Tara Sash like station s" — and produced a site she liked, apart from the photograph at the top, which she said is her Stripe badge photo from when she was about 19
Writing as Thinking Never Gets Automated; Writing as Reporting Always Does
Prompted by a question Brie Wolfson sent in about her briefs, Seshan split her writing in two. Writing as reporting is a status summary or a launch plan; writing as thinking is a brief arguing for a product, a strategy or a spicy take.
Reporting she automates as far as it will go. "But writing is thinking is something I never will automate." Outlining, turning it into prose, cutting and editing it is how she gets her ideas in line
She said both absolute positions are wrong — never using models for writing, and always using them — and that the line falls between the two kinds
Her method at Stripe was to write a brief, then shop it around and have people attack the ideas and poke holes before taking it to the next person. Stripe, she said, is "incredibly oriented as a writing culture"
A manager's rule she still uses: "write a doc to 70% completion and then take it to the people that you need buy-in from and get it from 70% to 100%" Ideas bounce off a perfectly polished document; something with rough edges invites people to polish it with you
What changed at OpenAI is the artifact, not the writing. A long document is no longer proof that anyone thought hard, because a long document is easy to produce "I am way more on mocks not docs or prototypes not docs" — better still is a result from an A/B test "I still write hundreds of docs all the time, but I do it for me. And I no longer do it for other people really." She called that the biggest personal change of this era
On AI brain rot, she said the writing-as-thinking discipline is her main defense, along with two personal rules: "if I'm going to make someone read my document I have to at least read it first that number of times", and for a meeting, "I need to have prepped the collective amount of time that people are going to spend in that meeting before the meeting"
She does not use models to polish her prose or write a first draft. "I start myself and I end myself" — a model may come in the middle to research, pull data or push back on ideas
What Sutter Hill Taught Her: Product Marketing Fit Comes First
Between Watershed and OpenAI, Seshan spent time as an entrepreneur in residence at Sutter Hill Ventures, the firm behind Snowflake among others. "Sutter Hill is an iconic firm and is intentionally a very illegible firm." Its website shows nothing; it operates deliberately quietly and is behind some of Silicon Valley's most iconic outcomes.
She went because her career has been a series of attempts to find product market fit — as a founder, in new products at Stripe, at Watershed — and Sutter Hill has worked out how to find it repeatedly in business-to-business software
Her central claim about the firm: people treat finding product market fit and building a very large company as a dark art, and the partner who started the incubation model has done it repeatedly, so "there's like clearly a way to do it there's clearly a roadmap for making that possible" The playbook covers how the enterprise sales team is set up, how the product is positioned and how the founding team is built She said their recruiting is unparalleled, helped by an internal tool called Reticle that maps everyone the firm has interacted with and the ten best people each of those people has interacted with
The takeaway she named: "product market fit is sure important but actually I really underrated product marketing fit" How a product is talked about and marketed can precede building it, and should come from understanding both the technology and the enterprise sales process The sequence she described is to pitch 100 people, refine the pitch, get the narrative of why the thing is transformative right, and only then commit to the product shape She said she had always dismissed product marketing as glue between functions, and now sees it as something that can decide whether a company succeeds
Knowledge Work Cannot Be Verified the Way Code Can
The last thing Seshan raised before the lightning round was a difference the team found while building ChatGPT work: "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". Knowledge work has no such check.
She cannot look at a finished deck, see the success rate it reports and simply believe it. "I really need to think about the process and the inputs and the reasoning and how it went along the way"
What that means for the product: making ChatGPT a collaborator — showing in-progress work, citations and inputs, so the user arrives at the end output already knowing it is right
Open questions she named include how much of the citation trail should appear in the reasoning and chain of thought, and whether a thread, which suits coding, is the right surface for knowledge work at all
Rachitsky compared it to pitching an executive, where most of the work is showing how you got there, and added the other half of the problem: whether the model has the context and access it needs, down to whether it can see your email
"as we think of maybe bringing in human collaborators into your work, we also need to think about how we can make the model more of a collaborator with you as you get things done together"
Lightning Round: Two Books, and a Film She Says Is About AI
Barbarian Days by William Finnegan, about a New Yorker reporter who fell in love with surfing. What she took from it: "one can be deeply passionate and dedicated and have something be your life purpose without you being good at it"
Anna Karenina, which she has been rereading along with other classics, and which she called a book of layers — the plot at 13, the class and European history at 17, and at 30 a story about a woman and about humans She tied that to the era: the challenge ahead is transforming yourself and looking at the same thing through different lenses as you grow
Simon Haisell's Substack, where he runs slow reads of long books chapter by chapter — War and Peace, and Wolf Hall by Hilary Mantel. She said that is the only way to read something like The Power Broker Rachitsky, who is reading The Power Broker, offered the 99% Invisible book club as a companion: 13 episodes of an hour or two, a couple of chapters at a time, with guests including Pete Buttigieg and Alexandria Ocasio-Cortez, and Robert Caro himself appearing a couple of times
The Odyssey, Christopher Nolan's film, which she recommended and read as a film about AI: "It is about AI as or my hot take is that it's about AI." She said Nolan has bridged artistry and commercial success in a way no other modern director has A friend got them tickets at 10 p.m. at the Metreon earlier that week
Rashomon, Akira Kurosawa's film, which pioneered telling a story through several people's perspectives. What struck her was what he did under constraints, in black and white in the 1950s "I have a hundred times the power and tools that he had making that film in my iPhone. And like what's my excuse for not elevating my ambitions and making better stuff?"
Cozy Software: Her Favorite AI Products Are the Ones Her Friends Made
Asked for a favorite AI product outside OpenAI, Seshan named the category rather than a company. "I'm such a huge fan of like the cozy software movement where you like make software tools for like five of your friends and you guys use it together."
A friend named Sebastian built her an app that turns anything into a podcast and drops it into her Apple podcast feed. Rachitsky said that is exactly the version he wants and asked how to subscribe
The same friend built a private social network for their group called GATS, which she described as private Twitter for a small group of friends, and where she said she learns the most interesting things
"It is exactly what I think the future should be which is people should make software that exactly meets their and their friends needs."
Toni Morrison's Four Rules, and the Fellowship That Changed What She Thought Was Possible
Her life motto is a set of four points from Toni Morrison's essay "The Work You Do, the Person You Are", which she pulled up on the call and read out: "whatever the work is, do it well. Not for the boss but for yourself." Then that you make the job and it does not make you; that your real life is with your family; and "you are not the work you do, you are the person that you are." Rachitsky said it is pinned to her Twitter profile.
On the Thiel fellowship, which Rachitsky summarized as $100,000 to skip college and build something instead:
"the Thiel fellowship was an inflection point in my life I wouldn't be where I am without it" — someone came to her and told her she did not have to take the path she was on
"there are key moments where you can tell people to elevate their ambitions and they do and that changes them"
It was 20 people a year, she said, under the 20-under-20 framing, and she believes it is still running
Her year was made into a documentary for CNBC, so her pitch on stage as a 19-year-old is on YouTube
Two fellows she named: Ari Weinstein, who founded Sky, acquired by OpenAI, and now leads much of OpenAI's computer-use work; and Dylan Field, who she called both an incredible talent and a very kind person
The two also spent a moment on how the fellowship's name is said — "Teal. Teal."
Her Plug: Try Work on Mobile, Then Walk Away From It
Asked what to point listeners to, Seshan said to download the ChatGPT desktop app, toggle over to work on the web, and ask it to do something — build a site about yourself, or a visualize block of your own ChatGPT usage.
The version she cares about most is mobile. Start a task, take a long ride with no service, and "when you pop out after having no service, the thing is done for you. That's the part that feels super duper magical."
Rachitsky assumed running in the cloud was a mobile-only feature. "No, it's everywhere. It's everywhere.", she said
Seshan's bottom line is that the product manager's job has narrowed to the parts a model cannot do — defining the hypothesis, steering the agents, raising what the team thinks is possible, and doing the writing that is actually thinking — and that anyone building on top of these models should aim at what they will be able to do in two or three months, because building for today's models and for next year's fail the same way.
Products, Companies & Tools Mentioned
ChatGPT and Codex (The two products she leads. Codex is what runs underneath work mode in ChatGPT; the interface differs, the capability does not, and the stated goal is that users stop choosing between them)
OpenAI (Her employer, which she describes as run like a company of founders with unusually little top-down direction and no internal strategy bible)
ChatGPT sites (Her main personal use: presentations, team games, a backpacking-trip tracker, dashboards on internal data — built from a prompt, shareable, and auto-updating)
Slash-visualize in Codex (Her second tip: it pulls in the underlying data and builds the visualization, which she called surprisingly delightful)
Stripe (Six years, joining as one of the first five product managers; the source of her writing-culture habits and of a badge photo that turned up on the site built during the recording)
Sutter Hill Ventures and Snowflake (The deliberately illegible firm whose incubation model she went to study, and one of the companies it is known for)
Reticle (Sutter Hill's internal tool mapping everyone the firm has met and the ten best people each of them has met, which she credits for its recruiting)
Watershed (Where she led product before OpenAI)
Claude Code (The comparison point for the shift in sentiment toward Codex that the host raised)
Slack and Google Docs (Named in her argument that a cloud agent without access to a company's systems is as useful as a colleague locked in a room)
Notion (Cited as one of the tools people have used to approximate configurable personal software)
GATS (A private social network a friend built for her group of friends — her example of cozy software)
Books & Resources Mentioned
Barbarian Days: A Surfing Life – William Finnegan (Her top recommendation: a life of devotion to something you will never be excellent at)
Anna Karenina – Leo Tolstoy (A book of layers she has read at 13, 17 and 30, and read differently each time)
The Power Broker – Robert Caro (Rachitsky is reading it; both recommended taking it a couple of chapters at a time)
War and Peace – Leo Tolstoy, and Wolf Hall – Hilary Mantel (The books Simon Haisell has run slow reads of)
Simon Haisell's Substack (Chapter-by-chapter slow reads of long books, which she called the only way to read them)
The 99% Invisible book club on The Power Broker (Rachitsky's recommendation: 13 episodes with guests including Pete Buttigieg and Alexandria Ocasio-Cortez, and Robert Caro appearing)
The Odyssey, directed by Christopher Nolan (She recommended it and reads it as a film about AI and the collapse of morality)
Rashomon, directed by Akira Kurosawa (Her reference for what is possible under constraints, and the source of her line about having a hundred times his tools in an iPhone)
"The Work You Do, the Person You Are" – Toni Morrison (The essay behind her four-part life motto, which she read out on the call)
patrickcollison.com/fast (Patrick Collison's list of unreasonably fast projects, which she argues should now be growing exponentially)
Tyler Cowen's website (Source of the line about asking people for the more ambitious version of what they are doing)
The Thiel Fellowship ($100,000 to skip college; she called it the inflection point of her life, and her cohort was filmed for a CNBC documentary)
Get the latest market chatter as it happens: