Steering, Not Rowing: Tara Seshan on Building ChatGPT Work for the Era of AI Coworkers
Lenny's Podcast
Tara Seshan leads product for ChatGPT Work at OpenAI. Host Lenny Rachitsky also introduced her as leading product for Codex. She joined OpenAI after more than six years at Stripe, where she was one of the first five product managers, and after roles at Watershed and as an entrepreneur-in-residence at Sutter Hill Ventures. In this conversation she lays out how she thinks AI products are evolving. She describes a first era of chat, a second era of agents, mostly coding agents so far, and a possible third era of persistent AI coworkers that people steer together. She also discusses what this means for product managers, why she thinks ambition has become the bottleneck, how OpenAI decides what to build ahead of model capabilities, and why she refuses to let AI do the writing she uses for thinking.
"Founders-led" rather than founder-led
Asked what surprised her most after about a year at OpenAI, Seshan said much of it felt familiar. The talent, urgency, and hyperscaling intensity resembled the fast-growing companies she had worked at before. The difference she noticed was structural. Every previous company she had worked for was founder-led. She described OpenAI as "founders-led": nearly everyone acts as a founder within their own area, and top-down direction is, in her words, extremely limited compared with her past employers. She found this energizing, partly because she came from a founding background herself. She also said the distance between an individual and the market is very thin. Unlike at some large companies, where teams can feel insulated from user demand, at OpenAI each team does whatever it takes to find product-market fit for its own product.
The second surprise went against her expectations. She had assumed there would be a "treasure trove" of secret OpenAI strategy, something like the internal "payments bible" she might find at a payments company. Instead, she said, "OpenAI is open." Nearly every idea about how the world should look, how products should be built, or how models should behave quickly becomes part of the public product or public messaging. When Lenny joked about a secret room with an AGI and a master plan, she replied that she is "at least not in that room." What inspires her, she said, is that the cycle from internal thinking to something users can touch is faster at OpenAI than anywhere she has seen.
From theoretical to empirical product work
Lenny asked what a longtime PM loses in this new environment. Seshan contrasted it with slower, more established markets. She used payments as an example of a dynamic but established market. There, she argued, you can do "grand strategy" work: reason rigorously from first principles, anticipate a competitor's likely bet, and map out next moves. In such markets, she said, rigor is mandatory. Winners out-think everyone else, and a lack of rigor looks like carelessness, because many outcomes could have been predicted.
In AI, she said, what will emerge is very hard to predict. The market is emergent and fast-changing, and above all it has to stay closely tied to research. So being prolific and empirical matters much more than being academic or theoretical. She described many of her past companies as academic, theoretical places. At OpenAI, instead of writing something like "a PhD thesis" on the plan for the coming period, the question became how to get to something testable with users as fast as possible. She admitted the switch felt jarring. At first she worried she wasn't doing her due diligence or being thoughtful enough.
The thinking hasn't disappeared, in her view. It has become more concentrated. The key step is defining the core hypothesis as sharply as possible. She cited what she attributed to Shishir Mehrotra as the "eigenquestion": the single most important thing to test. Any other grand strategy you invent, she said, is not relevant.
What hasn't changed about the PM job
Seshan argued that the core of product management has always been the same loop. You identify the question that will decide whether your product works, figure out how to test it, read the results, and feed them back into a sharper hypothesis. Building that hypothesis means combining an understanding of users, the market, and the technology. She said this loop has become the most important capability in the company, and that engineering managers, engineers, data scientists, and designers all now work this way. For PMs, she said, this is good news. Many of the job's "trappings," such as managing execution timelines and producing documents and presentations, have fallen away, and the part that remains is the part PMs have always cared most about.
From rowing to steering
Lenny raised "loops," the idea of telling an AI what success looks like and letting it iterate until it gets there. He asked whether that pattern would spread from software engineering to all knowledge work. Seshan answered with her central metaphor: work will increasingly look like steering rather than rowing. Agents will do more of the rowing, and people will point the ship in the right direction. She expects the level of steering to keep rising. It started at the level of accepting a line of code with Tab, moved to directing larger tasks, and may move to the goal level and beyond. Still, she said, a person has to decide which direction to go and where to take the work next as new feedback and data arrive.
She stressed that steering is not only about reading data. Much of it is making an opinionated call. She said we underrate intuition and what she called a kind of positive determinism about the future you want. You might choose a product direction not because the alternative is less viable but because you want the world to move that way. For now, she said, that opinion still has to come from a person. She also predicted that steering will become social: groups of people steering a shared set of agents together.
Lenny suggested that if everyone has the same tools, the human becomes the differentiator. Seshan compared it to fashion. Functional clothes do the job, but what you wear makes a statement partly by contrast with what others wear. She cited a line she attributed to Patrick Collison, or possibly John Collison: software is not like real estate, where money in reliably produces value out. It is more like filmmaking, where a big budget doesn't guarantee a good film and there is an auteur-like element of opinion and artistry. That, she said, depends on you or your team having something interesting to say. Lenny tied this to Marty Cagan's point that an initial product idea is rarely what ships. Seshan agreed and added that these discovery loops are getting faster, so the ability to form intuitions, gather what you need to form them, and act on them with people and agents is the key skill.
What agents need beyond intelligence: persistence, collaboration, and plumbing
Asked what internal ways of working might become common in the next few months, Seshan pointed to two shifts. First, people are working with agents at higher levels of abstraction. They give some steering and then "let the agent cook." Agents are starting to feel like persistent teammates. You work with them the way she works with someone on her team: they do a block of work, she gives input, they continue, and the two sync up at intervals and review each other's work in progress. She called this coworker model a much more natural interface and said OpenAI already sees a lot of it internally.
Second, most agent work so far has been one-on-one. It's her and her agent, maybe spawning some subagents, cut off from what colleagues are doing with their own agents. She described a period when people inside OpenAI would send each other screenshots of their Codex threads on Slack to explain how they had arrived at a number. That isn't a natural way to collaborate, she said. As more work runs through agents, the question becomes how people can work with agents together, and what interface makes that natural. She described the ideal as "a multiplayer game."
Lenny described the pace of progress as a "slow takeoff" of gradually growing trust. Seshan countered that the models are very smart, but said many of the gains that let agents take on larger tasks come from "meat and potatoes" work rather than intelligence alone. Local agents are convenient because they can see everything on your machine. Making an agent work in the cloud requires a large amount of infrastructure. Agents also need access to third-party systems where your data lives. She compared it to hiring a colleague and locking them in a room with no Google Docs, no Slack, and no company database: they wouldn't be very useful, and neither would an isolated cloud agent. She said data access, cloud infrastructure, and reliability feel more prosaic than questions about intelligence, but matter just as much for real-world effectiveness.
Ambition as the new bottleneck
Lenny brought up a recurring theme on his podcast: when easy things become trivial, what separates people and companies is how ambitious they are. Seshan said the people she sees using AI most effectively don't just automate rote tasks. They expand what they can do. She pointed to the old "unicorn" hire, a PM with strong product sense who could also code and design, and who was valuable because they removed the translation layers between functions. She said everyone now effectively has that superpower. She can generate designs, build a first prototype, work out a pricing model, and model scenarios herself. This lets her act more like an auteur and realize her vision at higher fidelity. Your ambitions, she said, are no longer limited by what you can personally execute or communicate.
The hardest part, she said, is expanding your own sense of what's possible in an "unreasonably short" time frame. She pointed to Patrick Collison's page at patrickcollison.com/fast, which lists historical projects completed with remarkable speed. What stands out to her now is that all of those projects happened before anyone could learn to build something almost instantly, summarize a difficult book in one question, or produce a CAD model of an idea. Her reasoning: if those projects were possible with older capabilities, we should expect an exponential increase in such work with what AI provides.
She also cited a point she attributed to Tyler Cowen: people underrate the effect of asking someone, "What's the more ambitious version of what you're doing? Could you do it faster, or at 10x the scale?" She said raising other people's ambitions has become a major part of the PM role. When a team proposes a timeline or a first version, part of the PM's job is to ask whether the ceiling is actually much higher.
Three internal memes: accelerate, mainline, feel the AGI
Lenny mentioned a meme that former guest Nick Turley had described: "Is this maximally accelerated?" Seshan confirmed it is an OpenAI meme and added one she and Andrew Ambrosino often ask their team: "Are you mainlining it yet?" That means: are you using the product all day, every day, to do your actual work, and bringing all your taste to judging whether it works? She described three questions as the memes of product development she wants to spread. Are we being as ambitious as possible about scope and scale? Are we moving as fast as possible? Are we mainlining it? Lenny called mainlining the new dogfooding.
She added a broader cultural one: "feeling the AGI." She described it as awareness that AGI is coming. She acknowledged there are many views on what that will look like, but said a shared belief in making AGI both real and beneficial for humanity is a big part of what brings people to the company.
Building for models two to three months out
The product principle she says she keeps in mind constantly is to build 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," she said, and called both outcomes "equally wrong." Build too early and you're wrong. Build around a past model's capabilities and you're entirely wrong. Her conclusion is to put model capability at the center of the product and "get out of the way of the model" in the product constructs you create.
Asked how anyone can know what the models will look like in two or three months, she said it is not a pure black box. Research has focused efforts, such as improving the model at coding or writing in specific ways, so staying in close contact with research and aligning product development with the research agenda and roadmap is essential. Lenny recalled Kevin Weil's line that today's models are the worst they will ever be. Seshan called that reality "absolutely absurd."
Chat, Work, and Codex: one product in transition
Lenny walked through the current ChatGPT app: a dropdown for ChatGPT versus Codex, and a toggle between Chat and Work. Seshan said the north star is that users shouldn't have to choose at all. You type a task into a box, such as building an app to help podcast guests prepare, and the system picks the right harness and model. Understanding the product's limitations shouldn't be the user's job.
For now, she explained, choosing between ChatGPT and Codex is about whether you want a more development-oriented UI. Codex users aren't missing anything and should keep using Codex. People who come from ChatGPT and are curious about agents should use ChatGPT mode. Within it, Chat mode is the familiar conversation-and-search experience with improving models. Work mode is Codex under the hood, with some coding UI removed. For example, a worktree won't suddenly appear. Lenny asked whether Work mode includes anything beyond Codex, such as harness changes. She said the difference is really at the UI level. If you ask Codex for a complex financial model or a six-month revenue forecast, it will do as well as Work mode. What differs is how the chain of thought is displayed and how much technical detail you see. She gave OpenAI's corporate finance team as an example of Work mode users. She said they have turned tasks that used to be manual, or that depended on one person's deep expertise, into things the whole team can do, and have raised the team's ambitions on timelines and scope.
On balancing a product used by about a billion people with fast-moving experiments, she said a goal of bringing Work to ChatGPT on the web and the desktop app was to give those users the power of agents. This is where she laid out her three eras. Chat was the first. Agents, so far mostly coding agents, are the second. A third era, which she said "might come soon," is working with a persistent coworker that gets things done with you, possibly alongside other people. She said the near-term challenge is introducing agents to a billion people who may never have used one, without making them understand concepts like harnesses.
She contrasted this with her earlier career. At previous companies, "polish was king." Getting every interaction right mattered more than shipping early, because timing didn't change outcomes much, and if every corner wasn't polished you might as well not ship. In this era, she said, when you are convinced something is transformative, getting it into users' hands beats perfect. She acknowledged the team has a lot to do to make Work easier for chat users, including people who use ChatGPT for consumer tasks rather than productivity. When Lenny noted the early confusion around the launch, she added that it's always possible to iterate more before launch and that there is plenty they can do better. The key, she said, is iterating quickly and listening to the right signals, ideally before launch.
Why Codex's reputation shifted
Lenny said he had noticed a shift on Twitter from Claude Code to Codex over the past few months and asked what had changed internally. Seshan cited the saying about chopping wood and carrying water before and after enlightenment. In her account, nothing fundamental changed. The team that built the Codex app had always been user-focused, iterated tightly, and mainlined the product. What changed was that the market and users noticed. She said everyone building the app is a developer who uses it for development and fixes their own problems as well as those of colleagues and users. She gave the team, and especially the desktop team, full credit. She described them as independent, founder-like people who notice something that should be better, build it, test it internally, iterate if people don't use it, and then ship it externally. When Lenny remarked that the answer was about humans rather than AI, she agreed.
Blurring roles and the unresolved question of craft
Lenny asked about the confusion people feel as roles overlap. Designers, marketers, and PMs are unsure what they are now responsible for. Seshan said she has always liked startups because roles have few boundaries, and she described Stripe as very much that way, with anyone able to do anything. She sees AI capabilities as finally catching up to that mentality. What matters to her is that someone is clearly accountable, a DRI, for whether the product is used, wanted, high quality, and effective. Whether that person is an engineer, designer, or PM doesn't matter, and other work can be picked up based on affinity and capability.
She also expressed real uncertainty. She loves the craft of product management that people like Shreyas Doshi, Marty Cagan, and Shishir Mehrotra have written about, and she understands the worry that fluid roles mean losing the chance to hone a craft. "I truly don't have an answer for that question," she said. Some parts of craft are being abstracted away by models that may do them better than individuals, and craft shifts to applying skill elsewhere in the product or discipline. She said she is still working out how to balance her enthusiasm for how effective people can now be with her love of craft. She noted that engineers used to enjoy handwriting code and rarely do so now, and that people mourn the flow state of writing code by hand. She called it a tough transition.
Where humans remain valuable
Asked where human brains will stay most valuable over the next few years, Seshan gave three answers, while noting she can't predict what the models will do. The first is accountability. You can think of your agent as a report, but a person still owns the outcome and decides whether it was high quality and what was intended, especially in regulated industries or settings that need a direct human interface. The second is expression. She returned to the film analogy: the best films aren't the biggest-budget ones, and software has authorship. Choosing what to build and how it feels is, in her view, a deeply human question. The third is how people care for and relate to each other. Talking with teammates, building shared enthusiasm, learning together, and raising each other's ambitions has become more important in her work than ever.
How she uses AI: Sites and /visualize
Lenny noted the "overhang" between what AI can do and what people actually use it for, and asked for concrete examples. Seshan said she now builds Sites constantly. She described Sites as a product for creating hosted sites with a database. She uses them as presentational artifacts at work and for much more. She built a game the team played together. For a recent backpacking trip, she built a site showing the route and elevation, where everyone on the trip entered their food. She said Sites realize Alan Kay's 1960s vision of malleable personal software. Earlier tools like Notion tried to get there by configuring blocks, but with Sites, she said, it is literally a prompt. Sites can be shared, can update automatically, and can use internal data for things like metrics dashboards, and she called them a more dynamic presentation surface than a slide deck. You just ask Codex, for example, to "create a site that is a mafia game for my team." Hosting is handled for you, and you can make a site public, team-only, or private. She said you can create them through Work, Codex, the web, or mobile, and that making sites has replaced much of her former habit of producing docs and sheets.
Her second tip was the /visualize command in Codex. For example, asking it to visualize your ChatGPT usage so far pulls in your history and produces a visualization. She said she has lost count of the times she needed to turn charts and data into something that told a clear story, and that /visualize makes this "surprisingly delightful."
During the conversation Lenny started a Site about Seshan. By the end it had produced a designed personal page with quotes and themes such as "big conviction, small teams" and "start with the buyer." She liked it, though she was amused that the header photo appeared to be her Stripe badge photo from when she was about 19.
Writing as thinking versus writing as reporting
Lenny passed along a prompt from Brie Wolfson, who said a "Tara brief" is iconic. Seshan distinguished two kinds of writing. Writing as reporting covers status updates, weekly summaries, and launch plans, and she happily automates it with models. Writing as thinking covers briefs arguing for a product, a strategy, or a "spicy take," and she said she will never automate it. For her, outlining, turning an outline into prose, cutting, and revising is how her ideas take shape. She thinks most people paint with too broad a brush, either always or never using models for writing, and considers both wrong.
Her process at Stripe, which she described as strongly writing-oriented and one of the few places where a brief could go viral internally, was to go deep on a brief, refine it, then shop it around and ask people to attack it. She would strengthen it and take it to the next person, repeating the cycle. She did this many times to start new product areas, propose changes in direction, or analyze problems. She mentioned Jeff Weinstein as another Stripe colleague known for writing briefs. A former manager told her to write a doc to about 70% and bring it to the people whose buy-in she needed to get it to 100%. Very few great people want to engage with a perfectly polished idea, she said, because their own ideas "just bounce off of it," while rough edges invite them to shape it with you.
At OpenAI, she said, she still writes to think, often hundreds of docs, but mostly for herself. She said a long doc no longer signals careful thinking, since anyone can now produce one without having thought something through. She described this as perhaps her biggest and most jarring personal change. She used to think in a document and then translate it into a presentation as proof of thinking. Now she favors "mocks not docs" or "prototypes not docs," and ideally results: "we ran an A/B, here are the results, this is why we should go in this direction."
Avoiding "AI brain rot"
Asked how she avoids over-relying on AI, Seshan said the writing-as-thinking discipline is her main safeguard. She also follows a personal rule: if she will ask people to read a document, she should invest at least the combined time they will spend reading it in writing it. Likewise, she preps for a meeting for at least the combined time attendees will spend in it. She said she doesn't rely on the model to polish her prose, adding that she doesn't think it really does that well, and especially not to generate a first draft. She does use it heavily for summarizing and converting content between formats. Her summary: "I start myself and I end myself." In the middle she may use AI to research specific points, pull in data, or push back on ideas.
The lesson from Sutter Hill: product-marketing fit
Seshan described Sutter Hill Ventures as intentionally illegible. Its website shows almost nothing, and it operates quietly, yet it is behind major successes such as Snowflake through the incubation model started by partner Mike Speiser. What struck her was that people treat finding product-market fit and building a tens-of-billions-dollar company as a dark art of luck and chance, while Speiser has done it repeatedly, which suggests there is a playbook. In her account, the firm has learned to be right about calling shots and about the daily compounding work of building a company: setting up enterprise sales, positioning, and assembling founding teams. She singled out recruiting and mentioned an internal tool she called Reticle, which maps the people the firm has worked with and the ten best people each of them has worked with. She said she went there because her career has been about finding product-market fit repeatedly, and Sutter Hill had figured out how to do that for B2B products.
Her main takeaway was that she had underrated what she called "product-marketing fit." The way you talk about and position a product can come before building it. It should combine a deep understanding of the technology with an understanding of the enterprise sales process. Before committing to a product shape, you should pitch around 100 people, refine the pitch, and get the narrative of why the product is transformative right. She admitted she used to see product marketing as glue between functions, and came to see excellent product marketing as work that can determine a company's success.
Knowledge work is not coding: verification and the model as collaborator
Near the end, Seshan raised an issue her team is working through with ChatGPT Work: knowledge work and coding are fundamentally different. Coding is output-oriented, so you can check a result with tests or by running it. Knowledge work can't be verified by looking at the final deck. You can't see "90% success" in a slide and simply believe it. You need to see the process, inputs, and reasoning. So, she said, much of the team's work, now and ahead, is adapting the product for knowledge work: making ChatGPT a collaborator, showing work in progress, surfacing citations and inputs, and letting users follow the model's path so they can trust the result. She listed open questions. Should the chain of thought show more citations? Is the thread interface, well suited to coding, the right surface for knowledge work? Lenny compared it to pitching an executive, where showing the steps behind a plan is much of the pitch, and added that context, such as whether the agent can see your email and docs, is part of the same problem. Seshan said that as the team thinks about bringing human collaborators into agent work, it also needs to make the model itself more of a collaborator.
Lightning round
For books, Seshan recommended Barbarian Days by William Finnegan. She took from it that you can devote your life to something without ever being great at it, and can pursue excellence knowing you won't reach it. She also recommended Anna Karenina, which she has read at different ages. At 13 she followed the plot, at 17 the European history and class dynamics, and at 30 she saw it as a story about a woman and about people. She connected this to the need to see the same things through new lenses as we grow and change careers. She also said she loves The Power Broker and suggested reading long classics chapter by chapter with a slow-read Substack, which she said had covered War and Peace and Wolf Hall. Lenny added the 99% Invisible book club series on The Power Broker.
For film, she chose Christopher Nolan's The Odyssey. Her "hot take" is that it is about AI and how it transforms society, and she said Nolan bridges artistry and commercial success like no other modern director. She also recently watched Akira Kurosawa's Rashomon. She noted that it pioneered telling a story through conflicting perspectives while working in black and white in the 1950s, and she drew a lesson about ambition: with a hundred times Kurosawa's tools in her iPhone, "what's my excuse" for not making better work?
For AI products outside OpenAI, she named software her friends build for small groups, part of what she called the "cozy software" movement. A friend named Sebastian built an app that turns any content into a podcast and drops it into your Apple Podcasts feed, and a private social network for their friends, called GATS, which she compared to a private Twitter. Her motto comes from Toni Morrison's essay "The Work You Do, the Person You Are": do the work well, for yourself and not the boss; you make the job, it doesn't make you; your real life is with your family; and you are not the work you do, you are the person you are.
On the Thiel Fellowship, she called it an inflection point in her life and a moment when someone raised her ambitions and told her she didn't have to follow the path she was on. Her cohort's pitches were filmed for a CNBC documentary, so her 19-year-old pitch is on YouTube. She mentioned Dylan Field from her cohort and fellow fellow Ari Weinstein, whose company Sky was acquired by OpenAI and who now leads much of OpenAI's computer-use work.
For her plug, she encouraged listeners to try Work in the ChatGPT desktop app, on the web, or on mobile, for example by asking it to build a Site about themselves or to visualize their ChatGPT usage. She said the most magical experience is starting a task on your phone, going through a subway ride without service, and finding the work finished when you come out, because it runs in the cloud rather than on a laptop you have to keep open. She confirmed that cloud execution works across all platforms, not just mobile.
If you think about the first era of AI products as chat, the second era of these products working with agents, that third era that might come soon is how do you work with a persistent coworker who is able to get things done with you.
There's this idea of the overhang of what AI is capable of and what we're actually doing with it.
So hard to understand what is going to emerge in the future. 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 outcomes are equally wrong. Only way to build is 2 to 3 months.
What if you had to most adapt to and adjust in how you operate as a PM in this world?
Being prolific and empirical is way more important than being academic or theoretical. Rather than writing out some long reasoning doc, instead, it's like how do I get to something I can try out and test with users as fast as possible. It feels like not only are we able to be more ambitious, we almost need to be more ambitious, which is not natural for a lot of people.
Elevating others' ambitions or reminding them of what's possible here is a huge part of the product management role.
I'm curious what's most surprised you about what it's actually like to work at OpenAI.
I came into the company expecting that there was a treasure trove of OpenAI secret strategy, and actually OpenAI is open.
Today my guest is Tara Seshan. Tara leads product for both Codex and ChatGPT Work at OpenAI. I believe this is the fastest growing and arguably most important AI product for knowledge workers today. Tara works alongside Andrew Ambrosino, who was a recent podcast guest. He's her engineering manager. Prior to OpenAI, Tara spent six years at Stripe, where she joined as one of the first five product managers, and for many of those years she was named one of the top three Stripes across the entire organization of Stripe. She also led product at Watershed, was a founder and a Thiel Fellow, and most importantly of all, Tara was one of the three Lenny's Newsletter Fellows, which is a program that I ran a few years ago to highlight some of the most amazing up-and-coming product leaders. I am so excited to see Tara in this new incredibly important and impactful role. Before we get into it, don't forget to check out lennisproduct.com for a free year of the hottest and most beautifully crafted AI products in the world, available exclusively to Lenny's Newsletter subscribers. With that, I bring you Tara Seshan.
Tara, thank you so much for being here and welcome to the podcast.
Thank you, Lenny. I'm so glad to be here. It's so nice to see you.
I'm even more glad. So you've been at OpenAI for just about a year now, which in most places would be a very short amount of time. In AI time, that's like a lifetime.
Yes.
I imagine when you joined OpenAI, you had a sense of what it was going to be like to work at a frontier lab. I'm curious what's most surprised you about what it's actually like to work at OpenAI, and ideally both good and bad stuff.
So many things about working at OpenAI felt familiar to me because I had worked at other places that were, you know, high growth, high talent, high intensity, hyperscaling mode places before. And so some of the things like, oh, my colleagues are so awesome, or the urgency is really high, felt very familiar.
The part to me that actually felt the most surprising is that many companies I've worked for, in fact all the companies I've worked for in the past, have been founder-led, and 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 top-down direction at OpenAI is extremely limited relative to places I've worked for prior.
And so I think when I first got to the company, that was both delightful, and I had come from a founding journey before, and I was like, yes, I can continue to feel like the founder of this product area or this team. And you know, the distance between me and the market is very, very thin. And sometimes at a larger company you feel insulated from what users want or feel insulated from what the market demands, but actually at OpenAI you do not at all. You are doing everything it takes to get product market fit for your product, akin to how a founder might.
But the counter to this, or maybe the more surprising side of this, is I came into the company expecting that there was a treasure trove of OpenAI secret strategy that I would be able to understand, akin to how, you know, at past companies you come in and you're like, ah yes, this is the payments bible and this is how we think about payments and operations. And actually OpenAI is open.
Every sort of thought that exists in terms of this is how the world should look, or this is how product should be built, or this is how the model should operate, very, very quickly becomes a part of the public product or a part of the public messaging. And so that to me was incredibly both positively surprising and just a change in my operating mode for sure.
Telling us there's not the secret room with the AGI running there with the master plan that has all the answers.
Or at least I'm not in that room for sure. But I think the piece that is really inspiring to me is that 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 have seen.
This episode is brought to you by our season's presenting sponsor, WorkOS. What do OpenAI, Anthropic, Cursor, Replit, Sierra, Clay, and hundreds of other winning companies all have in common? They are all powered by WorkOS. If you're building a product for the enterprise, you've felt the pain of integrating single sign-on, SCIM, RBAC, audit logs, and other features required by large companies. WorkOS turns those deal blockers into drop-in APIs with a modern developer platform built specifically for B2B SaaS.
Literally, 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. Whether you are a seed-stage startup trying to land your first enterprise customer or a unicorn expanding globally, WorkOS is the fastest path to becoming enterprise ready and unblocking growth. It's essentially Stripe for enterprise features. Visit workos.com to get started, or just hit up their Slack where they have actual engineers waiting to answer your questions. WorkOS allows you to build faster with delightful APIs, comprehensive docs, and a smooth developer experience. Go to workos.com to make your app enterprise ready today.
You've been a PM at a lot of different places, a longtime PM leader. What do you lose in this new world?
When a market is more static or a market is more slow-moving, you have the chance to actually do some grand strategy-esque work because it's more predictable, or you can at least understand all the pieces. As an example, payments is certainly a dynamic market to some extent, but it's also an established market, and you're able to say, ah yes, if I take this bet, my competitor might take this other bet, or reason from first principles very rigorously through what all the next actions might be. And in fact, the nature of that market mandates that you do that. Winners will think more rigorously than everybody else. And if you aren't thinking rigorously, it shows up as sort of carelessness, because a lot of those decisions that you made could have been predicted.
But in this market, it's so hard to understand what is going to emerge in the future. It's very emergent. It's very fast changing. It's really dynamic. And most importantly, it's very, very important to stay tied to the research. And so actually being prolific and being more empirical is way more important than being maybe more academic or theoretical.
And I think lots of the past companies I've worked at have been very academic and theoretical places. And it was a real switch to go from, rather than writing out some long reasoning doc, almost like a PhD thesis of what I think should be the plan for the next N 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.
And so yeah, that switch from theoretical to empirical felt very jarring at first. I was like, oh, am I not doing my due diligence here? Am I not being thoughtful enough? Shouldn't I be thinking through all of this in a ton of rigor? But actually you got to try stuff and learn as much as possible. 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. To use the Shishir Mehrotra phrase, the eigenquestion. What is that specific most important thing to test? And everything else, any other grand strategy you concoct, is not relevant.
I'd love to hear more about that, because that's really interesting as almost like, here's the thing of the PM role that is not changing. So much is changing. The world is changing, but there's still this piece that is even more important. Speak more to that, of what specifically you think people need to focus more on.
Yeah, there are so many trappings around the PM role of, you know, running execution on time and writing all these specific docs and presentations, etc. But the core of it has always been about what is the most essential question you need to ask about your product. What is the thing that will determine whether your product works or doesn't work? How do you test that? How do you look at the results? And how do you feed that back into a loop of refining your hypothesis and running it again? That truly has always been the PM job. And that involves of course trying to understand users, trying to understand the market, trying to understand the actual technology you're building, pulling those three things together to make the most sharp hypothesis you can, and then making the test as fast and effective as possible.
And I think that is not only not changed, but it's become the most important thing at the company to be able to do. EMs are thinking this way, engineers are thinking this way, data scientists are thinking this way, designers are thinking this way. Everyone has sort of moved to focus their efforts on this really, really important problem definition and testing loop, the what are we actually doing and how do we know if it's working thing. And from a PM standpoint, it's great, because PMs have always been really focused on trying to get that stuff right. And that has always been the core of the job, and actually many of the other trappings of the job have fallen away, and that remains the key thing to get right every time.
You mentioned this idea of a loop, and there's a lot of talk these days. Loops were so hot, I don't know, a few weeks ago on Twitter, and it feels like it continues to be a topic of discussion for knowledge work broadly. And the way I understand a loop is essentially: AI, here's what success looks like, go off and build and figure it out until you achieve success. How do you think about this idea of loops expanding from just software engineering to product management, all knowledge work? Do you think that's going to be a thing?
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 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.
And to that point, I think that that steering might grow higher and higher and higher level. The steering used to be at the level of: I wrote this line of code, press tab, to: oh wait, now I'm directing something a little bit more comprehensive, to maybe the goal level, maybe to an even higher level. I think the steering will continue to maybe go up layers of abstraction, but ultimately I think it's still on a person to be able to find which direction are we pointing this in, and given feedback and additional data, where do I want to take this thing next.
Some of steering I think is about certainly what the data tells you, but a lot of it is about making an opinionated call. I think sometimes we underrate that power of intuition, or even sort of positive determinism of what we want the future to be like. Picturing, hey, I would like the product to look this way, not because the converse is not an equally viable strategy, but because I would like the world to look like the direction that I'm pushing it in. And that I think will always remain an opinion, at least right now, that is required from a person.
And so I think that loops are awesome. Running agents in increasingly larger loops where they're doing more and more of that rowing for you is great, but right now you really still need to steer. And I think work will also look like steering with other people over a group of agents that you guys work with together. Bringing in other teammates into that interaction between you and the agent, where it's rowing and you're steering, feels also incredibly valuable.
That's such an interesting way of describing it. There's two thoughts here that come up. One is, if everybody has access to the same tools, the thing that will separate you is the human, the person, basically. Otherwise, we're all just going to be building the same thing. Everyone could be asking, how do we win? What do we do? And then the thing that's almost an unfair advantage is the human brain.
Yeah, I think it reminds me a lot of fashion actually, in some ways. There are certainly functional clothes that everybody can wear and that get the job done, but so much about what you wear, or at least how I think about what I wear, is about what statement I want to make about my individuality, or how I want to reflect to the rest of the world. And a lot of what makes that compelling is how it contrasts with other people's expression. The shirt I wear makes a statement only because it is maybe different than what everybody else is doing, or different than what some cohort of people are doing, or makes a statement about my group membership or something of that kind.
And I think a lot of the products that we build feel similarly opinionated and artistic. Patrick Collison has this really nice statement, or maybe it's John Collison who has this really nice statement about software, which is that software is not like real estate. You don't 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 auteur statement, or some opinionation and artistry, that goes along with it. And I think that relies on you having something interesting to say, or your team having something interesting to say, about your product.
There's something Marty Cagan is big on, which is this idea that when you have an idea for a product or feature, rarely is that idea the thing that it ends up being. There's this whole process you go through to kind of figure out what the hell it actually should be. And it feels like that's kind of what you're saying here, that you need to go through that process as a human to understand what it really is and what people actually want. It's never going to be like, "Okay, got it. Go build this thing. I got it from the beginning."
Yeah. For sure. For sure. And those loops are moving faster and faster and faster. And so your ability to form those intuitions, get the information you need to form those intuitions, and then use that with people and agents to put that into
action is the key.
I'm curious what you think the next shift will be in how we work just broadly as knowledge workers. It feels like not only do you have access to the most advanced tools that other people don't yet, also you work around the most AI-pilled, AI-forward people in the world. How are people working internally that you think will become kind of a more normal way we all work using these AI tools in the next, I don't know, 3 to 6 months?
Yeah, I think there's two aspects to this. One is continuing to work with agents at higher and higher levels of abstraction. So letting the agent do more and more for you independently, coming in, providing that steering, and then letting the agent continue to cook. Like, let the agent cook and provide details at higher orders of abstraction feels like the way 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 they do work again, and we sync up at different cadences, look at each other's in-progress work, and provide more and more feedback. It feels like that coworker model is the way that things are certainly going. It feels like a much more natural interface for us to be able to work with agents, and we already see a lot of that internally as well.
The second is that a lot of my work with agents thus far has been one-on-one, that I work with my agent. Maybe it's spawned some subagents to get some tasks done, but it's me and my agent together. And that is potentially divorced from what my colleagues are doing with their agents. And so there was a time where everyone internally was just sending their Codex threads, screenshots of their Codex threads, to each other on Slack. We're like, okay, well, I wanted to share with you how I got to this number. Here's how I got to this number. Here's a screenshot of what I did. But that's also not quite the most natural way for someone to collaborate together. And so as more and more work gets done with our agents, shouldn't we be able to get work done with our agents together? And what is the most natural interface to make that happen? And those are some of the things that we're thinking about.
Chat is what I'm picturing. That makes so much sense. It's like, okay, here's Tara's agent. Here's my agent. She did some work on some analysis. I'd be, hey, my agent, Lenny's agent, go check, make sure this is legit and connects to the way I think 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 rowing, tactical tasks.
It's interesting how it's just been this slow progression of trust and just awareness that this can be how we work. Just this, okay, go work for longer, you can take on more. There's been this talk of the slow takeoff, the fast takeoff scenarios, and everyone's afraid of this fast AI takeoff where it's way too smart and now we're in big trouble. It feels very much like we're on the slow takeoff scenario, which is good, where it's just slowly iterating. It doesn't feel that slow, but in a sense, you know, we're not some 300 IQ AI, you know.
I mean, the models are incredibly smart, but I think a lot of the things that have enabled us to then work with our agents together or have the agents take care of higher and higher order abstraction things certainly are about the intelligence, their ability to perform long-running tasks and how long they can stay on tasks, but also actually there are very meat-and-potatoes tactical things that make this possible. Like, agents working locally are really convenient because they have access to all the data that's on your machine. To make an agent successful in the cloud, there is a ton of cloud infrastructure that you have to build to make that possible. And just access to your systems, like how can agents talk to all these third-party systems that have all of your data. Just like a colleague who you hire, who you lock into a room, never give them access to Google Docs and Slack and, I don't know, the company database, would not be that useful to you. Similarly, a cloud agent that is similarly isolated will not be that effective. And so a huge part of making these agents useful and achieving some of these futures is 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 much more prosaic than some of the broader intelligence questions, but matter in some ways just as much for end effectiveness.
This touches on something else that has been coming up a bunch on this podcast: this word ambition. I know you think a lot about this too. It feels like not only are we able to be more ambitious because of these AI tools, we almost need to be more ambitious, which is not natural for a lot of people, because everybody can now do all these easy things really easily. Like the easy stuff is super easy. The hard stuff is easy. And the thing that separates people now and companies now is just how ambitious they can be. Talk about what comes up when I talk about the need and kind of the emergence of this need for ambition.
Yeah, 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. Like back in the day, you know, before all this AI stuff, the unicorn person was someone who was a really thoughtful product sense person who also happened to be an engineer, who may also have been a designer. That person was always the unicorn hire because they were able to really flatten the layers of translation needed between all these functions and were able to build something or ideate something really quickly and easily themselves and get it up and running, and then were able to work with a team and collaborate with the team on it.
And I think the most compelling thing I found, that certainly I try to be able to do with these tools and I've seen some of my most successful colleagues be able to do with these tools, is really expand the set of things that are quote unquote within their range of possibilities, so that they can start realizing more and more of what's in their head into reality, the way that someone who was previously a jack of all trades was able to do. We kind of all have that superpower now, that I can spin up a set of designs on something and I can go build an initial prototype of it and I can, you know, figure out the right pricing model for it and model out all the scenarios. Really the set of possibilities has widened dramatically. And actually what that means in so many ways is that I have the ability to, to that point earlier about film, be more of an auteur as I try to get something done and realize my vision, maybe to higher fidelity.
And that to me is part of what can elevate your ambitions while pursuing new ideas and new products: that because all of these things are now within reach, because this new set of capabilities is now within your reach to be able to try and access, you're not really limited. Your ambitions are no longer limited by what you're capable of executing yourself, what you're capable of communicating. It can be so much wider. I think the hardest part about doing this is simply just expanding your thinking. Actually, the capabilities have expanded so dramatically. It's really expanding your thinking of what's possible in an unreasonably short time frame.
And to me, the best way of trying to do that is, Patrick Collison has on his website, patrickcollison.com/fast, I think, which is all of these projects that were unreasonably ambitious that were executed in a really, really short time period. And what for me is now remarkable about that list of projects is that they all existed before these tools made it possible for you to learn how to build something almost instantly, or ask it with one question, hey, can you summarize this very complicated text or this very complicated book for me, immediately? Or can I try to do all of these things that were previously impossible to me, but now I'm able to do? Like, can you make for me a CAD model of this idea that I might have? Really, capabilities that were truly beyond my reach are now in my reach. And so if those fast projects were possible before with the capabilities we used to have, shouldn't we just see an exponential increase in the number of those types of unreasonably quickly and effectively executed things with what AI has given us?
To your point, the hardest part is just remembering to even try, just to be like, "Oh yeah, well let me see if Codex can do this for me." It's just a new habit, a new thing we have to build in our brain.
Tyler Cowen has this statement on his site, which is that most people underrate the impact of going to someone else and saying, hey, what is the more ambitious version of what you're doing, or couldn't you try this faster, or couldn't you try this at a 10x bigger scale? And in some ways, again, when I think of what PMs do that is incredibly effective now, or what they can do that is incredibly effective now, 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 version of it," part of your job now is to elevate everyone's ambitions and say, "Actually, isn't the possibility ceiling meaningfully higher?" Like, shouldn't we be more ambitious about what we're attempting here? Or, couldn't we try this faster? And I think that's, yeah, it's a great place to be. It's a great place to be in terms of what you can build, what's possible, and in terms of, yeah, how exciting the job becomes.
That is so interesting. I remember Nick Turley was on the podcast, who maybe had the role before you. I think he's working on enterprise stuff now. He had this meme internally, "Is this maximally accelerated?"
Yes.
There's an emoji, I think, inside the Slack. "Is this maximally accelerated?"
"Is this maximally accelerated?" is totally an OpenAI meme. The other OpenAI meme that Andrew Ambrosino and I love to ask the team is, "Are you mainlining it yet?" which is, are you using this product all day every day to get your thing done? And I think that, in combination with: are we being as ambitious as possible, which is about the scope and the scale of what you're trying to do; is this maximally accelerated, are we moving as fast as possible on it; and then, are you mainlining it yet, are you using it, and are you bringing all your taste to bear on whether this thing works and is something that people really want, and tightening that feedback loop as much as possible. Those to me are the three memes of product development that we just have to spread as much as possible now.
I love that. That's the new dogfooding. Instead of dogfooding, you've got to mainline it.
Yeah, exactly.
And that shows so deeply in the tweets. This is mostly how I see your team communicate, of just how obsessed they are with the product, and just constantly asking what can we do better? What's bugging you now? Here's the thing we're building. It's very clear how, to your point earlier, everyone is just the founder of their product, and it's very clear how they act, as an external observer. Are there any other memes internally? Those are so interesting. Any other, I don't know.
Yeah, I'm trying to think if there's other good cultural memes. Certainly a really important one is feeling the AGI, or just being conscious of AGI coming. There are so many outcomes for what it could look like or how one thinks about it. But a huge part of what puts most people at this company is believing in that mission of AGI being beneficial and trying to do whatever it takes to make that possible, both the realization of AGI and ensuring that it is beneficial for humanity.
And in building products, another just constant refrain I have to keep in the back of my mind is, are we building for where the models are going to 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 outcomes are equally wrong. And I'm sure many people have talked about this, but both outcomes are really equally wrong. If you're too early, you're wrong. If you build something that was overly focused on a past model's capabilities, you're entirely wrong. The only way to build is two to three months out, and having this meme of, models are going to get way better. I need to think about the model capability as the center of this product. I need to get out of the way of the model in terms of the product constructs that I create. How do I ensure that this is right for the model in two to three months' time?
How do you know what two or three months is like? It's a challenging understanding, especially when we're on this exponential. Is it just a gut feeling? Is there anything the researchers give you a sense of? How does that work?
Yeah, certainly communicating really tightly with research on where they think things are going is incredibly important. These things aren't entirely a black box, in that you kind of know, hey, we're focused on these particular things, like we would like models to be better at coding in these specific ways or better at writing in these specific ways. So we certainly have focused efforts on making the model better at specific capabilities. And so knowing where that is and ensuring that product development is as tied as possible to what research has as its agenda and its roadmap is really important.
A quote that I'll never forget is when Kevin Weil was on the podcast. He was chief product officer at that time. He said that this is the worst the models will ever be. And it sounds so simple, but it's hard to just wrap your head around that, that this is the worst they will ever be. It's such a cliche almost now to say that, but it's true. It's absurd.
Yeah, it's absurd. It's absolutely absurd.
Oh man. Okay. I want to talk about ChatGPT the app briefly. Okay, so I have it open right now.
Yes.
Okay, so here's what I see in it: ChatGPT, and then there's a dropdown, and there's ChatGPT and Codex, and then there's this toggle, Chat and Work. Tara, what is going on? What are all these things? Help us understand what each of these things are for, and where do you think this goes? Is it going to stay like this? Is there a next step that you're imagining already?
Our north star here is that users do not need to make decisions picking between all these different options. There is no toggle here; you go to the box, you type in your task, like, I would like to build a really awesome app that, I don't know, helps my podcast guests do research before episodes or something, and it will just pick the right harness. It'll pick the right model for you to be able to get that thing done. Ideally the choice here is not on our users to have to pick between all these different concepts and understand not only what they are trying to do, but understand the limitations and capabilities of our products. So that is certainly where we want to go in the near term.
Picking between ChatGPT and Codex is really a choice for: do you want to stay in a sort of more development-oriented UI, or do you want to have the same power and capabilities in the ChatGPT mode? And
So if you're a Codex user, keep using Codex. You're not missing out on anything. Continue using it as much as possible. But if you're a ChatGPT user who is like, what are these new agent capabilities, you should probably be in ChatGPT mode. And then when you're in ChatGPT, if you want to have conversations, if you want to search, that's where chat mode is the right thing. It's the same chat mode you know and love, with better and better models and newer and newer capabilities every time.
But in work mode, that's where under the covers this is Codex. We've removed some of the coding UI, like you're not going to see a worktree pop up all of a sudden in work mode, but it is the same power to get things done, for example, to generate a really complex financial model. That's all possible in work mode. And we see people, especially, I mentioned our corporate finance team, use work mode to do incredible, incredible things. Things that were previously either manual or required deep expertise from one person on the team become things that the whole team can execute, or it just elevates the ambitions of everyone on the team in terms of timeline or capabilities or the frontier of what they can get done.
Okay, that's really helpful. So there's kind of like these three modes currently. There's the engineering mode, the chat mode, and then the do-knowledge-work mode. And the knowledge work mode is actually Codex doing all that work, but people may not know what Codex is, may be afraid of it. Is there anything in that work mode that's not just Codex? Because that's actually really interesting. Are there additional harness tweaks to make it feel a little different, or is it just the same thing with a little different UI?
It's really at the UI level. So work mode and Codex mode: if you go to Codex and ask it to generate an amazing financial model to price your product or something like that, or like, predict my revenue for the next six months, Codex will do as good a job as work mode. It's really about, whilst it's doing so, what kind of UI do you want to see in the chain of thought? What kind of technical detail do you want exposed to you? It's similarly powerful, and so Codex users aren't missing out on anything by not switching modes. In fact, we do not want them to... like, stay in Codex and do all the stuff you want to do in Codex, and we will show you the appropriate UI based on the things you asked for.
Truly, our north star is to merge all these things so that users don't have to make any of these decisions. The separation is really more about how can we meet people where they are as much as possible, in terms of the products that they use, in terms of their familiarity with concepts, and make sure that we are enabling everyone to take advantage of working with agents, which has entirely transformed the way every single developer works. We should do the same thing with knowledge work.
It makes sense. Because things move so fast, I imagine somebody's like, let's try Codex, this is going to be awesome, and then it takes off and there's 10 million monthly active users, and then they're like, wait, what are we doing here? We've got ChatGPT, we've got Codex, how do we... So it makes sense why these things, you know, it's not going to feel obvious and perfect for a while, because you have to kind of adjust as things work and things don't work, and there are these transition periods of like, okay cool, now let's get people moving towards this vision of the super app. Okay.
I imagine one of the hardest parts of your job is balancing this 100 billion MAU product, ChatGPT, maybe the most successful consumer product in history, with Codex, which is this new thing, and other new things that you guys want to try. How do you think about that? Just balancing these very innovative, fast-moving teams and products with this, like, okay, there's a billion people using this, we can change this dramatically.
Yeah, I think one of the most interesting things here is that 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. 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 that has been coding agents. We'd like to bring it to more domains, certainly like knowledge work, and that is part of the goal of giving all these billion chat users the power of work.
Certainly the product challenge that's on us is how do we not only bring it to them but make it natural and easy to adopt. Make it not a decision they have to explicitly make; we can just help them do the right thing. How do we decomplexify it so they don't need to think about things like harnesses, which feel like crazy concepts for a billion consumers to understand. So that is primarily the challenge. And then of course, that third era that might come soon is how do you work with a persistent coworker who is able to get things done with you, maybe collaboratively with other people.
And so part of this challenge in the near term is we're introducing agents to a billion people who may not have experienced them yet. How do we do so in the easiest, most natural and most usable way possible? Certainly, there's a lot more for us to do to make that happen. But part of this is also a lesson I've had, maybe contrasting the pre-AI era or past product experience with this one, which is that at previous companies, polish was king. Getting every UI interaction or 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 such, if every corner wasn't perfectly polished and everything wasn't exactly correct, you might as well not ship it.
But I think what's been really compelling and interesting about this era and this product experience has been that getting the product in the hands of users, when you have so much conviction that, hey, it's transformative, is way better than perfect, and that urgency and that introduction of the product is so important. So we have a lot to do to make it more usable and easier for chat users certainly, especially for folks who are maybe not even using it for productivity but using it for consumer tasks. But yeah, done is better than perfect, and we have so much more to do.
Yeah, I remember when this app first launched, there were a lot of comments about the confusion, and seeing how quickly the team iterated and responded to the feedback is exactly what I'm hearing here: get it out, figure out what's not working, how people are using it, iterate quickly. Feels like that's the model now.
And of course, there are things where you can continue to iterate and get that feedback prior to launching. And there's a lot that we can and should always do better. But iterating as quickly as possible and listening to the right signals, regardless of whether that's pre-launch or post-launch, ideally pre-launch, is the key thing.
This episode is brought to you by Mercury. Radically different banking, now with Spend. I've been a Mercury customer for so many years now. I switched all my business banking to Mercury and honestly, I could not be happier. It's what online banking feels like when it's built by product people, not by bankers. And now with Spend, you can give your team individual cards, set spending limits per person or per team, and have expense receipts automatically pulled in from Gmail or over text. You can even give your AI agents their own cards with their own limits and policies.
Most founders start out the same way. One card used by everybody at the company. It works until it stops working. Someone goes over, a receipt disappears. You spend two days trying to figure out who spent what and why. Spend is expense management built directly into Mercury. All your team's cards, budgets, and reimbursements all live in the same place as your business banking. No chasing, no manual reviews, no end-of-month scramble. The result is a team that can move fast and a founder who is no longer the bottleneck. Learn more and get signed up at mercury.com. Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., members FDIC. The IO Card is issued by Patriot Bank N.A., member FDIC, pursuant to a license for Mastercard International Incorporated.
Something I've noticed on Twitter is there's definitely been this vibe shift from Claude Code to Codex in the past few months. It used to be everyone was Claude. More recently, it just feels like people are leaning towards Codex, at least on Twitter, which is a bubble, but it's where a lot of tech people are. I'm curious what's shifted internally in the past, I don't know, three to six months, other than Tara joining and shaping up the ship. Is there anything that you can share that's just like, okay, we figured this thing out, we shifted this, we cut this thing? What helped shift the vibes and help Codex become as successful as it is becoming?
You know, I think there's this phrase: before enlightenment, carry water, chop wood; after enlightenment, carry water, chop wood, sort of thing. And actually with the Codex app, the team who initially got it up and running and were working on it were, again, super user-focused, with a tight iteration loop. They really dogfooded the thing, mainlined the app as much as possible to get everything right. Folks started to realize that was happening externally and on Twitter, and users started to really notice, but the team was always really focused on users, really focused on that iteration, and it was merely, to some extent, the market catching up that was the change.
And that process has not changed internally. Everyone still constantly uses the app. Everyone who's building it is obviously a developer using it for development, and is constantly fixing not only their own problems but trying to listen to other people in the company's problems and user problems. Actually, what's sort of remarkable is that the mode of operating hasn't changed. It's always been the same thing I mentioned earlier: are we elevating our mission sufficiently? Are we maximally accelerating progress, and are we mainlining it as much as possible? And I think it's great that users and folks on Twitter have noticed, but that operation, full credit to the team, hasn't changed.
What's really interesting about this answer is it's the very human part of it. It's you, it's Andrew, it's Tibo, it's the team just being obsessed with the customer, the product. And it's not like AI was the answer. It's the humans that made the difference.
Yeah, the team deserves full credit here. Everyone on the team is incredibly thoughtful and independent, to the point where, you know, there are many founders at OpenAI. Almost everyone on that team, the desktop team especially, acts like founders and cares about every piece and every detail, and when they notice an area that should be better, they go build it very independently and get the thing up and running. And if it doesn't test well internally, if people aren't using it, if people don't find it useful, they'll iterate on it and then finally ship it externally. But for that loop, full credit to the people on the team and individuals for making that happen.
Something you touched on is this idea of roles overlapping, this idea of, you know, engineers doing PM work, you probably shipping prototypes and building, maybe shipping to production. It feels like that also creates a lot of challenges. I hear from a lot of people: what is my job now? As a designer, what am I responsible for, what am I not responsible for? As a marketer, what am I doing? Is that something you notice? Is that something that you're dealing with? Just any thoughts along those lines.
I think the thing I've always liked the most about working at startups, and sometimes I've started at a startup that accidentally grew into a large company, but primarily working at startups, is that there are very few boundaries around your role. Everything and nothing is your responsibility; ultimately you're accountable for success. Actually, Stripe was very much this way, where there were no boundaries around what an engineer could do versus a product manager versus a designer. Everyone could do anything. And so actually it kind of feels like I've always really loved that mentality, and now finally capability is catching up to that.
But the thing I really care about is that someone needs to look after, or have core accountability for: is this product being used by users? Is it something that people want? Is it high quality? Is it effective? And whether that person is an engineer or a designer or a PM or whomever, someone is the DRI, and then whatever work needs to be done to make that possible, certainly people can pick it up based on their affinity, based on their capability. But I like a team that doesn't really mind what the boundaries are between individual roles, where everyone's just focused on making the outcome happen.
The converse of this is I also really love the craft aspects of being a PM. There are so many aspects to PM craft that folks like Shreyas or Marty Cagan or Shishir, all these people, have really espoused, that I think are wonderful. And maybe some of these questions come from: wait, I so love the craft of my domain; by taking this more fluid approach to teamwork and collaboration to get something done, do I lose out on getting better and polishing my craft? And I truly don't have an answer for that question. I think it's something we're all experiencing together, which is that some pieces of our craft are actually getting abstracted by models being able to do them 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 other part of the product or the discipline.
But yeah, that is still a question I'm thinking about, which is how do I balance my desire to be part of a team and use these tools, and feeling so compelled by how effective one can be now with all these products, with my love of the... Yeah, it's really fun handwriting code as an engineer all the time, and one doesn't really do that anymore.
Yeah, that's where I was going to go. It's just unbelievable how different the engineering role is now.
Yeah.
It's like you used to write code all day. That was your job, and that is no longer your job.
Yeah.
And that happened so quickly.
People mourn the flow state of writing code.
Yeah.
Manually, yourself, versus what one does now. But I think it's a tough transition.
Yeah. And you know, some people love it, some people don't. And that's a whole other topic. Kind of along those lines, something I like to ask people at the frontier of AI is: where do you think human brains will continue to be valuable in the future? It's impossible to predict long term. Will we need humans? Hopefully. But I'd say in the next couple of years, where do you think human brains will continue to be most valuable?
I think humans will continue to be most valuable certainly as an entity of accountability. So who ultimately owns the outcome here? In some ways, you can think of the agent that you're working with as your report. Ultimately, who owns what the end product was? Was it high quality? Was it the thing
that you wanted it to do and say, like, that will certainly remain a person, at least for now. And especially in industries and places that are highly regulated or require a direct human interface, that makes a ton of sense to me. I think the human brain is also really valuable for expression. I'd mentioned earlier that analogy of software is not like real estate. It is more like a film, where you could put money in and a great film does not come out. The greatest films are not the ones with the biggest budgets. And given that there's a certain artistry and opinionation and expression in building software, where you feel like there is some authorship by a person or a group of people, that part remains to me so human. What you choose to build and how it feels feels like such a human question.
I also think the human brain continues to be valuable in how we care for each other and relate to one another. That piece of my work has remained so human and has actually become more important than ever: the part where you talk to other people on your team and collectively figure out how you can be enthusiastic about an area, how you learn and work together, how you elevate each other's ambitions. All of that feels and remains such a human thing to do. Yeah, I think the human brain will continue to be so valuable in that regard. That said, I can't predict what'll happen with the models, but those pieces feel to me to be incredibly, incredibly human.
I love that answer. You talked about this idea of this overhang of what AI is capable of and what we're actually doing with it. It feels like one of the biggest gaps is, okay, what should I do with it? I'm curious, what are some ways that you use AI in your work that may inspire people, like, oh wow, I didn't think about using it that way? There's kind of two buckets here. One is how your PM job has changed most thanks to AI, where you're like, okay, now I use AI for this stuff. And then, are there any super interesting creative uses of AI recently where you're like, oh yeah, I should try this.
One of the most exciting ways that I use AI in work is I actually build sites all the time now. I don't know if you've tried building sites in
I haven't. Talk about sites.
Sites is a really fun, amazing product. You can basically build a site, certainly in work as a presentational artifact, but I also build sites for literally anything. I built a site for the team as a game, where we all played a game together using a site, because sites have a database. I actually built a site because I went on a backpacking trip recently. I built a site of the route that tracked the elevation of everywhere we were going. Everyone on our trip inputted all their food. It was super fast and effective. Sites kind of realized the dream of malleable personal software that Alan Kay flagged in the 60s, of like, the true personal computer is one that has personal software. In some ways, sites are the tangible way to make that possible. We had all once dreamed of making personal software, and certainly people with tools like Notion, etc. tried with all these blocks to configure what that could be. But with a site, it is literally a prompt. I literally, with a prompt, say, build me this exact tool that I need to get this thing done, and it just does it.
They're sharable. They can auto update. You can use internal data to build a dashboard, for example, with lots of metrics. And rather than painstakingly laboring over some sort of slide deck, a site is just a way more dynamic surface for presentation.
How do you use a site? Do you have to do anything special, or you tell it, create a site?
In Codex, be like, create a site that is, I don't know, a mafia game for my team, and it will just do it.
And like, I'm thinking capital S site, but doesn't matter, I imagine.
It just knows what sites are.
So it's, yeah, because it used to be, here's some source code, go figure out where to deploy it. Yeah. And what you're saying here is it just hosts it for you, and
Hosts. You can choose whether it's public, you can choose whether it's with your team, or choose whether it's private to you. They're great. The easy reach of building a site all the time has changed what my day-to-day looks like, which in previous worlds used to look like creating lots of artifacts like docs and sheets and whatever it might be. Now I just make sites all the time.
And you could do that through, I imagine, Work, or can you do it through all the services, Codex, Work?
You can do it through Work. You can do it through Codex. You can do it in the web. You can do it on mobile. [laughter] You can do it anywhere.
Okay. I just kicked off a create a site about Tara Seshan.
Great.
Is that how you pronounce your last name, by the way? I haven't asked.
Tara Seshan, like station.
Seshan. Okay, cool. Okay, cool. Sites. Okay. Any other quick tips while we're on this topic for people? That was a great tip, because I don't think a lot of people know about sites. That's very useful.
Yeah, sites are awesome. The other thing I really love is using visualize in Codex. Have you used /visualize?
No.
Oh, /visualize is incredibly exciting. You can just do /visualize, visualize my ChatGPT usage until now, or something like that, and it will pull in all the things that you've done and create an amazing visualization for it. The number of times that I've been thinking about how do I not only pull in a bunch of charts and data but present them in a way that is understandable and useful for the story I'm trying to tell has been infinite, and visualize makes that incredibly simple. It is surprisingly delightful to use visualize.
These are such good examples of there's so much power here we don't even know about or understand, and that's the challenge you have here. Help us, for sure, help us know all these things. That's why podcasts like this are also useful. Can't put it all in the product. I'm going to go in a totally different direction. I want to talk about writing. I asked Brie Wolson, who knows you well, what to ask you. Funny enough, she suggested questions for the previous podcast conversation I did with Adam Ward from Ger. So she said, "Okay, you should ask her about writing/thinking. A Tara brief is iconic." Help us understand what makes your writing, your briefs, iconic, and any tips that might be helpful for people that are maybe trying to get better at writing and writing documents.
I really strongly believe that I do two types of writing at work. One is writing as thinking and the other is writing as reporting. Writing as thinking is me writing a brief about why we should build a certain product, or why we should take a certain strategy, or maybe a spicy take. But writing as reporting is things like, oh, I'm summarizing the status of what our team has been up to this week and I'm sending over a report about it, or this is our plan for this particular launch or announcement or something like that. Writing as reporting I happily automate, or I use the models all the time to make that as simple as it can be. But writing as thinking is something I never will automate. I really strongly believe that, at least for me, the act of going through and outlining something, turning it into some level of prose, cutting it and editing it, continuing to iterate on it, is one of the most important steps for me to get my ideas in line.
I think most people will actually paint with a really broad brush, like, I will never use the models for writing, or I always use the models for writing. And actually, to me, both those broad brushes are wrong. 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 shouldn't replace your thinking with it. But my briefs in the past, because I write so much as a way of thinking, is that I will go into a hole, write a brief for a new idea or a product, spend a ton of time refining that particular idea, shop it around with people and have them attack the ideas in it as much as possible and poke holes, make it stronger, and then take it to the next person and do the same thing.
And so at Stripe, this is something I did many, many, many, many times over, whether that was to kick off a new product area, or to suggest a big change in direction, or to analyze a problem and suggest a path forward. And Stripe is incredibly oriented as a writing culture. And there are many people like Jeff Weinstein who are also very into writing and sharing briefs at Stripe. Stripe is one of the few places where a brief will go viral inside the company. And so writing as thinking there is really prized, and that's where I did the majority of that writing work.
At OpenAI, I think I still write as thinking all the time. But the sharable artifact here is not really a long doc or a sort of proof of work in that way. Partially because times have changed, and a long doc is not a signal that you thought through something, because you can easily produce a long doc that indicates that you haven't. And so maybe one of the biggest changes I've experienced personally in my day-to-day, which has been a big, maybe jarring change, is that I used to think in a document and then do some translation of that into a presentational artifact, and that would be my indication that I thought through a problem and this is what we're going to do, and the team moves in that direction. And now I am way more on mocks, not docs, or prototypes, not docs. And if I have something that people can try and interact with, or even better, I have results where we tried this, we ran an A/B, here's the results, this is why I think we should go in this direction, that is a way better communication tool than the doc itself. And so 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. That no longer is the best way to talk and communicate. That is probably the biggest change I've experienced personally in this era versus the previous era.
That is so interesting. I really liked your tip of getting tons of feedback on a doc. It sounds obvious, but you can get to an iconic doc, brief, by just cheating almost and getting lots of feedback on it as you're iterating. Yes, to make it stronger and stronger and stronger, versus, cool, here it is, first time, and it's rarely
I previously had a manager who told me that the right thing to always do is 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%. And that still is a thing that I do all the time, because very few great people want to interact with a perfectly polished, finished idea. With a perfectly polished idea, their new ideas just bounce off of it, versus something that has more crags and more rough edges that they too can polish with you together. And I think that bringing people into the process that way, where a doc is an underlying artifact for that, is one of the best ways to collaborate that I've found.
How do you think about AI brain rot and starting to over-rely on AI? This is just a challenge everybody's going to have. Why not use this magic to help look at something, and then we start to lose our ability to write, read long documents. Is there anything you do to try to avoid that?
Yeah, I think this writing as thinking discipline is one of the main pieces that I employ in my day-to-day to make sure I'm not overly atrophying my thinking abilities. I will again outsource all writing as reporting as much as possible to the model, but writing as thinking I have to do myself. And I have this 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 going to call a meeting with a set of people, I need to have prepped the collective amount of time that people are going to spend in that meeting before the meeting. And so when it comes to keeping my thinking sharp, I do that writing for the document myself first and make sure I've invested the collective amount of time I expect people to read it, at least in writing it and producing it. And I don't really rely on the model either for polishing my prose, which I don't think it really does, or especially not in generating the first version. But I do, of course, have the model help me a lot when it's summarization or translation of content from one format to the other, all the time.
So what I'm hearing is, write the idea, the brief, the plan yourself as a human. Write it yourself. Don't start with AI. And even don't use it to improve on the writing. Just keep that all human.
Yeah. At least for me, I start myself and I end myself. I might use AI in the middle to research specific elements, or drop in some data, or go pull some data, or help
push back on some ideas.
Yeah, push back on some ideas, but start yourself and end yourself with a piece of writing, and that doesn't deteriorate your thinking.
Okay, one last question I want to ask, about Sutter Hill. You had this very unusual career step. You were this PM, founder person, and then just, okay, EIR at Sutter Hill Ventures, which is an iconic VC. People can look it up. A lot of amazing companies came out of Sutter Hill. It has a very unique way of approaching founding, where basically they incubate companies. Snowflake as an example. What was that about? What did you learn from that experience?
Sutter Hill is an iconic firm and is intentionally a very illegible firm. If you go to the Sutter Hill website, you will see nothing on the website. It is a firm that doesn't operate loudly. It tries to operate as under the radar as possible, as modestly as possible, yet is somehow responsible for some of the most iconic successes that Silicon Valley has seen. And they have this very unusual incubation model, which Mike Speiser, who is one of the amazing partners there, started and has rolled out success after success.
I think the thing that was most iconic to me about Sutter Hill is that people look at finding product market fit as a dark art, or building a tens of billion dollar company as a dark art, like, oh, it's luck, oh, it's chance, oh, it's all these things that must come together. Yet Mike Speiser has done it multiple times, and so there is clearly a way to do it. There's clearly a roadmap for making that possible. There's a set of things one can do to get this repeatably. It's not just luck, it's not just a dark art. There is a playbook, as it were, and that playbook lives inside of the firm Sutter Hill. And they have figured out how to be right a lot in terms of calling shots and making bets, and they've learned how to be right a lot in terms of the daily compounding things that one does to create a successful company, whether that's how you set up your enterprise sales team, how you position your product, how you build the initial founding team. Recruiting at Sutter Hill is an unparalleled, excellent thing. They have a secret tool called Reticle where they have a map of everyone 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.
So I went to Sutter Hill because in some way my career has been about how do I try to find product market fit as many times as possible,
whether that was as a founder or in starting new products at Stripe or in joining a startup like Watershed. And so Sutter Hill is a place where they've figured out how to find product market fit on B2B products. And I wanted to learn what I could from them.
What'd you learn? What's one thing you took away from that experience other than they know how to do it?
They definitely know how to do it. I think one thing that was very surprising to me that I learned there is that product market fit is sure important, but actually I really underrated product marketing fit. The idea that the way you talk about the product and the way you market it can precede actually even building the product. It should probably come from some sort of bringing together of understanding the technology deeply and then understanding the enterprise sales process, and then that product marketing fit, that narrative, that positioning is actually, even before you build a product experience, the right thing to test. So you should go pitch 100 people, figure out how to refine that pitch as much as possible, get the marketing narrative of why this thing is transformative, right, and then and only then go commit to, okay, this is exactly the product shape. And Mike Speiser is unbeatable at this art.
Previously I'd always kind of underrated PMM work. I was like, it's whatever, it's the glue between these functions, it's fine. And then I realized how transformative that work done excellently is to a company's outcome, and in fact can be the element that makes a company successful.
Amazing. I so agree with that, positioning. We talk a lot about that on this podcast.
Mhm.
Okay, I'm going to show you what sites created real quick. It was running while we were talking. Check this out. Look at this.
Oh man.
As like, make it more awesome, and it made it more awesome.
Multi-product.
Beautiful. Look at this. This is like legit design.
Look at you, got quotes. Big conviction. Small teams. Start with the buyer.
How do you feel about this being your website? Your new website.
I do think that the picture of me at maybe 19 years old at the top is really funny. But yeah, otherwise I love, love the site.
Okay, good job.
Looking good.
I think that was my badge photo from Stripe.
Oh wow. Amazing. I already unshared it, but I love that it built a whole little thing around your head. So cute. Tara, is there anything else that you wanted to share? Anything else you want to touch on before we get to our very exciting lightning round?
Yeah, one thing that we've been thinking about a lot in product building, especially with ChatGPT Work in this new era, is how knowledge work and coding are actually fundamentally different. And one of the surprising things we learned as a part of that is that 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. There is a way to validate it based on the output. But knowledge work is different in that I can't simply look at the deck in the end and see the numbers, oh, it's like 90% success or whatever in the deck, and actually believe that. I really need to think about the process and the inputs and the reasoning and how it went along the way.
And so in terms of how that looks in the product, 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 go on the journey with the model to get to that end output, such that you know in the end that, oh wait, this thing is right, this thing is good, this thing is useful.
And that shows up certainly in the UX of the product quite a bit, but also should show up in things like the reasoning and the chain of thought. Should you see more citations along the way, for example, of how it got to that end state in that data? Is the surface of a thread, which is so suited to coding, the right place for you to see all of that for knowledge work as well? There's so many big important product questions. And so 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.
That is such a good point. I'm imagining an exec meeting where you're trying to pitch the exec on here's what the plan is, here's what I think we should be doing. So much of that is helping them see here's the work I did to get there, here's all the steps. And so it makes sense that you need the AI to show you that same sort of work that it did, the proof of work essentially, versus engineering where, okay, I don't need to know all of the little architectural decisions you made, just what does it look like, is it passing all the tests that we have. So that is a really good point, just how different those two models are. And also there's the context. Does it have the context it needs to do the thing that you want it to do? Can it see your email? Can it see all your Notion docs?
Mhm.
Such a good point. So, I see the challenge in your job. I think all this work is one product. Tricky, tricky. Amazing. Anything else before we get to our very exciting lightning round?
Yeah. Let's jump into it.
With that, we've reached our very exciting lightning round. I've got five questions for you. Are you ready?
Yes.
What are two or three books that you find yourself recommending most to other people?
One book I really recommend to people is Barbarian Days by William Finnegan. I don't know if you've read it. It's about the life of a man who is a New Yorker reporter, but how he fell in love with surfing as his passion. The thing I took away from the book is that one can be deeply passionate and dedicated and have something be your life purpose without you being good at it. And it is about the art of falling in love with surfing and his striving for excellence and perfection whilst knowing that he will never reach it. It is such a compelling and transformative story for how I think one should continue to live our lives. I really, really love that book.
Another book that I might recommend as a book that people should read: I really love Anna Karenina. I've been rereading the classics lately, and I love Anna Karenina because it's a book of layers. And I think that a huge part of what we're going to have to do in this new era is transform ourselves, or take ourselves on a journey to do different things than what we were used to. And when I think about that book, I think about when I was 13 and I read it, I understood basically the plot. When I read it at like 17, I understood the European history dynamics and the class warfare. And then when I read it at 30, I was like, "Oh, this is a story about a woman and humans." And it just reminds me of growth, and that it is possible to look at the same thing through multiple different lenses as you continue to grow. Which I think is kind of the challenge that's ahead for all of us as we consider our careers as well.
It's interesting, on both these I could connect to AI in the time we're living in now too. I also recently read Anna Karenina.
What did you think?
Earlier this year. Amazing. I've never read it before. I saw it on a book list of here's what the smartest people in the world have read, and it's a whole list of books, and that was one that I hadn't read. So I'm like, I got to read that. Yeah. It was amazing. Someone gave away the ending, which kind of made it less surprising. I don't want to give anything away. No spoilers. And I also felt like it was very long. But now I'm reading The Power Broker, which is like set the new precedent for long. I've been reading it for half my life at this point.
I love The Power Broker. Another thing I highly recommend to people is if anyone follows Simon Haisell's Substack, where he does a slow read of important books. So he did one of War and Peace, and he's doing one of Wolf Hall I think, or he did one of Wolf Hall, which is a Hilary Mantel book, take it chapter by chapter. That's the only way to read something like The Power Broker or War and Peace or even Anna Karenina, chapter by chapter.
Speaking of that, someone, I forget who told me this, there's a 99% Invisible book club breakdown of The Power Broker where it's 13 episodes, an hour or two each, and they go through a couple chapters of the book one at a time and talk about it, and they have special guests like Pete Buttigieg and AOC and folks that lived in that area, and they talk about the story. And it's so fun to read and listen to their analysis of it, and then they have Robert Caro come on a couple times.
That's amazing.
Yeah. Hot tip. Okay, we'll keep going with our very lightning round. Favorite recent movie or TV show you've really enjoyed, if you've had time to watch anything.
Of course, I watched The Odyssey. I found it to be an incredible, incredible film. It is about AI, or my hot take is that it's about AI, or Christopher Nolan's view on how AI transforms society, which I loved, and I highly recommend watching The Odyssey. He's just an incredible director and has bridged artistry and commercial success in a way that I think no other modern director has done.
I also recently watched the film Rashomon, which is the Akira Kurosawa film that for the first time did that technique of telling a story through multiple people's perspectives where you never know what was true in the end. That technique in film was pioneered by Kurosawa, and it reminds me what one can do under constraints. That film was made in the 50s, it was black and white, there's a guy holding the camera, and yet it is so perfect and such a tasteful, innovative, amazing example of creativity. And what I'm reminded of watching that film is I have a hundred times the power and tools that he had making that film in my iPhone. And what's my excuse for not elevating my ambitions and making better stuff?
All comes back to ambition. On The Odyssey, I'm still trying to get tickets. It's so hard. I slept on it, and now it's impossible for like a month. There's no seats anywhere.
Kevin Clark got us tickets at 10 PM at the Metreon earlier this week. It was so good.
Next time, call me. I'm in. Oh man, I have bots running on it. I have a person working on it. I have a friend. We're all trying to find a...
It's amazing. You're going to love it, and I can't wait to hear what you think after you see it. If you agree with me that it is about AI and the collapse of morality.
Okay, no spoilers. Hopefully by the time this comes out, I have seen it. But if not, if anyone has hookups, please tell me. And I'm trying to do the IMAX full power Metreon sort of thing. Yeah. Okay. Next question. Favorite or interesting AI product that you've recently discovered, ideally not an OpenAI product, but you can also go there if you want.
Ooh. I mean, of course, my favorite AI product is ChatGPT and using cool sites and visualize stuff in Codex, which is amazing. But outside of OpenAI products, my favorite AI products are products that my friends make for me, because now actually people can do that. I think that's so cool. I'm such a huge fan of the cozy software movement, where you make software tools for five of your friends and you guys use it together. And so I have a friend named Sebastian who made a really cool AI app that turns anything into a podcast and puts it in a little podcast app for you. And he also made a really great private social network for our friends, and it's called GATS. It is exactly what I think the future should be, which is people should make software that exactly meets their and their friends' needs.
What does GATS stand for? Is that some inside joke?
It is not. Or at least if it is an inside joke, I don't know it. But it's like private Twitter, maybe, for a small group of friends. And I learn the most interesting things on that product.
It's like a WhatsApp, but not.
Yes. Exactly. Exactly.
The podcast app is interesting, but I feel like the version that I would love is it's an actual podcast in your feed of podcasts, and then new episodes get added of things you want to read or whatever.
Yeah, that's what it does. It drops it in your Apple Podcasts feed.
Amazing. I want this. It's great. Help me subscribe to this.
For sure.
Okay. Amazing. Okay. Two more questions. Do you have a favorite life motto that you find yourself coming back to often in work or in life?
My life motto that I come back to all the time in work is actually Toni Morrison's three takes on work. Let me pull it up really quickly.
Amazing.
Okay. It's four things. It's from her essay "The Work You Do, the Person You Are." The first one is: whatever the work is, do it well, not for the boss but for yourself. The second is: you make the job, it doesn't make you. The third is: your real life is with your family. And the fourth is: you are not the work you do, you are the person that you are.
I got tingles. Wow. So good. And I think that's what you have pinned to your Twitter profile. Yes, because I remember seeing that.
So cool. Okay. Maybe we'll show that on the screen as you're talking about that. I love that. That's a great way to remember something. Just stick it to the top of your Twitter, because every time I go to Twitter, oh, there it is again.
Mhm.
Okay. Final question. You were a Thiel Fellow back in the day. Teal fellow. Teal or Theel? Teal. Teal. Yeah. What an alumni group. Holy moly. It's such a great idea and program. Any story from that time that might be fun to share, something that's like, oh wow, that was crazy? I don't know, any other Thiel Fellow that you're proud of, what was the interview like, anything along those lines.
Yeah, the Thiel Fellowship was an inflection point in my life. I wouldn't be where I am without it. Maybe to the point of, there are key moments where you can tell people to elevate their ambitions, and they do, and that changes them. That was a moment where someone came to me and elevated my ambitions and said, no, you can do this, you don't have to take the path that you were on. And truly, I'm eternally grateful for them being able to do that.
One of the Thiel Fellows that I get to work with all the time now is Ari Weinstein, who founded a company called Sky that was acquired by OpenAI. And prior to this, he founded and worked at Apple for a while, because they acquired his previous company. Ari is just one of the most creative thinkers I've ever seen and is truly the expert on what are all the cool things you can do on a Mac. And so Ari leads a lot of our computer use stuff at OpenAI, and he's shipped a whole bunch of great things for computer use. But yeah, his creativity and his joy in what he does and his love of his craft really inspires me, and Ari's a cool guy. But I'm trying to think of a good story from that time that feels...
As you think about it, I'll explain the Thiel Fellowship for people that don't know this, and correct me if I'm wrong. Basically, Peter Thiel is like, "Hey, people shouldn't go to college. Instead, they should just try building something that they want," and you get $100,000 to not do college and instead just go follow your ambition. Is that roughly correct?
Yeah, that is exactly right. And you're with 19 other people. At the time, it was 20 people every year, because it's 20 under 20.
And
How many years did it go on for? Is it still...
I think it's still going, but I think it was constrained at the 20 number for like the first four or five years or something like that.
Yeah. I think a really crazy thing that happened my year is that I was the second ever year of the fellowship. They decided to make it all a documentary on CNBC. And so our whole... my pitch for the fellowship, getting up on stage and presenting the idea I was going to do, all of that is unfortunately live on YouTube. So if you really want to see me as a 19-year-old doing something embarrassing, it's there.
Of course, one of the most amazing and successful people who came out of that batch of the fellowship is Dylan Field, who is not only an incredible talent, but also a very kind person. And yeah, feel very lucky to be able to work with those folks.
Amazing. Yeah, it's interesting that Dylan's like the guy I think everyone thinks of when they think of Thiel Fellows.
Yeah. Yeah.
What a brand. Okay, Tara, this was incredible. Is there anything you want to plug? Anything you want to point people to? And how can listeners be useful to you?
Anything I want to plug and point people to? Maybe they should use the ChatGPT desktop app. They should use ChatGPT in the web and try Work. It's unfortunately a little toggle. They can toggle over to it and try out Work. Ask it to do some cool thing. Ask it to build a site about you, maybe to start, or ask it to make a little visualize block of your ChatGPT usage. It's a really cool way to start experiencing the power of this stuff very intimately. And the list of use cases they can do from that, you know, are infinite. And I'm happy...
Here's a better idea. Here's a better idea. Ask it to build a site to tell you what you could do with Work.
Great.
That will work.
Solve all the problems. Okay. I interrupted you. I apologize. What else were you going to add or say?
Yeah, my main plug is, yeah, go download the ChatGPT app. Go use it on web. 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 wandering around with your laptop open the entire time. You've finally got these things running in the cloud doing real work.
Yeah, that last piece I was going to bring up, but that's I think a really underappreciated element of the product today. On mobile, is that just a mobile-only feature? The cloud piece.
No, it's everywhere.
It's everywhere. Okay. So amazing. So on your mobile app, you can go to ChatGPT, toggle Work, ask it to do some work, and you don't need to actually have the... it's not running locally. It's running in the cloud. It'll go keep doing work until it's done, and then you could chat to it. So that feels really simple, but that's a massively powerful thing. Okay. Anything else, Tara, before we let you go?
No, that's it.
Okay. Thanks, this was awesome. Thank you so much for doing this.
Such a pleasure.
What a journey since the fellowship back in the day. I'll talk about that more in the intro.
Yeah.
All right. Well, thanks for being here.
Thank you.
Bye, everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.
Article published · Updated