Codex from 0 to 10M Users: Building ChatGPT Work

[gentle music] Okay. We're here in the studio with Akshay from OpenAI. Welcome. Thank you. And with our trusty co-host, Vibhu. Uh, we, uh... So you recently launched ChatGPT Work. You lead Core Product Engineering. You know, it's been a long journey, uh, into, into all this. I find it very interesting that you started with no code or low code, uh, with Walrus and Airtable, and it's, to some extent, ChatGPT Work is kind of like the super app of super apps of, well, here is the ultimate no code, you just write, uh, a prompt.

[chuckles] Yeah, yeah. It's, it's, it's funny how things come, like, full circle. I mean, I think for a long time in my career, I mean, I started my career working consumer fintech, but then after that, like, there's this hypothesis that, you know, the things that we were able to do with code, like, as engineers, like, if we could bring that to many more people in a more, uh, accessible way, then that would be truly magical.

We were working on a startup, it's actually funny, like, before LLMs, before vision LLMs, on how to do automated testing with AI. Um, and [chuckles] it was just kinda jank, um, back then, but, you know, doing what we can, and then, and then worked at Airtable for a while on, you know, the same thesis that, like, if we can bring a database or the primitives behind a database to people, that'd be really useful to them.

Um, but once I think LLMs came onto the scene, it became clear that, like, this was, like, the missing piece, like, the missing technology required to, like, bring the magic of code to everyone without them having to know what's going on underneath the hood. And so, like, I think this launch and, you know, a lot of the stuff that we've been up to is, like, um, the manifestation of that. How was stuff when you joined? So you joined OpenAI 2023.

Now we've got, you know, so much more stuff. So ChatGPT, Codex app, ChatGPT for Work. Have things changed? Actually, I think the more interesting thing is how things haven't changed. Like, I guess, like, one, I, I joined-- I remember when I joined, it was, like, 500 people. One thing I was worried about was, like, I was looking for something, you know, more early stage and, like, was it- [laughs]... gonna feel startup enough?

And I joined and I was like, "This feels even more startup-y than I could ever imagine." And, like, that, that really hasn't changed even till now. I mean, I think the, like, level of, like, bottoms-up ambition and, like, the ability of anyone to, like, you know, do anything or have an idea and, and ship it is, is really cool.

But on the, like, sort of mission side, I think what was really compelling to me is this mission of, you know, bringing frontier intelligence to everyone. Like, building AGI and then bringing it to everyone. And, um, I think acknowledging back then that, like, that vision is gonna, you know, not be a linear progression.

Like, we're probably gonna, like, try different products and, and have different things that, that succeed and don't. Um, but the vision has stayed the same, and the mission has stayed the same, and we're starting to see the pieces, um, fall together and, and that, that's really cool. Uh, you worked on Enterprise.

What-- A lot of people never touch Cha-ChatGPT, ChatGPT for Enterprise, God. Um, what is something that you learned from there that you're bringing into your work now? I think how there's no, like, one-size-fits-all solution in Enterprise. Um, I remember in the early days of ChatGPT Enterprise, like, when we talked to customers and, like, everyone...

That was, like, when-- I think it was a year after ChatGPT was released, and everyone was so excited to bring, um, you know, AI into their enterprise. And, like, like, there was all these teams stood up, that were being stood up. It was, like, you know, the AI deployment team with, like, these enormous budgets.

And if you asked anyone, like, what were they excited about? Like, what were they excited about solving? Like, at first, you'd get, like, you know, kinda like the, the baseline answers of, like, "Yeah, we have all this context and data and all this stuff." But then if you ask them, like, you know, "What was, like, a discrete use case that, like, they want AI to enable in their, in their workplace?"

You get such a different, like, variance, like, explosion of, uh, different types of answers. And it's interesting, like, you know, you-- using, like, these models and these products, you, you have this box, and you can say anything to it, which is the magic. But on the flip side, it also means that, like, you don't know what to do with it.

And in Enterprise, I think a big part of that is, like, actually meeting the users where they are, like, what use case were they trying to solve, and then actually teaching them how they can use AI to, like, gain leverage there. Do you meaningfully differentiate that from forward deployed engineering or...? I, I think there's, like, the, the, like, go-to-market side of it- Yeah... and then there's, like, the product side of it.

I think- With Neo- You need someone on the product side. Right. Yeah. Um, and I think, like, however good we get at FDE motion, like, I think at the end of the day, if we have a user who's, like, looking at their computer or looking at their phone, like, it's our job in the product to, like, be enabling them and showing them where to go.

Um, so we're really excited about that. Do you think there's been changes, you know, over the past three years of adoption? So there have been, you know, step function changes. You have reasoning models and whatnot. Is there still the same problems of Enterprise has black box, don't know what to do with it, or have things changed? I mean, we're seeing now that, like, there's this huge uptake, right?

Everyone's ex-ex-extremely excited about it. It feels like, you know, many peop-- like, millions, hundreds of millions of people are using ChatGPT. They understand, like, how generally to work with AI. But then, like, every time, like, a new capability gets unlocked, so now, like, we're seeing with agents, like, there is probably a contingent of, like, early adopters still who, you know, truly get it, who are like, "You know, you can do anything.

You just have to make sure the right context is there, it's connected to the right tools, and then that you're supervising it, but, like, anything is, is, is possible." But then there's, like, this, like, ten X or a hundred X bigger market or, like, they don't yet get that, or they don't yet see that. Um, and so I think that's the next stage here.

Um, so I guess to answer your question, like, I think the adoption is there and, and growing fast, but I think the opportunity is, like, far, far bigger than that. That's where we wanna play, especially with ChatGPT Work. Yeah. Uh, well, let's, let's, uh, skip ahead to ChatGPT Work. Uh, on-only, like, a month ago or so, uh, announced.

Uh, what was the sort of decision process that led into it? Um, you know, there was this, uh, overall merging of the super app. Is, is that what we're officially calling it? Um, you deprecated the browser as well. Just, I guess, summarize, uh, your last, like, couple months of working on this thing. Yeah. It feels like forever now, but I guess it's only been a few months.

I think maybe the one, one, like, impetus that, like- Is most salient is when we re-release Codex or e-even internally add Codex, like, it was really surprising to us, um, I think we recently put out some stats on this, that there was this, like, real inflection of, like, adoption among non-developers at OpenAI.

And, um, I, you know, through this product development process, like, would go to, like, these UXR sessions to talk to people internally. And the thing that stuck out to me is, like, one, like, you know, you, you go talk to, like, strategic finance or marketing or whatever, and they're all using Codex for, you know, their use cases.

That, that part, that part's cool, but the thing that really stuck out to me is how proud people were that they were using Codex. Like, how, like, not- It's like, "I'm not supposed to be using it, but I am." It, it, it was that. It was, like, that they were, you know, early to this, like, new thing, but it was also this thing of, like, they felt like they had a superpower, right?

And, um, what we recognized then is that, like, the, the power of Codex, the power of agents, like, we already had this massive distribution base of people who have, you know, come to know and love ChatGPT. Like, how do we show that to them? Like, how do we bring it to them? Which is, like, a hard product problem, and it's, like, a tricky, tricky thing, right? There's many ways you can go about it.

Um, and so that's what we called the Merge and the Super App over time, and, and, and ultimately launched it in ChatGPT Work, is how do we do that? But it, it came from that initial realization that, like, the, the power was not only for developers, like, much, much earlier than probably even we thought. Like, it could be extended to, to everyone.

How do you see the products differently? So, like, who is it for, right? So Codex started out even CLI, then app. Now there's a merge of ChatGPT, Codex and ChatGPT Work. So is it the opening for the average user, for enterprise, for work? How, how do you position it? I think we want to get it to position it for if you're doing worky related things, for lack of a better word, right?

[laughs] I think productivity is like, uh, is, it's actually what, like, the pillar that I, that, that I support. Like, that's the name of the team. Um, and the reason for that, the reason we call it productivity and not, like, you know, enterprise or, or, like, work or something like that, is because there's also personal productivity, right?

And, like, I think ChatGPT Work is-- I've seen people do things in their personal lives that you wouldn't classify as, like, work technically, but, like, these agents are, you know, super capable for. Like, one, one recent example that someone posted about, um, on our Slack is, like, someone had, like, a missed package, like, they didn't receive it, and then they got, like, the picture of it, um, you know, from Amazon or whoever the courier was, and they, like, asked ChatGPT Work to, like, find out where that package is.

And, like, the agent, you know, is extremely tenacious and, like, like, took the image and, like, looked at a bunch of, like, listings around their neighborhood and figured out exactly the apartment complex in which the package was, like, gave them some information. And so, like, I think there's all these things that, like, you, you know, worky or productivity-related things, I think that's what we want the product to be. You asked about Codex.

I think we think Codex is, you know, a durable brand, but we have a principle that, like, the user-- you know, we don't want a user to get stuck in a tab or an experience where they don't get the power of the product. And so, like, basically everything that you can do, you know, in the Codex portion of the product on, on desktop, you can do in ChatGPT Work and vice versa.

But we made some opinionated product decisions on, like, you know, how much of the Git state, if you're in a Git repo, do we wanna expose to the end user? Or how much do we wanna make the, the experience of seeing the agents thinking, like, diff forward so that you're g- you get exposed to the diffs out of the box.

And then, like, on the safety side, like, how do we wanna think about, like, sandboxing and making sure that we have the right defaults in one, one state versus the other? So, um, there's, like, some opinions that go behind that, but we do, we do want-- We don't want the user to need to choose which experience they're in. That is a good goal for AGI, right?

Like, people don't want, like, to hide-- to choose what version of AGI they want. They just want the AGI to decide- Yeah... for them. Um, uh, can I get an answer or, like, uh, it's not super clear to me. Is the Codex harness and the ChatGPT Work harness the same? Is it just UI affordances, or are there actually prompt level or even w- even deeper differences? So the harness is the same. The harness is shared.

Um, on-- In both of the products, we made improvements to the harness to make it good for knowledge work, especially as it relates to plug-ins or computer use or artifacts. You get that power regardless of which experience you're in. On the UX side, there's opinionated takes that we have when you're in Codex mode, what the UX should be-- how the UX should behave, and some stuff around the sandbox like I mentioned, but the underlying harness and capabilities should be the same.

Actually, I'm just kinda curious. Maybe we can, uh... Is there a query that we can run that would look different in the, the two modes? Yeah. Like, try to create, like, ask it to create, like, a retirement calculator spreadsheet or something, um, in, in both, in both modes. And then in Codex mode, um, you might have to be in a, in a repo for this, but you'll see, like, the diffs of, like, the, the sheet that it's creating and stuff like that, um, and the file edits.

Um, but in Work, you won't be able to see that. I think that's, that's super clear. And then also the other thing I wanted to dive into was your, uh, the productivity team. Uh, what else is there? Uh, first of all, you know, what, uh, what, what are the top-level teams other than productivity? Isn't productivity everything? So, you know- Science?... we have a team focused on, uh, on ChatGPT- Yeah...

like the, the core chat experience, um, for consumer, which is like, you know, not, I, I think all productivity. Like, there's-- People are using ChatGPT every day for search to, you know, figure out how to write messages to loved ones, to think about, um, how to, like, learn a new topic, et cetera. And so there's so much more inside to create images.

And there's so much more in chat that, you know, the hundreds of millions of users are using that, um, you know, obviously that, that warrants, like, a, a very dedicated effort. Um, and there's teams focused on enterprise and infrastructure and API and stuff like that, so. I will bring it up. Yeah. So I have them both running. Yeah. This is Work. There's a Codex version here.

I picked [chuckles] "Five Little Things Soul," so this will take a while. Uh-huh. Um, I think, I think we'll just keep it in the background and, you know, as, as they finish, we'll look into some of the differences. Yeah. But immediately, I think if you flip back to the Codex version- Yeah... you'll see that, uh- That it assumes, it assumes Git. Exactly. Yeah. Yeah. The, like, dynamic island assumes that you're in a Git repo. Um- And you might miss some stuff because some of it is, like, in the actual chain of thought, but with those changes and how we display that, but yeah.

Is, is there an unintuitive-- like, is there a thing that you wanted to ship and then you got feedback, and you were like, "No, let's not do it?" Like, what's the thinking behind that? In, uh, ChatGPT work? Yeah. I think one direction we could have gone with this is, like, keeping the experiences, like, completely separate. So it's like, why- Different apps.

Exactly, like different apps or even in the same app, like different, completely different experiences. Like, why merge it all? Like, what is... You know, Codex obviously people love. Like, why, why bring these products together? And I think the intuition here is that, like, all of our jobs are, like, changing dramatically with AI.

Like for, like every few months, like I feel like I wake up, and I'm like doing a completely different thing than I was doing a few months ago. And my, my hypothesis here is that, or I should say our hypothesis is that, like, part of what we're, we're building, and this, this technology is giving people leverage.

Like, you know, the things, maybe it's the more mundane parts of your job or, or parts that like, if you were able to automate, you'd be able to share more ideas faster or whatever, like you're able to do now. And because of that, like that might actually blur the lines between someone who's like only writing code or creating strategy docs or, you know, planning events or, um, helping with marketing or doing podcasts or whatever, right?

And so, like these things are gonna get blurred over time. And so, like trying to draw a h-hard boundary based on like the who you are is, is, is gonna be, is gonna be tough. And like, we should enable users to choose, but we shouldn't box them in. And so a lot of w-the work that went in here, like, you know, keeping the primitives the same, like for example, plugins are like unified across, um, this product and ChatGPT in the cloud, was because of that.

It's this, this thesis that like, eventually things are gonna come together, and, and we don't wanna be-- like we wanna be prescriptive about when to be in either experience, but we don't want to box anyone in. I wonder if there's users who are very tuned to the old ChatGPT harness that is effectively now replaced by the, the Codex harness. I can't imagine what that was, but maybe they're more the more con-conversational side.

Can you com-compare and contrast the, the two harnesses? 'Cause only you've seen it. [laughs] Yeah. I, I mean, I think ChatGPT, the, the existing harness like still exists today, like exists in this app, um- The classic, right? The cla- uh- You just start a new chat, and you don't go under Work, right? Yeah.

If you start- So-... a new chat and go to chat, then- Yeah... you're, you're talking to ChatGPT with the instant model. Oh, we can technically do another. Yeah. But I guess, you know, on instance. Yeah. So this one's not gonna code or it's gonna- Oh... be in line. It's on a s- in line in a sandbox. It'll actually- Oh, that's cool... it will try to push you to, to, to go to Work if you're creating a spreadsheet. Yeah.

And this, this is a router decision? Sorry? It's a router decision? This is the decision that, you know, the model is making, and then like- Right... you know, it sees that you're able to... or you're trying to do something that would be better served in Work mode. But I, I think your question was like, what, what are the advantages of like the, the chat, like ChatGPT chat harness?

Uh, it's more broadly like I wanna, uh, basically do an oral history of harness engineering. Mm. Right? Uh, you know, the ChatGPT harness lasted us from, let's call it the '01 era, uh, until now, and now it's being replaced by the Codex harness- Mm-hmm... effectively. And they're, they're overlapping somewhat, but I'm curious what changed if, you know, if there is.

[laughs] My perspective on this is like there's, there's, there's al- there's sort of like a, a constant process of like divergence, convergence, divergence, convergence. And in Chat, like many of the use cases I was talking about before, like, you know, search or learning, I think we're, we're really optimizing for latency and optimizing for personality and like different things that, um, over time, like the product-- the reason people love ChatGPT is because we've been optimizing for those things and working on them for so long.

Codex, what we learned was that like if you give the agent access to this infinitely flexible environment as a computer, um, it can do really, really powerful things. And so when we think about like, okay, well, for, for knowledge work, like what is... which mode should we choose? It was like it felt more natural to us to bring that to this like computer environment and, you know, maybe abstract some of the details of this computer away from users who might not be used to that, but like give them that same power.

Um, but ultimately, I think that we want the power in, in all places, right? We wanna meet people where they are. So I'm sure there'll be work down the road in order to, to get things to be, um, equivalently capable in, in all scenarios. Um, but it's just a question of like what we've been focusing on the product on historically and what we're focusing on now.

I think alongside that, outside of just harness and when to use Codex, ChatGPT, or Work, there's also the new models you've released, right? Um, any guidance there? So people love to min-max what to use, like only use Terra on high reasoning versus, uh, for this, you know, you wanna use Sol here. Ignore all these- There's 32 options. Yeah, yeah, yeah.

Um- [laughs] But, um, that being said, you know, for people that are expanding, so, uh, productivity stu- trying stuff for work that don't have the breakdown of what all this is- Mm-hmm... um, what's, what's the advice, right? Well, I mean, I think before the advice, like the first thing is like none of this would be possible without these models.

Like I think you asked earlier like, you know, what was like the inspiration for Work and like, you know, early on, like I, I mentioned like what we were seeing with Codex, but that was also because the, the models were getting infinitely more capable. That's happening again. I think it's like another step function jump now. And to answer the question on advice, like we want this default to be the best possible.

Like we wanna be opinionated about the default, and so we've ch- we've chosen a default that we think is gonna be the best for everyone. And, you know, we have for power users options under the hood. We could-- one could argue that there might be too many right now and we're, you know, working on simplifying it.

Um, but you can extend, you know, the reasoning level, and you can change between the different model classes if you need to, but the default should be the best for, for most, most use cases. So my advice to most people would be to stick to that. And then, you know, if you reach a situation in which you think that you could, you wanna try, um, a different configuration, if you're not seeing either the, the efficiency on the, on the cost side or, or the, the quality on the intelligence side, then you can change the defaults and see if you can get something better.

But, but we think that the default should be good enough. I have, uh, I-I'm just gonna run something by you since you, you have way more experience than me. I've recently been doing Sol Lite, but with goal, with the idea that the goal basically augments the reasoning effort, but with more terminations and turns. Mm-hmm.

Is that a good way to think about it as opposed to Sol Ultra or Sol, you know, Extra High? Yeah. It's hard to say because- Yeah. It's like an interaction effect. Ex-exactly. It's like there's a preference on, you know, for you as an individual, like how do you like to collaborate with the models? Like how many of those like terminations, as you call them, do you want where, you know, you can in, uh, steer or make sure that it's doing the right thing? I think generally people should try whatever works for them.

Um, I think that like using Ultra or the like multi-agent setups are best for like when you have like tasks that are either incredibly complicated, like open, open explorations or very parallelizable. I think even for tasks-- Using Goal, I think is best for, for tasks that you know that you'll be able to make consistent progress in a way that's verifiable over time.

But I think for most tasks, they actually don't fall into either of those buckets. [chuckles] Um, and so like at least when they're starting, and so that's why I think the best first step is like trying it with the, the default configuration and then seeing like where you wanna go from there. Right. You guys worked on a slider, which is actually, actually is super helpful for reducing the amount of panic. [chuckles] Yeah. Yeah. It's nice on mobile at least. There's a nice slider there. It's nicer.

I haven't tried it. So you, you, you have, you have the advanced view there, but if you click advanced view... Yeah. Yeah. Ooh. Just a nice slider. Yeah. Very, very pretty, very colorful. Yeah. The idea was here, here was to like reduce it to like one dimension even though there's multiple dimensions, right? Try to project it onto a single dimension for the user. Yeah. Of like, you know, something from...

that represents like, you know, speed and efficiency on one side- Yeah... and then like sort of like quality and thoroughness on the o-other side. Um- I, I am just puzzled that it uses Sol so much, like the lower- No, no... I would've used-... I think this slider, if I'm not mistaken, is- Terra. Oh, it is. Yeah. See? Nice. Nice.

So they, they preset Terra to only be the light one. I see. But like I think a lot of people actually would... more people should use Terra. One, because Sol keeps running out of capacity. [laughing] I'm the reason. You know, here's ten minutes of our- There you go... retirement calculator. Oh, that's the Excel thing we're displaying. This is, uh- Oh my God, look at that...

this is work, and then Codex is still cooking, so we'll get back into it. I think it, it'll be interesting to actually see the thought process, the reasoning, and also, you know, I guess this is eight minutes on work. Codex is still cooking. Yeah. And by the way, uh, I... So I've, uh... Do you know Gabriel Chua? He's part of the OpenAI Singapore team. He showed me this, and I was like pretty shocked that this looks like Excel. Yeah.

Uh, it, it edits Excel files. You never paid an Excel license. [chuckles] Right? Like, but somehow this, this is like kind of workable and it's agentic Excel. Yeah. I mean, one of the big like pushes that we made for this launch was like artifacts, right? Mm-hmm. Like both on the, on the model side, like I think if you compare this with, with 5.5 and, and, and 5.4 before that, you'll see that the-there's been pretty dramatic improvements in the quality of these artifacts and then also on the product side.

The UX side is also crazy, like hosted sites and whatnot. No, no longer needing to host your own little webpage. Like it- Oh, I have a story about that. Yeah. I can, I can do, uh, a separate thing. I, I'll need to take the, the, the visuals here, but we'll, we'll, we'll cut to that later. Was there co-training, I guess, uh, because you are moving-- making this big move, and you launched 5.6 on the same day as ChatGPT Work, was there influence between the model training teams and the harness teams, or did they-- did the launch dates just happen to line up the same day?

Uh, I think the... You know, we collaborate heavily with the research teams, and I mean, I think that's like one of the most magical parts of the job, like the, the most fun parts of the job. Um, but yeah, I mean, just u- just using artifacts as an example, like, you know, a lot of what you're seeing, like underneath the hood, there's a lot of work that went into making sure that like, you know, we had the right infra to be able to train the models to get better at this, and then on the product side, like had the right experience for users to be able to collaborate with the model on an artifact like this.

In fact, like this whole viewer, like the intuition here is that like, you know, it's not necessarily that you wouldn't need an Excel license. This is stage one, right? Like this is probably not what you meant when you're like making a retirement calculator. Yeah, you can iterate very easily. Yeah. You wanna iterate and like when you s- when you're seeing it, and if this thing is high fidelity to like what you would actually see in, in or what your, your coworkers would see if you were to send this to Sean, like that, that I think makes it so easier and makes you trust the, the, the product in, in terms of iteration.

When you say coworkers would see, do you see a multiplayer, multi-team collaboration with artifacts? Any, any things you guys think about that? You can already share it, right? Yeah. Yeah. It's inter-- It's something that, you know, we're actively thinking about. Um, one thing that, you know, we've noticed internally without talking too much about the roadmap is that like I...

there's many times when someone will ping me about something, and I will ask ChatGPT Work the question, and then I'll ping them back the answer. [chuckles] And then I'll be thinking like- Like the cyclist would be, you know, the three of us are just all on one hosted. Exactly. But I'll think about like, you know, like was I required in this loop or...

And then maybe it was, you know, rephrase like what they were asking or pulled from certain context or whatever, but like, you know, when I gave them back the answer, that process was also lossy, right? Like I gave them just like my interpretation of what ChatGPT Work cooked up. But like underneath the hood, there's so much context like in the rollout and stuff that, that could be interesting.

So like the answer was preemptively respond to every inbound request? No. It was just like literally like this is what I do sometimes as my job. I know you copy-paste- Yeah... and then you, you know, you're just a message forwarding service- Yeah. Yeah, exactly... from AI to AI. But I think it's interesting, right?

It helps people understand the capability of what you can ask and delegate that- Yeah... oftentimes people don't realize until they try or someone shows you, and then you're like, "Oh, okay. Okay, I see." Yeah. I think it's als- there's also like a, um, light security issue, uh, where like basically you're the permissions layer.

Like yes, I could query everything that you query, and I could get an automated response, but maybe I'm not supposed to see it. Yeah. And that- [chuckles] Yeah... there's no way I would know because I'm not supposed to know what I don't know. Especially as like, you know, with ChatGPT Work, we're, we're asking you to connect your plug-ins and, you know, it's pulling from your local files and stuff like that.

Like the, the amount of context that the agent has access to is like- Deeply personal and, like, that's something that, like, we need to preserve. So that'll be definitely a challenge. There's Excel, there's PowerPoint, there's Docs, you know, the, uh, grand trio of work. What other formats of work do you, do you think about? You know, like, obviously you worked on Airtable.

Is there a future where there's, like, open AI Airtable? Like, you know, like, what, what, what does that look like if, if you ever ended up doing it? It's a really good question. I think, um, I mean, o-one that you didn't bring up was Sites, and I think that- Sites... was a core part of this launch.

Um, there's one, one side of Sites that I think people commonly talk about, especially on Twitter and stuff or X, of like, you know, this sort of like prototyping tool. And actually, like, we saw that happen with this launch. Even the, the model slider that you guys were referencing earlier, like, that was developed almost fully in a Site. Like- Mm...

you know, the, the collaboration between design and engineering and product on that was, like, on a site where we play with, you know, the, the, the affordance and, and, and figure out how it feels and, and all of that. Um, but the other aspect that, that I think is a little bit less talked about is, like, Sites as, like, an, an artifact for, for knowledge work.

I was actually talking to someone the other day who was on, like, our, our corporate finance team, and, like, they were mentioning how, like, now, when they have these reports that they're, they're working on as a team month to month, historically, those things were in, in slide decks and in spreadsheets, and now they're just in Sites.

And, like, Sites is the mechanism that they collaborate across the team. And the reason is 'cause it's like, it's like somewhat higher bandwidth. Like, you know, at so- these tools like PowerPoint and Excel are, like, infinitely flexible, but at some point you reach the boundary of, like, either as a human you may not know how to use some feature or something, or the product itself doesn't support it.

But with a Site, you can kinda do anything. You ask for anything, and you can get that. Um, once people see that magic, I think it's been really valuable. Yeah, uh, let me show you my, my case study. Uh, this involves all the hot topics, including ChatGPT work, but also 5.6, token billionaires and token maxing and Sites and auto research. I'm a fan of this game called Strata.

It's, uh, basically it's like a little board game that you- Mm... that you play with, uh, physical blocks. Uh, they come on top of it like that. Uh, so over the weekend, I, I, I took, like, thirty photos and just threw into ChatGPT. Uh, one point seven billion tokens later, out comes this site with a fully playable thing- Wow...

with, uh, with, uh, with 3D, uh, block placement and everything. Because it requires physical blocks and I needed friends to train on it so they can get better, so I can play against them. But also, I could also, uh, do things like train an AI on it, and that, that's, uh, that, that gets- That's your auto research... that gets into auto research.

So, uh, you want to train your own AIs, uh, and then make sure they self-play against, uh, each other. Uh, I need to set both AIs. Um, so this is AI versus AI, uh, and they're, they're gonna self-play. Obviously, the, the AIs start out bad, and then you want to define a loss function and, and get good. I wasn't gonna supervise all this. I was at, uh, I was down in San Mateo, uh, attending a conference.

What I ended up doing was, uh, auto researching and, and on this and creating benchmarks, and that there was just way too many parameters for me to, to read. Uh, so I started asking it for a Site, um, and it's created this, uh, uh, this, uh, this, this lab, uh, panel. Where... Is there, is there a, is there a shortcut for a Site that it's created?

Uh, you should be able to go in the sidebar to Sites, top of the sidebar. Uh- The left sidebar. This one? Oh, left? Yeah, just scroll all the way to the top. Oh. Oh, it says Sites. Yeah. Oh, there you go. Yeah. Ooh. So it create, it creates the Sites. Um, I don't, I don't think this is, uh, it is exactly what I, what I wanted, um, but let me, let me show you what it popped up, right?

Like, I think as a, as a research artifact, um, it is very important to communicate, uh, exactly, uh, what is being done. Outputs this, uh, this thing which I eventually started publishing. So I, I moved it off of Sites because I wanted more, uh, database and infrastructure than Sites afforded me. Uh, but this is, this is, like, a research output that you can start to mess with and, like, try to think about, like, what hyperparameters are you tuning for training AIs.

Uh, and, like, I was trying to make, like, scaling laws and everything and doing all sorts of, like, game optimization stuff. Uh, and the fact that you can just kind of throw this up as a research artifact, like, I no longer need to read ChatGPT output. I read Site output. But then there's also a huge sprawl.

Like, look at how long this thing is. There's so many numbers. Um, it is pretty overwhelming. So then I have to start pruning it from there. But, uh, it's an interesting transition from Markdown- Yeah... effectively that you're putting out to, uh, you're putting out a whole functional site. I think Markdown just isn't that optimal for people to read, right?

Might as well just write HTML website and, I don't know, I think you can do a lot with customizing this, right? You have your skills that explain what you want. Like, I notice they're quite verbose. I don't need a lot of this information. It's very verbose. So... And then the nice thing of having a Site side by side is, you know, you just iterate on what you want and what you don't, right? Yeah.

I, I don't know if, uh, any... that, that triggers any stories for you of how it's run internally. Am I doing this right? Yeah, I mean, I think that this is, like, a workflow that we're seeing, like, all different types of teams use, where, like, the, the canonical artifact that was previously a deck or something is now becoming a Site.

And, like, with a Site, you-- because it's just HTML, you can, like... It's infinitely flexible and so, you know, if you, if you want to give more prominence to a certain thing that, like, in a slide deck would, you know, feel like it was buried, like, you can do that. You can have it be, like, the hero image, right? And so I think that, like, um, people are starting to see that.

Um, there's obviously more work to be done to make these things, like, much more easier, easy to collaborate on. Um, you mentioned that they're very, they're long and verbose, could be broken up. I'm sure that there's still something to do there. They're super long. Yeah. Um, yeah, but I, I think we're starting to see that, like, there, there is this aspect of this is a really interesting, uh, format, um, for, for people to use, um, that's, like, much more flexible than, than what they ever had before.

I think your job also comes-- becomes kind of meta. You're not designing the products. You're designing a product f- to make products, and I'm curious how you manage that. I, I think one thing that we've been Like when we look at the UX, like that, that we've been thinking a lot about is how can we balance like simplicity with capability?

Like if we, if we're designing a product, like you said, that like is, is made to make up-- build other things, right? You can build so many different things. Um, but we can't put that all in front of you because you'll get overwhelmed. Yes. And so we had similar problem or, or similar challenges even Chat-with ChatGPT, but especially now, like when there's so much that can be done, I think the balance that we're constantly trying to strike is like, how can we give the user enough of a UI surface where, you know, they can be expressive, they can, they can tell the, the, the agent what they need.

They can verify that it's using the right tools, it's pulling from the right sources, et cetera, but then it gets out of the way. And then how can we build the right system such that we can show them instead of telling them what can be done? Because so much of this is gonna be like, how do they discover the next use case and the next one after that if they really want to, you know, to, to be super powered by the AI. Yeah. It's interesting.

I feel like everyone also just has a different way to do it, right? I made a similar version of this same game. I didn't take any pictures of board or rule game. I threw in at goal eighteen minutes, fifty-three seconds later, a lot of tokens later, I've got a similar version. Obviously, not with all the auto research and whatnot, but you know, um- You gotta do all the latest trends. And yeah, [chuckles] I did it with, uh, did it with Codex, not work, but it's interesting, right? Yeah.

And this is obviously GPT image generating the [chuckles] pro-avatars. Very good for game design. Like- And-... a lot of game designers were like really into GPT image for assets. I, I will say like the, the broader takeaway probably is the reason that we do this is more so just to test the tools, right? Like, um, this was also a test for 5.6 came out. I had done the game on 5.5, right?

The ability for me to no longer need it to-- I had to feed it the rules. It's, it's a pretty niche game. It couldn't find how to do this on its own. Oh, yeah. 5.6- It is auto distribution. This is why also very keen- Mm. Yeah... on testing the 5.6 capability. But you know, this is just as, as work comes out, as new things come out, these are just our side ways to test things, right? Yeah. I-- It's some kind of private eval, I guess.

Yeah. That is not this private [chuckles]. But also valuable because now you can send this to your friends and I mean, I learned about this game through seeing this. [laughing] It's a hard game. He's very good. [laughing] Uh, uh, g- It's good to when no one is competing with you. [chuckles] Uh, but yes, it's a classic RL problem of like self-playing, bootstrapping your game AI.

Um, yeah, you see how easily work becomes personal and personal becomes work because- Mm-hmm... the thing I do for personal, it actually directly informs people I work with because I showed it to them. They were like, "Oh, you can do that with [chuckles] GPT?" Yeah. Which like I, I imagine is the growth strategy. Yeah.

The, the show not tell is a big piece that, you know, I think we've-- we're not still not fully cracked of like, you know, showing people all the, the things that they can do with the product versus like trying to teach that to them through like, you know, articles or onboarding or whatever. Yeah. So meeting them in the moment. It's a career risk for me, uh, because I used to be in developer relations, right? Where your job is to show, and then you're like, "What do you mean?

You, you don't, you don't need..." Uh, actually, your job is to tell. Mm-hmm. And then, and then... But the product people are like, "Well, we don't need you if our product is intuitive enough." So [laughing]. Yeah. I mean, that's the magic of the, the models. So you can tailor the telling or the showing to like specifically what the user needs, like what, what they care about, what they've done in the past, exactly where they are on the adop-adoption journey.

So I think that's like gonna be a super big opportunity. Seems easier and easier now to tailor custom showing, right? People have different use cases. As much as you said you don't wanna segment different people into different buckets, right? It's also not that hard to for people that are in different categories.

But the question, I guess, is you said your team is more broadly on... What, what was the term you used? Productivity? Productivity. Yeah. Productivity. So how- Which is now work, basically. Is it work? Is there another distribution that we're not hitting? Is there a group of people that will have something different than ChatGPT, Codex or work? Is there, is there more that the mass isn't targeting?

I see it as like a sequencing, like, you know, the, the vision is like bring useful agents to everyone. We started with like developers. Like developers historically are like early adopters that are willing to put up with more friction, set things up, et cetera. Like that's where, you know, Codex started.

I think the next opportunity is like sort of what we call general knowledge work, you know, all the other functions around developers. I think when, when you go from developers to this segment, like there's inherent challenges obviously with like, you know, this, this show not tell thing that we're talking about, um, making the product more understandable, um, bringing in new capabilities that matter more for, for this cohort than matter for developers, things like artifacts, things like computer use, et cetera.

And then I think like the, the same learnings, like similarly how we took the learnings from developers and brought it to, you know, general knowledge work, the next stage will be like taking the learnings from general knowledge work and bringing it to everyone no matter what they're doing in their lives.

Um, and we're already seeing that a little bit. Like this, this game example that you have is, you know, something that's like on the border of like fun and personal life to, to, you know, your, your professional life. I use ChatGPT work full-time at home for everything, like for, for whatever I'm doing.

I used it the other day to come up with a meal plan and like, you know, save that on, um, on the like computer environment that it has and something that I can continue going back to. Like is everyone doing that yet? Probably not because the thing says work on it, but eventually, you know, we wanna get people there. Um- ChatGPT life. Yeah, exactly. ChatGPT cooking. Um,- [chuckles]...

but I think there's a lot of, uh, there's a lot of opportunity there, but I, I see it as like, you know, we're, we're built-- we built the foundation in, in, in, in software engineering, and we're gonna take the same learnings that we take from software engineering to knowledge work, knowledge work to, to everyone.

Do you have any power user advice? I feel like, um, there's a group of people that will live it, use it for everything, stay on it twenty-four/seven. Yeah. Um, and then there's a bit of a gap between that crew and people that, you know, okay, I use it for work. I use it occasionally. Sometimes I type questions.

Uh, any advice, any learnings, anything you recommend or just, you know, takeaways that you found that help bridge that gap? I think a couple things that I've seen is like, one, that it really helps to broaden your imagination of what's possible. And this has been a learning even for me. Like, you know, the technology has progressed so fast that, you know, something that, like, even three months ago, like, no way, no way that models can do this. Like, now it's like, wow. It's like it actually can.

Like, um- Give an example... we're going through right now our, our, like, review cycle internally, and, you know, people always talked about this as, like, kind of a, a thing that the, the models are good at. And like, you know, there's a cliché of like, okay, like, uh, no one wants to be writing reviews and, like, we can just use AI to do it. But I mean- And it also evaluate it as well. [chuckles] Yeah, exactly.

In all seriousness, before it was like, just like slop basically. And like, I, I think it was helpful, but, you know, not super productive. Now I've found out, like, the, the model can do a much, much better job than me, especially in this environment of, like, pulling context on, like, what people are up to, how they've like-- the things that they've done to make a difference, highlighting like, you know, wins that they've had that, like, I might n- may not even have seen.

You know, it has access to, like, everything, right? Like, the code, like, you know, things that they've caught, uh, reviews, Slack, everything. And so it's, like, incredibly powerful in that domain. And, like, just like six months ago, the last time we did this cycle, like, I didn't even-- I tried using it, but it was not at all helpful. And this time it's been, like, incredibly helpful.

And like, so I think continuing to push the, the frontier of imagination of what's possible, even if you tried something before, I think is maybe the-- my biggest piece of advice. The other, I guess, thing is, like, the more, the more you put in, especially in this environment where, like, you know, the model has access to, to everything on your, on your computer or in ChatGPT work, like, you can create, you know, artifacts over time and save them in your library and, like, the model will continue having access to those.

Like, the more information you give it about whatever domain you're in, whether it's your life or your work, the more valuable it becomes. And it'll become both valuable in, like, ways that might surprise you. Like, it might pull from context in a way that, you know, may be proactive and that you might not even have thought about. But it needs to have access to those, um, to that-- those tools or that context first.

One thing I just wanna talk about the review stuff because I'm still-- that's a very sensitive thing and you're, you're a founder, you've managed people, you've hired people. As manager myself, I'm very reticent to put out any LLM-generated things wh-- especially when it comes to people, 'cause it feels like you don't care. Mm-hmm.

Presumably at OpenAI, people are obviously more open to being eval-- basically [chuckles] rated by GPT. Uh, but are there any unofficial rules around this? Like, what's the etiquette? Oh, I mean, I think the etiquette is that, like, I would never write something via, like, solely via AI and, like, present it as, like, a review for someone. What I was talking about is more, like, gathering context.

Yeah. That's the place where it's- So it's just search... incredibly helpful. It's agentic search. Yeah, exactly. Yeah. Like agentic search, but you know, that you can tailor and steer much more capably than you could before. And 'cause, like, the, the thing is it- it's all-- there's sort of a flywheel happening, right?

Because of Codex, people are able to do, and because of ChatGPT, more people are able to do so much more now than ever before. And if you're able to do so much more, it's easy to miss things as well. Um, and so, like, I think we need to use these same tools to keep up with all- Yeah... the impact that people are having and, and, and understand, you know, where, where we can be helpful.

I think the, the, the thing, like, obviously I, I run a small company, so easy to search, but at the scale of OpenAI, with the amount of messages that you guys put in Slack, do you think that it misses things? Probably, but I think that I also miss things. Like, it doesn't matter, right? I think sometimes it's- Like it's, as it needs to be human level... it's all relative, right? Yeah.

Sometimes it's nice when it finds things you wouldn't, right? Like right now, my Codex system prompts, they're set up in such a way that every project I have has a secret- separate, uh, notes MD, and it just writes learnings to there. And then the, the global one can pull from all these. So sometimes it'll be like, "Oh, there's this project you did like four months ago. Here's a note that we had," and it, it randomly pulls it back into context.

I would never do... I haven't thought about. Exactly. And I'm like, "Okay, this is quite superhuman," right? Like, stuff that would... A- and you know, it'll save like hours on chunking of stuff or find something that's already been done. I'm like, as much as it might miss stuff, I would too, but it's very useful when it finds stuff.

And I have like a very, you know, non-super engineered solution to this. It's just marked down files that get pulled whenever they want. Yeah. I actually have a funny anecdote about, about this. Like, recently gearing up to this launch, you know, the team has been, you know, really cooking on it for, for, for a couple months, and over that time, like there's so much conversation and chatter going on in Slack and Docs and elsewhere.

And, uh, one of, one of the members of the team set up this, um, scheduled tasks, like automation to like look at everything that's going on and like come up with the best memes and then post it in one of our shared channels. And like, there are two cool things about this. Like, the first is like, I think the models are, you know, over time, like actually starting to become like- Funny?... funny. Nice.

Whereas like, you know, a year ago, like that was not at all the case. The second is that it was what you were saying, like they find things that in surprising ways that you may not, not have thought of and like create connections that you may not have thought of. And that really helps with like the meme generation [chuckles] because then you can see something that, you know, genuinely surprises you and, and, and, um, is funny in that way.

So yeah, I mean, obviously that's like not like the most productive, uh, use of this-- the, the technology, but it, it does-- it doesn't cover this like, this capability that's emerging, which is just like to find information that you otherwise would not know of. Talking about the, the launch, I, I think, uh, I have pretty much said this is the most successful launch in a long time. I think even more successful personally than 5.0, and they're announcing ten million users.

Does it feel different? You've been through a lot of launches. I think it feels like a culmination. Well, I, I think two things. One, it feels like a culmination, like I was mentioning earlier, like this like vision mission that we've been on for a long time. Like I said, we saw the magic of Codex internally, and then we're like extremely excited to bring this to many more people and to see it working, to like see us reach, you know, the distribution goal, uh, I mean, um, numbers that you mentioned.

Like, I think that's like huge and, and, and super exciting. The flip side of that is like, there's so much more to do too. Like, that's also really exciting. Like, you know, ChatGPT as a whole, like the, the-- this product that, you know, everyone almost equates to AI and like loves, you know, has hundreds of millions of users.

And so like ten million is really cool, but like we, we need to get this to everyone. Like we need everyone to feel this magic. Um, and so that's the next step from here. But yeah, I think extremely pumped about how, how it's going so far and, and the opportunities. Uh, awesome. Uh, I did want to also-- Because I've, I've, I've been tracking the, the number closely, it transitioned at some point from just Codex users to Codex plus ChatGPT Work, obviously because the same harness, the whole point is that you don't, you, the, you can't, uh, count them separately.

Do you have roughly a billion, uh, ChatGPT users? Why did it just jump to one billion right away? Like, isn't that the default on ChatGPT or no? We don't default you into ChatGPT Work if you're on ChatGPT- If you're free. Yeah... um, it's also only available to, to paid users right now. And, and I think there's, like, a process of, you know, educating users of what is the value of this product, having them try it, learning from their feedback, and making it better over time.

But, I mean, the goal is to, you know, get as many of, of the people who, who love ChatGPT today to, like, feel the, feel the power of ChatGPT Work. But I think it'll be a, a journey. Yeah. And Codex will still be alive as a brand for the foreseeable future. Yeah. Um, and we'll just toggle between them as, as, as needed for UI stuff. Yeah, I think it's e-even stronger point than that.

Like, I think we fully intend to, like, you know, treat developer, like, developers have been, you know, a core market for us for so long. And, like, there's, there's so much more that we can do to make Codex great specifically for, um, software development, and we'll continue to do that. This doesn't take away from that at all. If anything, it should increase the utility of something like Codex, because now you can move seamlessly between writing a diff to creating an artifact or, you know- Yeah... doing a search for your character.

I, I do wonder how much this terminology leaks to the non-technical user. Like, do they have to learn to say artifact if I want artifact? Or you know. [chuckles] It's funny, like, we call artifacts internally 'cause that's what the teams call it. It's actually six, yeah. But, like, externally, like, no one says that. No one calls it an artifact.

I-- But I think that people, like, often, like, describe things, whatever they're used to, right? So if, you know, ChatGPT Work is good at creating slides, they'll say ChatGPT Work is good at creating slides, and that's actually what we want. One big-- Another, I mean, it's July of twenty twenty-six.

[chuckles] One big thing that also happens in, uh, for OpenAI was OpenClau, and that's, I think, a lot of people's first time really maxing an agent for personal stuff, but also crossing over to work in, in essence same way. As far as, as far as I understand, OpenClau is still independent, but did you go through your own OpenClau moments? Were there any lessons you took from OpenClau to Codex or back? Whatever.

I think there's a lot of inspiration. I, I did go through my own OpenClau moment. I, um- Yeah, tell the story... me and my, um, my, my wife, like, set up an OpenClau to, like, try to manage everything in our house. Not that there's, like, a ton, but it was, like, actually quite useful. We gave it a calendar. It started, you know, creating events for us and stuff.

At some point, the, the laptop that we were running on, it died, and [chuckles] then we got a chance to, to pick it back up. But there was a lot of inspiration there, like, um, you know, in ChatGPT Work, in web and mobile, like you, you get access to, like, persistent computer environment where, you know, you can store files, and those files stay around between sessions.

And the idea is to be able to enable use cases like this. Um, one of the members of our team actually uses ChatGPT Work for what they used OpenClau from before, and I feel like it has, like, completely transitioned, which is, like, uh, workout planning and, like, meal tracking. Um, which again, it's like a worky thing, right? It's, like, not work necessarily, but it's like in s- personal productivity space. But it has all the same primitives. So it has scheduled tasks.

It has the ability to store files on a file system. It has the ability to, like, reference those things over time. Um, and so you start to see the same types of use cases emerge, which has been really cool. Is there a point that ChatGPT Work completely replaces OpenClau? Obviously, they're independent, so. Yeah, I mean, I, I'm, I'm not close to it, so I can't speak to the- Yeah...

the Open Clau roadmap, but I, I don't think so. I think that there's gonna be, you know, there's always a need for, like, this, like, incredible, like, open source technology that, that that team has built. And I think that we can draw inspiration, um, in the product and, you know, ChatGPT, I think many more people have, like, heard about and use ChatGPT than have u- have used OpenClau.

And if we can take the magic from OpenClau and bring it to them, I think that'll be a success. I think that, like, one thing on the ChatGPT Work side that we feel strongly about is that, like, the core experience is that you come to this product, and you have a conversation, start a session, whatever you wanna call it, with this agent.

And the magic of the product is that you can do anything in that moment. And we would like to create a product where you don't have to click a button or to go to a different place, whatever, and you can get whatever functionality exists in, you know, your, your finances app or wh- or any other product, like, in this, in this one place.

And so that's the goal. It's like it-- we, we want an extensible system with plug-ins where you can connect to the tools that you need in order to be able to accomplish, like, a financial task. Where you can, you know, if, if you're doing, like, science work, like, we have an ability to, like, extend the system in such that you can, like, write the tech and it, and per-performs well.

There'll always be, like, products that we support that are best in class at those things, but we want as much of the magic as possible in that core experience. Yeah. Do you think that you can do everything you used to do with Wealthfront in ChatGPT Finance? I actually tried it. I mean, like, ChatGPT doesn't yet custody, [chuckles] uh, cash and, and assets for me.

So, so that part, no, not yet. But I, I mean, there was, like, a whole component of, like, retirement planning and, um, sort of like financial planning and budgeting and stuff that, that, um, we were looking into when I was there. And, like, with the finances plugin, like, that's all possible [laughs] with ChatGPT today. So, um, I feel like at least that component's replaced for me. I haven't really plugged it in yet.

Um, I'm somewhat scared to look at the answer. [laughs] Like, that's honestly, like, the same reason for health and finances. Like, I'm like, "Uh, I don't know." [laughs] It's really good. I mean, it's, it's really cool how-- I mean, we were talking about, like, the agentic search aspect a little bit earlier, but, like, it's really cool how, like, you know, in, in conventional UX, like, if the more power you wanna give to a user, the more, like, knobs and bells and whistles you need to add.

Like, you know, for, like, these finance and budgeting apps, like, there's always, like, a bunch of the different filters and, like, search bars and stuff like that. But, like, now, like, with the right- Connect-connectivity to the right data, you can have whatever you want. You can ask any question you want- Mm-hmm...

um, and into that box and, and get the answer, and I think that's super powerful. I think it's also nice to just have it centralized in one space, right? You have different health apps. I have one for a smart scale, a watch, all these different things. It's just nice to centrally co-locate it. Which is, uh, you know, part of the whole thing of OpenClaw, right?

Like that, that you would have, um, personal OS, um, which presumably ChatGPT wants to become. I do think that just relying on, like, just-in-time pulling of data for, let's say, through via MCP, CLI, API, whatever you, whatever you do, still not enough. Like I, like, I come from a bit of, bit of a data engineering background, like you still want like a data warehouse or some kind of caching or semantic layer. Um, do you feel that or do you already have that?

I can't speak to like all the details on how everything works, but I think it depends on the access pattern, right? Like if you want an answer immediately, then yes, it's very difficult to do that if you need to pull from all of these sources. But a lot of the like use cases that we want to enable in ChatGPT Work aren't necessarily something that you need immediately.

It's more like a task that you want the agent to go and do, and that's gonna take a certain amount of time. And, you know, with things like programmatic tool calling and stuff now, like some of that time and subagents and stuff, like some of that is also parallelizable. And so it's possible-- I, I think it's very possible that there's a, the ceiling on what can be done, you know, with MCPs and like calling out to these third-party services has, has been raised- Yeah... substantially.

So we're really excited about that. You mentioned subagents. I, I gotta double-click on that. Uh, Ultra is a new mode. Um, you have special affordances in ChatGPT itself to show off the, the agents. Can't really do much with them, to be honest. Like just, just watch. [laughs] Um, what have been, what have been your experiences, any design issues that you would call out to other builders building with subagents?

I think it's sort of goes back to the balance that I was raising earlier about like, you know, showing builders the power of the tool, but also creating enough of an abstraction to not overwhelm them. I think with subagents, the thing that we wanted to show is that you can take a task that, you know, has many parallel tracks or, um, is, is complicated in a way that, you know, subagents can handle, and this product is for you.

Like the, the model can, can accomplish those goals or try to accomplish those goals. Um, and so like that's the point of like showing them in the product and, and that's where we-we've gone with the design. There's another, you know, iteration of this where like you can see exactly what they're doing and, and things like that, which I think is like, you know, could converge on like overwhelming, um, with information.

And so this is like the deliberate trade-off that we made for now. I mean, you, you do display quite a lot of transcripts. Right. Right. I think it's hidden by default, though, right? You want to display more than that? No, no, it's hidden by default. Some, some people could want more. So I'm one of those people that will basically throw a lot of stuff at goal, and pretty much every goal I'll tell it to use subagents.

Seems redundant, right? But every time I'm like, "Okay, use subagents where possible." And I have a lot of people, a lot of friends that recommend and do the same. Whereas I'll sometimes talk to people that are like, "Okay, this is where I want you to use subagents for this subtask," and I'm sure they would appreciate seeing into how they're being used.

For me, it's primarily like two things, right? One is net time efficiency, so span out across subagents. Two is probably cost, right? Right. Uh, don't use big, expensive model. Offload to a lot of smaller, cheaper models. And some people want that level of control. So if you have repetition in what you're doing, right?

Say I want something built where I want it to consistently do this every day, I might want to go in and fine-tune subagents here, subagents there. So- Yeah... you can see both, but I think if I'm not mistaken, it's hidden by default. There's a dropdown that goes a lot where I'm like, okay- Right... I'm just gonna keep, you know, using.

You can, you can change the model that they use. I know I tell them to be steered. I'll say my-- I know Anthropic offers this in CloudCode. You can tell Fable to use Sonnet or Opus to use Sonnet as subagent, so pretty trivial thing. You know, you tell it to span out subagents with Sonnet, you know it's cheaper, faster.

I would assume if it's not there, it could be built there. But I think there's a side of- It's too many toggles. Hmm. It's not a toggle, actually. It's just, uh, you tell it in chat. You're prompting it, yeah. The way I do it is prompt it, right? And I think this is something that gets abstracted unless it's something you built for repetition, right? So if I'm building something, say that's a podcast prep, right?

Research into people, do a very, very deep extensive research, um, that I might wanna configure to cheaper, faster model just for web search, right? I can see a world in which you want both. I think the default is actually pretty good right now, where it's hidden, but you can drop down and get some more info into what's done. Mm-hmm. I know people talked a lot about it on 5.6's launch.

Uh, this thing loves to use a lot of subagents and causes the ChatGPT app to just crash because it's so processor heavy. But, um- For what it's worth, that's not my experience. Yeah, I mean, uh, you know, I, I, I haven't had a crash from subagents. I, I haven't either. I have-- we both have big laptops. [laughs] But I know, I know people brought it up.

There was a topic of discussion that we didn't see the same, but it is another vibe eval, right? People are like, "Okay, the amount of subagents Sol is wanting is crazy." And I'm like, "I think this is okay. I think it's good." But just stuff people bring up. I think when we launched the product too, we weren't as opinion about like who is Ultra for and like when, when should they be using it.

And since then, we've made some changes to like, you know, require you to turn it on and, and find it in the advanced setting 'cause that's who it is for. It's for like power users who understand what's gonna happen because it also, you know, depending on your use case, can, can use more of your, your limits as well. Um- Yes.

[laughs] So that's where I think a lot of the, the feedback was coming from. It's okay. Reset the limits. [laughs] Always reset the limits. Well, it's, you know, today we're resetting because of this. Uh, I wanna change topics to one last piece of the harness, memory. A lot of people are commenting on memory recently. ChatGPT's new memory system used to suck, it's not very good.

Uh, and then this guy also basically the same thing, and Samir, who you presumably work with- Mm-hmm... uh, talking about memory. What can you say there? I think that, you know, Samir and the team have made a ton of-- and, and then the research teams have made a ton of, um, updates and, and improvements over time.

I think when I talk to friends, family members about what they love about ChatGPT, like the fact that it knows them, that they feel like their ChatGPT is, is their ChatGPT, I think comes up probably- Yeah... number one. In ChatGPT Work in the, in the cloud, like by default, all conversations like inherit from your ChatGPT memory, so you'll know-- they'll know context about you, and they'll also be able to write back to this memory.

With it like a, like a, like a small text write, like you tell me when you're writing, right? Is it- No, it's part of the same like memory V3- Okay... system that, that we, we launched. Yeah, dreaming V3. Yeah. So I think that's been really powerful because, you know, going from ChatGPT to ChatGPT Work feels like an extension of what I've already been doing with the product for sometimes many years.

Um, so that's been awesome, and it's awesome to see that like people are recognizing- Yeah... um, the, the improvements here. Is there-- So it's, it's basically a retrieval problem, right? Like, are you retrieving the right things? Are you over-focusing on the wrong things? Is there like a more false positive or false negative, you know, if, if that makes sense? Like, what's the bigger problem?

So I don't work on memory directly- Okay... so it's hard to say what the bigger problem is with like certainty, but I, I think you're right. I think that like, you know, the, the-- there's two sides of it. It's like, you know, making sure it knows things about you, but then also having the EQ to like bring those things up at the right moments proactively or surprising you in ways that are positive, not negative. Yeah.

Um, so I think it's a very challenging problem, but something that I, I think we feel very-- is a huge opportunity to get right, which is like why we've made like big investments in it. How do you see the side of, okay, when you're building ChatGPT for work different than the regular chat app, different than Codex, managing memory across different projects, um, collaboration and whatnot, how do you see the, the side of what's separate from the harness, right?

So if I have four threads on one project- Mm-hmm... um, any learnings on how to build memory systems there, you know? For background as well, I guess to steer it a bit, is when you do chat style applications, I'd say you have a lot of one-offs, right? Yeah. When you switch to work- [laughs]... it might be something you're doing for a month, something you do a lot, right? Now, as I add more sessions, there's a lot more than just single-threaded, right? Sure. And there, there might be memory there.

I mean, I think first I'd challenge that like the depth of the memory or the like value of it is like fundamentally different across chat and work. Like it is true that like, you know, there are a lot of like shorter sessions on chat, but I think, you know, the ChatGPT, the product, has had like a ton of longevity, um, in, you know, as long as this, this technology has been around and, and people use it for work-y, like productivity-related things already today.

And so I think we found that there's a lot of value. I mean, I found this u- my personal usage, like all these one-offs add up over time into something like quite durable and like, like quite a good representation of who I am. I know like from time to time, something will go viral on, on X about like, you know, ChatGPT telling you everything it knows about you, and people are always surprised like how, how deep that is.

The fun roast me, you know? Exactly. So like I think like the-- That's all to say that like I think there's a lot of depth there in the existing, you know, ChatGPT product, and so that's why I think we think it's valuable to bring into the work product. But the other reason I brought that up is because I think like hopefully we can use some of the same fundamental primitives and systems to extend memory here as well, and I know this is something that the, the team that focuses on this is like working, working through right now.

I wanted to bring up one element of memory which I honestly don't really use much, and I'm curious if you do: Chronicle, which was-- is up on screen right now. Uh, it's kind of a super memory or like what, what is it? [laughs] I think the idea is that like it can learn from, you know, how you're using your computer and like it's ano-another input source, um, into memory.

And, um, I think it's, you know, experimental right now and something that like isn't default off. But I'd recommend that you try it. I think that it's like quite interesting how-- It, it goes back to a conversation we were having earlier on like, you know, you were asking like does it-- can ChatGPT miss things?

Like does it, you know, on Slack, when it's searching, does it miss things 'cause there's such a volume of stuff, right? And like it's-- I, I-- You can ask the same question about like everything that you're doing on your computer. Like is it gonna know everything that you're doing? Is it gonna capture the intent and stuff like that?

Probably not, but like it probably will find things that you might not know about, and then if it can surface those to you in relevant times in proactive ways, like when you're doing tasks and... I, I found at least that it can be quite helpful, so it's worth trying. So mostly for insights and longer term. Yeah, exactly.

Like insights and it builds context that, that makes-- that can make you more productive- Mm-hmm... on, on certain tasks, but it's, it's, it's hard to describe without feeling it. Um- [laughs] I will say you can feel it pretty well. Like the idea of what they're saying here, right? Just check through my memories or check through my logs and add skills. Yeah. Pretty underrated, right? But that's automations.

You, you can repeat that u-using a cron job. Checking through your memories and creating skills? Yeah. But I think the creation- I see... of the memories from Chronicle itself is like what's different. Okay. It's like you have much deeper memories because you have Chronicle on. It's there. I don't use it much, but maybe I just... I need more examples.

I, I imagine you guys use a lot of it internally, so I'm always fishing for use cases. Yeah. [laughs] I would just try turning it on and then like- It just auto works? Like it- Yeah, and seeing like where- Right... where it might start helping you. I think you'd be surprised. Yeah. Amazing. I think that was, uh, about it in terms of like the, the overall, uh, coverage of ChatGPT work.

I think there's been a lot of like good progress and discussion on building and all these things. There's a lot of like ex-founders in the, in the community, uh, in, in OpenAI as well. Do you think that things have changed a lot? Uh, I guess, like your overall reflection of building, uh, pre-AI and post-AI. I mean, I think things have changed a ton.

I think it's, it's like super exciting to see how quickly you can go to-- from idea to something real today. Yeah. Um, whereas like even before, like I think, you know, five, 10 years ago, like it w- it was- It's fast if you were scrappy and, you know, like, r-willing to build the, the minimal viable thing.

But, like, now the extent of what you can build is, like, much, much broader. And I think that also, like, what we've seen internally building is, like, that gives you an opportunity to validate much more quickly, to talk to users, to talk to, to internal doctors, et cetera, and, like, make sure you're on the right track.

And, like, that loop, I think, has been-- has become more closed than ever before, and that's, like, a win for product development. I think it's a win for, for consumers and users too because ideally that means they're getting much more better-- much better products out the gate. Does it mean your teams are smaller?

I think there's much more to do now. Um, so I think people can accomplish more individually or in a small team than they were-- that would require more people than before. But there's-- at the same time, there's also more to do, so I think the teams are much more ambitious. You seen any changes in scopes of roles and building teams and how we used to have teams, say, a few years ago versus what ideal teams look like now?

I think we've seen a blurring in the lines between, like, the typical product development functions, like between, like, you know, EM, BM, engineer, um, designer, et cetera. Like- Yeah, I wanna bring up this quote. There will be, uh, only four jobs left in tech. There's AI, uh, slop cannon, the people who just, like, uh, they'll burn a bunch of tokens. Uh, and then there is, there is SRE, uh, who-- people who are more responsible.

There's grown-ups who sell things, and then there's hot people. [laughing] This is an interesting take. I think my, my suspicion is that there's everything, everyone will be, like, T-shaped in a way, and that, like, AI will enable everyone to become a generalist. Yeah. Like, you know, things that, like, I, I never would be able to, like, come up with a design before.

And, like, even now, like, I don't have maybe, like, the visual taste required, but I can iterate on something with the help of AI. But then people will have a specialty, and that's, like, the, the, like, I guess, the straight line in the T or the, the upward line in the T. Um, and so, like, you can have a specialty that you're interested in.

With the help of AI, you can go deeper and become better at over time, but then you'll also be a generalist. And so with that foundation, what you can accomplish is, like, almost limitless. What are you bottlenecked by in terms of specialties? Like, do you need more designers? Do you need more slop cannons? Do you need more hot people?

[laughing] I think the bottleneck some-- becomes, like, sort of, like, ideas and taste, I guess. Um, I think because anyone can, can build now, I think, um, it really is the era of, like, bottoms-up ambition. And because there's so much to be built, like, you're always gonna be bottlenecked by, you know, the amount of ideas and amount of things that you're doing at any given time. Do you think models help solve that? Models?

Yeah. I mean, I have the example of, like, I have a front-end design skill that's like, they give me four drastically different examples of what this looks like. Sure, it burns a lot of tokens, but, you know... And then I'll mostly just condense down, "Okay, I like this part. I like this part. Let's draw these together." And it's like, yeah, I had a vision, but, like, I don't know.

I would say that the one automation that I would love to work and it doesn't work is bring me new ideas, right? Uh, somehow LLMs are just not it. One interesting part about ideas is, like, they're not, like, in a vacuum. It's, like, not-- They, they usually come from somewhere and, like, you know, in, in product development, like, they're coming from talking to users or reacting to, you know, friction that you're seeing or feedback, building on some foundation that you already had planned out before, whatever.

Um, and so I think that's where, like, I think there, there will always be value in, in these, like, generalists that we talked about. Like, you know, closing that loop and, and then having-- coming up with those ideas that are grounded in, in that feedback or talking to users, whatever it is. Cool. Uh, you were gonna-- You lead the productivity team.

How do you define productivity? I think our mission is to make it possible for people to do things that they weren't able to do before. And right now we're thinking about it from the, the perspective of knowledge work. And so when I look at knowledge work, I think about people are no longer siloed by their roles. They're no longer siloed by maybe the, the, um, background or training that they have.

Like, no matter what function you're in, you can suddenly build things and suddenly get access to data that you otherwise might not be able to interpret, et cetera. Um, and then I think that extends to your personal life, where we want to give you leverage at the end of the day. Like, we want the models and the product to be able to give you leverage so that you can, you know, create time for yourself to do the things that you love. Does that also translate to a way to measure productivity?

Like, what is, you know- The end is the-... how do you measure leverage? I think we haven't figured this out yet. Part of the reason is it's so diverse. Everyone has different goals, and really the true measurement is, like, their ability to achieve that goal. Did we help you or did we not? Yeah. And it's very difficult without knowing what that goal is up front and also tailoring it for every individual. And the thumbs up and thumbs down from ChatGPT doesn't give you anything, right? Right. I mean- [laughing]...

you don't know if they're thumbs downing the, the content of the answer, the vibe of it- Oh, yeah... whether or not it helped them with their goal. Yeah. I think that's difficult. Um, but it's something that I think w-we will need to figure out and the industry at large will need to figure out because, you know, that's how we, we measure success, if this is what we're, we're- Do you think it's changed, productivity and how you measure it?

Basically, you said there's a lot more work that can be done, a lot more scope. Um, has it changed? I think it was always true that what you really wanted to measure is, like, you know, was your team, was the individual, were you personally able to hit the goal, or are you closer to hitting that, whatever your goal is, right?

But I think previously we used proxies for this. So like, you know, code commits or- Lines of code... lines of code or whatever, um- Story points. Yeah, exactly. Story points. And, like- They're coming back, by the way. [chuckles] May-maybe. I mean, but, but that is for a part of the change. And, like, I think with, with AI now, those proxies starting to fall apart.

Like, you, you know, the number of tokens you use or the number of pull requests you make are, like, no longer, like, maybe as hypercorrelated with that, is, was your team able to hit the goal or are they on, on track to hit their goals? So I think we'll need to come up with, with new, um, measurements. For the managers listening, uh, give them one thing to try. I think for me, what's important is like at-bats.

Are we as a team building the muscle to have not just quantity of at-bats, but quality? Like, are we able to go all the way from, like, generating an idea, building it out, getting the feedback, reacting to that feedback, actually validating or invalidating the hypothesis, going on to the next idea? Are we able to do that really efficiently?

And like that goes to like, you know, the actual like code that's being written or the designs that are being made or the specs that are being written, whatever, but also the culture of the team. Like, do we have the humility and, and, and, and, um, are able to like go through that process many, many times and stay motivated and excited throughout that?

Um, so that's the thing that like I think is important now, especially when we're on the frontier of this technology and like there's so much to build, there's so much to do. That's probably the most, most important thing that we look at. Any traps people fall into around measuring productivity, what your team work on? I feel like there's a lot of, okay, we added a lot of LLMs.

We have dashboards for this and that, but not much has changed, right? [laughs] That is the trap, yes. And you know, the, the broader source of the question is for, for the managers and teams building, you know, how, how should they approach this? I think maybe the trap is like conflating motion and progress.

I think motion is much easier now than ever before because of the tooling that we have. But progress requires you to be like very prescriptive and deliberate about like what you're actually trying to achieve. And it goes back to our question of measurement, right? Like you were-- we were talking about like, can we, OpenAI, like figure out how to measure productivity for our, our users?

That's, that's a very hard problem because of the diversity. But like as a team, like you should have a really prescriptive and deliberate view on like what progress looks like for you and for your team. And if you don't have that, then it's very easy to conflate these two things. I think at-bats is a really great thing. I, I'm, I'm really glad-- I, I like the discussion between motion and progress.

I think that's a quote that we're gonna feature on the writeup. You've been very generous with your time. Thank you so much and congrats on ten million. Yeah, thank you for having me. The next to, uh, next one at a hundred [laughs] in two months. Sure. [laughs] Two weeks. [laughs] Thank you. [outro jingle]

Codex from 0 to 10M Users: Building ChatGPT Work — Akshay Nathan, OpenAI

Kapitelklicken springt direkt hin

Shownotes

Transkriptvom Podcast bereitgestellt · Zeitmarken springen hin