OpenAI is building agents for computer work and personal life

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 50:58

   of transition to agents no question right people in software engineering you've been feeling this for let's say the past six months and the you know of the course of 25 i think there was a transition from yeah it's like kind of autocomplete to okay yeah you're kind of sidebar in your editor and that you'll start talking mostly over there but you're still mostly doing the same kind of software development you were doing before to now it's like actually you want a tool like codex that is really an agent management platform and the agents are going to take care of all the details the agents are going to do all the the nitty-gritty work and there's still probably 20 percent of how you fit things together and the the way that you will structure your code and exactly some of these these higher level things that you would normally put an architecture document that the human still

2. 51:41

   really cares about and wants to manage but the details of exactly the code like nope that's what agents are for. And so the question that we had is, first of all, how do we really rise to this moment? Because it's not just software, right? We see line of sight for every single vertical, right? For law, for finance. Some of the mechanical skills there of creating spreadsheets and presentations, how to make our models extremely good at those, right?

3. 52:07

   It's like you work with domain experts, you produce evaluations, you produce training data, you have the AI actually take its great domain knowledge and apply it in these verticals and get experience and get those, you know, yep, you did a good job in order to figure out what good looks like. so we have the mechanical the the vision of exactly how to do this but we need to make sure that we're building the right product surface to unlock all of it and one thing that we have found is that the models have shifted from being the product to being a part of the product right we used to have these very thin layers of software on top of them and you didn't have to like think that hard about how it was architected but now it's a very fat layer right that you have things like skills, connectors, you have exactly how you hook up to computer use, how you manage context

4. 52:51

   and memory and all of these things. And so there's just this deep layer of software, which is almost you could think of it as the AI. It's kind of like we have this brain in the form of the model. Now we're building the body. Both are hard. They have to be co-designed together. And so a lot of what we're focusing on is number one, getting together an amazing agentic platform. Like that is the number one focus that we are delivering on. We have teams that are executing extremely well on this. I'm super excited about what we'll be releasing over upcoming weeks. The second is where do you actually want to apply these agents? And that our priority there is really towards computer work, right? So that I use that term very deliberately, by the way, people like to talk about knowledge work, but no one thinks of themselves as a knowledge worker, right? Like

5. 53:29

   that's not a thing that people do. It's kind of a term that's like almost removed from what the actual thing is. But the thing I like about computer work is like, I don't really want to do computer work. That doesn't sound like the thing that I want, but you realize how much of your time, you do spend doing it, like chained behind your desk, typing away, hunching over your shoulders, getting your carpal tunnel, all of those things.

6. 53:47

   And so we are focusing on that, bringing Codex as it exists today, not just for software engineers, but really making Codex be for everyone. And that's something that's coming very quickly. We'll have some updates even coming today as of this podcast filming. And there's a lot of exciting things that are still in the pipeline for that direction. And then the third thing is really thinking about personal AGI, which is about – think about ChatGPT right now used by a billion users.

7. 54:12

   And every single person on the planet is going to want an AI that represents them, has their context, that they have built trust with. It's not just something that you build or it's not just something that you talk to one-on-one, but it can be out there doing things for you. For example, maybe it knows that you like a specific musician and that musician's in town and it notices it proactively. And tickets have just become available.

8. 54:34

   There's some great tickets that are very cheap that you could get for you. It just goes and buys them, right? And maybe it knows it's built trust with you, so it kind of knows, yeah, I'm allowed to do this without getting approval. Or maybe it realizes, ah, I should probably, I'm not quite sure I should check. And so we're building that as well. And if you think about these things, they all are kind of expressions of something that fits together into a cohesive whole.

9. 54:55

   Like in the end, you fast forward to where we're going. You really just want an AGI, right? You don't want a language model. You don't want threads. you don't want any of these like details you just want something that is helping you that is operating on your behalf that is able to help you solve problems that knows what your goals are and achieve those in work context personal context and so this is what we are prioritizing and building yeah and so you know to me one of the important questions you know people are asking how do we think about consumer how do we think about enterprise and the answer is if you take the definitions of these words as they exist today we care a lot about consumer we care a lot about enterprise. But I think that the meaning of these words will change and blur because what we're

10. 55:35

    doing is we're going to, so we're going to unlock this wave of entrepreneurship. Again, I think we're seeing the leading edges of it. Small companies able to get tons and tons of revenue that was not possible before. And that's been a trend for a while. It's just going to really accelerate. Is that enterprise? Is that consumer? It's kind of neither, right? And so I think we are really focused on solving goals across all contexts. And that is the lens that we look at Thanks.

11. 56:00

    And so that has meant we need to deprioritize other things that are also amazing on their own. So what got cut? Yeah. Well, Sora is the most obvious one. And why? Well, because it's a different branch of the tech tree, right? So if you look at the models that actually power Sora, that they're not unified with the core GPT series. And secondly, the use case is not quite unified either, right? It doesn't fall as far under this goal.

12. 56:28

    like there's creative expression there's something very important there i think it was like an incredible model the team does incredible work and i think that technology will live on for other applications but what we were really focusing on was this the product suite that we want to be delivering what do we want to be doing over the next three months six months 12 months and the thing i described by the way is just step one because we also see line of sight to much more powerful models like you look at what we're doing right now in mathematics it's okay it's actually kind of mind-blowing where this result that I just mentioned of solving a new Erdős problem that seems actually really significant, I was just someone using GPT 5.4 Pro. Like two years ago.

13. 57:08

    Yeah, we used to like train our model. We had a team of 20 people to try to train our models to go solve a computing Olympiad. And we got a bronze medal. A team of 20 people for like two weeks and lots of compute. And now it's just this model to be trained very casually. Someone is able to point internet problems and get this kind of result. What if you point that at drug discovery? What if you do take that team of 20 people and all that compute and you really try to push it for scientific discovery? And that's something no one is pricing in right now. So I think it's rising to the moment of agents, really trying to make sure that the product investments we're making are sort of well-structured, that we think about how all these pieces fit together, that we have connectors that work really well, and each of these pieces can be composed and really

14. 57:50

    also build an ecosystem because it's not just about what we build right that we want to build some example agents you think of codex almost as an example agent but it should be that if you're a developer if you're someone with a creative idea you can build your own agent right that you can build it for your application for your purpose you care about the specific math problem you should be able to apply the agent to that math problem and so we're enabling all of that but you i mean okay just bear with me for one second just to go quick through a couple things around this i mean