OpenAI’s GitHub Alternative, OpenClaw Craze in China, and the AI Chip War

0:13James Everingham:Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Wednesday, March 4th. We have got a fun show for you today. Before we get started, I want to direct your attention to the latest edition of our AI Agenda newsletter. The information is first to report that OpenAI is nearing the release of its latest GPT 5.4 model. We get into all of the details of what the company is working on, including an extreme reasoning mode that allows the model to think for even longer on super tough problems. This is, of course, an important story to follow as Google and Anthropic have rival models that have now become more fierce competitors to OpenAI.

0:53James Everingham:Be sure to check out that newsletter for more details. We will link it in the show notes. Elsewhere in OpenAI news, today on the show, we are digging into our exclusive reporting that the company is building an internal alternative to Microsoft's GitHub. We'll bring on one of the reporters who broke that story and also a former top meta engineering executive to unpack that news. We'll then bring on our Asia Bureau to share with us all the ways in which OpenClaw has stirred up a craze among developers in China. And finally, I am sitting down with the founding managing partner of Celesta Capital, a former longtime Intel exec, to talk about the current state of play in the chip sector.

1:33James Everingham:It's going to be a fun show, so let's get right on into it. OpenAI is building a tool that would rival GitHub, according to an exclusive report from The Information. It is the latest way in which OpenAI is competing with Microsoft, one of its biggest investors. It is also the latest in a sprawling array of product development efforts that Sam Altman's company is working on. To unpack what we know, I want to bring on Aaron Holmes, one of the reporters behind that story. And I also want to bring on James Everingham, the CEO of AI agents infrastructure company Guild AI, which just raised a big Series A from Google Ventures this week.

2:09James Everingham:James is also a former VP of Developer Infrastructure and Engineering at Meta. Aaron and James, welcome to the show. It's great to have you both here. Thanks for having me on. Aaron, I want to start with you. I want you to remind us what GitHub is for those of us who haven't used it as frequently as developers have. And then tell us what we know about this tool that OpenAI is working on.

2:31Aaron Holmes:Yeah, so GitHub is essentially a platform where companies can upload their source code and their developers can work together in a collaborative way, similar to how people might work together in a Google Doc for us non-coders. And essentially what we learned is that OpenAI has recently gotten a bit frustrated with outages that GitHub has been having in the last couple of months and is essentially looking at building their own platform where they can store their source code and work on it together and hosting it internally. And we still don't know if OpenAI is thinking about releasing that product as something that it would sell to its own developer customers.

3:09Aaron Holmes:That's something that they have been discussing internally as far as we know, but those plans could go either way at this point.

3:16James Everingham:And so do we know anything about how far along they are in the product development process or when this thing could be launched even?

3:23Aaron Holmes:As far as we know, this is pretty early. I think the soonest it would be ready is in the next month or so. but I think that they're still deciding whether this is going to be something that entirely replaces GitHub or whether it's just something that supplements that work. All of that is still unclear.

3:41James Everingham:And Aaron, do we know anything about how they plan to differentiate themselves against a giant like GitHub?

3:48Aaron Holmes:So I think the biggest way they would do that is by combining this offering with their Codex coding agent. That's something that they sell to developers to automate the process of writing and editing code. And Codex is powered by OpenAI's models. GitHub has its own GitHub Copilot product, which in the past has also mostly relied on OpenAI's models. So if this is something that OpenAI can make a bit more cutting edge than GitHub Copilot and draw in users that are already using Codex to generate code and integrate it more with that product, that could presumably be a way that they would differentiate it from the likes of GitHub.

4:25James Everingham:So, James, you've been an engineer for your whole career. What do you think of this story?

4:30Sriram Viswanathan:Yeah, I think it's interesting. Partnerships in technology often involve collaboration in some areas and competition and experimentation in the others. So I think the real interesting thing here is, like, how is AI software development evolving and what do the systems that need to support that look like? And that's the question that I'm wondering what the answer is here.

4:55James Everingham:And in terms of your use of GitHub, I wonder how disruptive these issues that GitHub has been having has been for you. And ultimately, I wonder what would OpenAI need to do to make its alternative compelling for someone like you to use instead of GitHub?

5:12Sriram Viswanathan:Yes, I think GitHub is a very successful platform. It's got a rich developer community and an economy around it. So I think it's going to be around for a long time. That being said, the way that we develop software is really evolving quickly. It's moving from people writing lines of code and files and working collaboratively to more of systems evolving. And I think that that work will require some different functionality. So it makes sense that there would be a new intelligent layer that would pop up on top of the source control to help manage that. So it would need to be some pretty differentiating functionality to help scale these systems and run them effectively.

6:00James Everingham:Any ideas on, you know, now that you're in the seat now as a founder, you're designing products and stuff like that. What sorts of differentiating factors, what sort of features would you want to make this compelling?

6:12Sriram Viswanathan:Yeah, absolutely. So moving, if you source control basically is a series of files. You can think of it as a file cabinet with, like, you know, where you store your code. Where systems, where it's going to is, like, more complex problem solving and system evolution. So, what maybe the repositories of the future might look like looks more of, like, a system graph or a knowledge graph. And so, I would need to see something that would support a higher level abstraction of this. Plus, there are other things that need to be taken into consideration as these software development flows evolve. Things like security and compliance and things like that because AI is non-deterministic.

6:57Sriram Viswanathan:You can't depend on the answer being correct. So feedback loops, different types of hooks into the system to allow these to get better will all need to be built on top of these repositories.

7:11James Everingham:Aaron, as James talks about this, I wonder, I mean, the people you're talking to, the sources, I mean, are these in the conversation of things that OpenAI is looking at or are we still sort of unclear on that?

7:24Aaron Holmes:Yeah, I think there's been a lot of, you know, discussion in the developer community about how workflows need to change now that AI is playing a bigger role in creating and editing code. And, you know, specifically, we have seen a lot of new startups pop up in the last year that are trying to solve this problem specifically. In fact, GitHub's former CEO, Thomas Domke, who just left in September, has just raised money for a new startup that is essentially trying to reimagine, you know, how developers will interact with code in the future when maybe AI is writing all of their code and they don't actually need to look at the code line by line anymore.

8:01Aaron Holmes:So I think there's definitely an appetite to sort of disrupt the GitHub model right now. I do think that one big question for OpenAI, if they were to productize this, is how do they convince companies to trust them to upload their source code into OpenAI's platform? I mean, then again, we've seen a lot of companies have no problem using products like ChatGPT Enterprise to deal with sensitive company data.

8:27James Everingham:James, how big of a challenge do you think that would be convincing companies to upload their source code into an OpenAI platform?

8:34Sriram Viswanathan:Yeah, that's a good question. We're definitely seeing across the industry companies are willing to take a little more risk. I think that the pressure of competition in building products faster and higher quality is a thing. So I think it's hard to tell. I believe that there's probably areas where there will be more sensitivity across the code base. And so it may not be a black and white answer for each company. There'll be areas where they're comfortable with it and areas where they're not.

9:08James Everingham:But, James, one thing I wanted to understand, and Aaron, you pointed this out in your story, is it is somewhat common for companies to sort of develop their own internal tooling like this at companies like Meta where you worked for a long time. Right, James?

9:21Sriram Viswanathan:Oh, absolutely. So, you know, getting back to why other reasons why they might be doing this is, you know, we can map that to some of the same reasons that maybe we saw at Meta. And, you know, it's no secret that we built our own source control system, mainly because the share scale of the system. It's billions of lines of code. You know, GitHub just purely couldn't handle it. So, look, if you're operating a very highly reliable business and you have a very large source base, you probably want to be in control of some of that so you can provide the reliability and the scalability that's needed.

9:57James Everingham:Aaron, what do we know about how Microsoft feels about all this? One of OpenAI's biggest investors.

10:03Aaron Holmes:Yeah, I mean, so far we have seen that OpenAI is not afraid to compete with Microsoft. They have been increasingly releasing products that they say can automate the work that white-collar workers might have previously done in office, products that can create Excel spreadsheets and PowerPoints, for example. And I think that this is probably potentially another front of that competition. Microsoft, of course, has also been branching out from OpenAI and has been increasingly striking deals with Anthropic and other competitors. So I think we're very much in a world where these companies are maybe more competitors than partners, or at least do both in equal measure.

10:44Aaron Holmes:So I don't think that Microsoft has any way of stopping OpenAI from developing products that compete with it. At the same time, I think it's safe to assume that Microsoft would be closely watching this and thinking about how it strengthens the pitch to developers that GitHub is still the best place to use as a code repository.

11:05James Everingham:And James, I just want to come back to you because some people listening to the show might not have as much technical expertise as someone who's been in engineering for more than 20 years now. But I just want to go back to the question of how OpenAI could make this compelling. If you were explaining those differentiating factors that you talked about to someone who wasn't in tech, you mentioned stuff like security. You mentioned stuff like context. I mean, just bring it a little more high level to what could OpenAI do to make this compelling? Sure.

11:35Sriram Viswanathan:Well, I think, like, one, it would need to – I think that it needs to evolve to be able to handle these new use cases. And, again, I'm going to go back to the whole industry is experimenting with this right now, and nobody really knows where it's going to land. But I think that adding functionality that would address some of these issues that we've talked about, such as reliability and scale, I think adding an intelligence layer on top that allows us to step back a little bit from looking at things in the form of files, commits, and comments, and also evolving the system to be able to understand the evolution of the more complex software systems, which is really where the work seems to be going.

12:23Sriram Viswanathan:Writing the code was never really the hard part. It was really the evolving the system and understanding how it all fits and operating together. So an intelligent layer on top of that that helps support that and evolve that would be pretty compelling. Great.

12:38James Everingham:Well, Aaron and James, I want to thank you for coming on. That is Aaron Holmes, our Microsoft reporter, and James Everingham, the CEO of Guild AI here on TITV. OpenClaw has caused quite the stir across Silicon Valley over the past few months. And that same stir is sweeping across China. My colleague Juro Osawa, a reporter in the Information's Asia Bureau, published a deep dive today on all the ways in which developers in China have started to leverage OpenClaw's technology, and also the ways in which the big cloud companies in China have also started to offer services based on OpenClaw. I want to play for you a conversation that our Asia Bureau chief Jing Yang had with Juro about his reporting.

13:20James Everingham:Here is that conversation.

13:24Juro Osawa:Hi, Jiro. You've been covering China tech for quite some years. Can you tell us how China's tech and startup circle has been reacting to the open claw frenzy? Sure.

13:35Jing Yang:I think this whole thing may be even crazier in China. So many tech founders became completely obsessed with the open claw back in January, and they immediately started looking for new opportunities that could come out of this. and they're developing new products or new business models using AI agents. So one example is a founder I spoke to in Hanzhou, and he was really inspired by Motebook, which is like a social network for AI agents that came out of OpenClaw. And his startup organized a hackathon last month to create all other kinds of apps for AI agents, like, for example, the equivalent of Tinder for AI agents or a recruiting site for AI agents.

14:28Jing Yang:And there are other types of people, like I talked to a guy in Beijing who is using OpenClaw to scale up his one-man business of running social media accounts for AI influencers. And basically, he built a whole fleet of OpenClaw agents running on secondhand MacBook Air that he bought, like a dozen of them. And so, yeah, it looks like all kinds of founders are jumping on this opportunity. And some of those people are saying this is because they feel like they have to do it right now because they know that all the other entrepreneurs in China will be doing the same. You know, your competitors will be launching something if you don't do it right now.

15:20Jing Yang:So they feel like they have to respond.

15:24Juro Osawa:So quite a, sounds like quite a bit of a fear of missing out going on there.

15:29Jing Yang:Absolutely.

15:30Juro Osawa:I just want to quickly circle back on the Tinder for AI agents idea. How's that supposed to work? Is there something built for AI agents to date each other, fall in love with each other?

15:42Jing Yang:So I think that one, the idea is that because AI agents are kind of working on behalf of the human owners, so if they understand the human owners, you know, preferences, love interests really well, then, you know, the agents can go meet with other agents also representing humans. And then I guess the goal is, you know, ultimately the humans can eventually meet based on what the agents have arranged for them.