Latest / The Index Podcast / Agentic AI Foundation: Will AI Be Open Like the Internet or Closed Like Social Media?
Transcript
- 0:12SPEAKER_00: One quick word from our sponsor, Triton One.
- 0:15SPEAKER_00: If you're building on Solana, your app is only as fast and as reliable as the infrastructure underneath it.
- 0:22SPEAKER_00: At Triton, we built the open source read layer for Solani ourselves.
- 0:26SPEAKER_00: The streaming engine that pushes data the moment it hits the chain, the transaction pipeline that lands your trades when everyone else is timing out, all of it running on bare metal in over 20 data centers worldwide, served by the team that wrote it.
- 0:39SPEAKER_00: We power some of the biggest names on Solana, but with Pay AsYou Go, anyone can start.
- 0:44SPEAKER_00: Sign up, get an endpoint, pay only for what you use, head to Triton.1, Serious Infrastructure for Serious Builders.
- 0:50SPEAKER_00: I've worked with Triton for years at this point.
- 0:53SPEAKER_00: They've been around since the very beginning of Solana.
- 0:55SPEAKER_00: They've contributed open source software that's the core of the infrastructure of Solana that makes everything work since day one.
- 1:01SPEAKER_00: I couldn't recommend working with a better team.
- 1:03SPEAKER_00: Triton.
- 1:06SPEAKER_00: Hey everybody.
- 1:07SPEAKER_00: Welcome to the Index.
- 1:09SPEAKER_00: I'm your host, Alex Kahaya, and today I'm joined by Monik Sertani.
- 1:13SPEAKER_00: He's the CTO and co-founder of the Agentic AI Foundation.
- 1:19SPEAKER_00: Uh it's the Linux Foundation's home of MCP, Goose, A2A, Agent Router, Agent Gateway, and Agents.md.
- 1:28SPEAKER_00: Monik, thanks so much for taking the time to come on the show.
- 1:31SPEAKER_00: I appreciate you.
- 1:33SPEAKER_01: Hey Alex, thank you very much.
- 1:34SPEAKER_01: Um to all of those projects and more soon, I hope.
- 1:39SPEAKER_00: Yes, uh, you and I both.
- 1:40SPEAKER_00: Um and we can talk about that.
- 1:42SPEAKER_00: I think before we get into what the Agentec AI Foundation actually is, because some of my listeners probably have never even heard of the Linux Foundation and much less what the AgenTech AI Foundation is, so we should talk about that.
- 1:56SPEAKER_00: But before we do that, I just want to understand, you know, what gets you out of bed in the morning.
- 2:00SPEAKER_00: Why are you choosing to do this?
- 2:03SPEAKER_00: And I guess you're gonna have to explain a little bit about a little bit about what it is you do so people can understand because it's a little bit different.
- 2:10SPEAKER_00: This is not like a company, you know, you're at an open source foundation, so I don't know, you take it from here.
- 2:14SPEAKER_00: I would love to hear.
- 2:16SPEAKER_01: Yeah, absolutely.
- 2:17SPEAKER_01: Um you are you are absolutely right.
- 2:18SPEAKER_01: It is very different from from most traditional CTO roles.
- 2:22SPEAKER_01: It's very different from any other role I've had in the past as well.
- 2:25SPEAKER_01: Um, what gets me going in the morning?
- 2:27SPEAKER_01: Well, look, I I I care very deeply about um the need for AI to be open.
- 2:34SPEAKER_01: Uh, I've got the strong perspective, right, that for AI to be genuinely useful and genuinely valuable to the world at large.
- 2:42SPEAKER_01: Um, I mean, no, it can be.
- 2:43SPEAKER_01: There's lots of value there to be accrued out of AI, but to make sure it benefits everybody in the world, it's got to be open.
- 2:50SPEAKER_01: There's a huge risk if open if AI is not open sourced, if it is closed, if it's a closed system, if it's a silo.
- 2:59SPEAKER_01: We've seen that happen in other industries.
- 3:01SPEAKER_01: We've seen that happen with, let's say, social media, where you don't have choice, you don't have transparency, and what happens?
- 3:08SPEAKER_01: Well, lots of bad things potentially happen, right?
- 3:10SPEAKER_01: And we've seen the opposite happen.
- 3:11SPEAKER_01: We've seen what happens when powerful technology is built on open standards, with open protocols, with transparency, with choice.
- 3:19SPEAKER_01: The internet, the internet is such a technology, and the benefits have been far-reaching and wide.
- 3:25SPEAKER_01: It's lifted entire countries out of poverty.
- 3:28SPEAKER_01: It's been a really, really powerful force for good.
- 3:30SPEAKER_01: Now, I want AI to be like the internet and less like social media.
- 3:34SPEAKER_01: So that's what gets me up in the morning.
- 3:36SPEAKER_00: And this is why I think you know you and I, I've enjoyed talking to you so much over like the past couple of months, the couple phone calls we've had.
- 3:43SPEAKER_00: I I I believe that AI and in intelligence is a human right.
- 3:49SPEAKER_00: Like everyone should have access to it.
- 3:51SPEAKER_00: And the only way to ensure that everyone has access to it is that it's built on open source software.
- 3:57SPEAKER_00: It doesn't mean that you can't use closed systems like anthropics and open AI's models, but there need to be competitive open source alternatives so that people have choice.
- 4:08SPEAKER_00: And so that like you know, the idea that like only a few companies would have access to this and complete control of this kind of important infrastructure is just really dangerous.
- 4:18SPEAKER_00: And the same thing goes with the governments, right?
- 4:20SPEAKER_00: Like, you don't want any of these actors to be the only people with with access.
- 4:25SPEAKER_00: And yeah, I have I uh my my biggest question is like how do we get there?
- 4:30SPEAKER_00: You know, I think and and maybe that's maybe that's a way for you to lead into like some of the stuff you guys have and kind of explain what all these different protocols and tools are that are housed and how they're housed.
- 4:40SPEAKER_00: But how do you achieve that goal of ensuring this is open and free for everybody?
- 4:45SPEAKER_01: So how we get there is partly why we started the Agentic AI Foundation.
- 4:49SPEAKER_01: And I can give you the origin story there and and the motivation there.
- 4:53SPEAKER_01: But um the key part of that is, as you said, it's not just open models uh and open source software around the models, it's also open standards and open protocols.
- 5:02SPEAKER_01: That's what makes it easy to swap one piece in for another.
- 5:06SPEAKER_01: Um I'll go back to the internet again as an example, right?
- 5:09SPEAKER_01: Like um, you can choose to use a proprietary email system, but you can choose not to as well, and you can switch.
- 5:16SPEAKER_01: And if you switch, because the protocol is open, you've not lost anything.
- 5:20SPEAKER_01: You can take your email history with you, you can take your contact, your address book with you, all this other stuff, which makes it feasible.
- 5:27SPEAKER_01: Now, you can't do that with social media, for example.
- 5:29SPEAKER_01: You can't say, I'm going to quit Facebook, but I want my friendship graph to move with me to some new system.
- 5:34SPEAKER_01: That's not going to work, which is why people don't do it, which is why you get locked in.
- 5:39SPEAKER_01: Now, I think that for AI, for it to work in AI, all the protocols need to be open at every step of the way.
- 5:45SPEAKER_01: The stuff that you use when you talk to a model, the way a model talks to a tool, the way the model talks to a user interface, the way it talks to an app, things like that.
- 5:54SPEAKER_01: If all of those touch points are open and standardized, then sure, you can use a proprietary model where it makes sense, and you can very feasibly vote with your feet and move to an open model later if you wanted to.
- 6:06SPEAKER_01: And that power is what keeps the uh infrastructure companies in check.
- 6:11SPEAKER_01: It's what keeps them honest, and that's what keeps them um transparent and so on and so forth, and doing good for the world.
- 6:17SPEAKER_00: Yeah, and it's interesting because the early adopters in enterprise are already seeing this.
- 6:22SPEAKER_00: Um like the best teams, like I I I had uh this guy, Max, from Franklin Templeton on uh the other day, and this is we're talking about an 80-year-old with a two trillion dollar asset manager, right?
- 6:34SPEAKER_00: A big, huge fund.
- 6:36SPEAKER_00: And they they've invested really heavily in AI, you know, and and they've taken this approach of making their stack modular and composable so that they can achieve the goal that you're talking about, and they see that vision.
- 6:49SPEAKER_00: And I and I know not everybody, not every team is kind of like rocking this out of the gate, but there are some seasoned people who kind of saw how the internet was born that are at some of these companies, these bigger enterprises, and they're realizing, oh, this is the way we're not reinventing the wheel here.
- 7:04SPEAKER_00: You know, this has been kind of done before in terms of the strategy.
- 7:09SPEAKER_00: Um it's just that AI is making everything move just so much faster, right?
- 7:13SPEAKER_00: And uh I I just I just find it interesting that there are some early adopters in enterprise that see the problem and they're like, hey, we need to be able to swap models really quickly.
- 7:23SPEAKER_00: And if we can't, then we're gonna have a big problem someday, you know, and it's easy to lose sight of.
- 7:29SPEAKER_00: And especially for individuals, and that's kind of where I'm kind of curious, is like there's enterprise and then there's the individual, like you and just person to people like you and me.
- 7:36SPEAKER_00: And an enterprise can afford to go buy a GPU rack of GPUs and train an open source model and protect their data, but what about us?
- 7:43SPEAKER_00: You know, how do we how at the model layer itself, you know, how do individuals keep that sovereignty?
- 7:49SPEAKER_00: I'm wondering if you guys are thinking about that.
- 7:52SPEAKER_01: Yeah, I mean that's something that the Agentic AI Foundation explicitly does not touch.
- 7:57SPEAKER_01: Um, the way I see it, right, like for the entire AI stack to be open, there are multiple layers.
- 8:03SPEAKER_01: You've got the hardware layer, for example, the actual inference chips and the training chips that are used for AI.
- 8:09SPEAKER_01: At the moment, that's not very open.
- 8:10SPEAKER_01: That's all very proprietary.
- 8:12SPEAKER_01: A couple of major companies run all of that, and that's okay.
- 8:14SPEAKER_01: In the future, maybe that opens up.
- 8:16SPEAKER_01: I don't know.
- 8:16SPEAKER_01: Uh the next layer on top, the training layer, the training layer, the software training layer, rather, that's actually surprisingly very open right now.
- 8:24SPEAKER_01: A lot of that is based on iTorch, on Ray, on a bunch of other technologies in that space, which are all open source.
- 8:31SPEAKER_01: They've got their own foundation, uh, works really well.
- 8:34SPEAKER_01: Everything from major uh frontier labs in the US through to open source labs in China all use the same tools.
- 8:41SPEAKER_01: They've all coalesced around the same set.
- 8:43SPEAKER_01: So that's actually a really powerful signal.
- 8:45SPEAKER_01: The next layer on top, the serving layer, that layer also is very open source friendly.
- 8:50SPEAKER_01: That's basically Kubernetes at this point.
- 8:53SPEAKER_01: The CNCF has been talking about um AI inferencing being the largest workload Kubernetes has ever seen.
- 9:00SPEAKER_01: All of this running at scale on Kubernetes, that's all a very open stack, very powerful again, lots of freedom there to be able to pick and choose and move things around.
- 9:07SPEAKER_01: The layer on top now, the agentic AI layer, uh, or rather the models itself.
- 9:12SPEAKER_01: The models themselves, that's where it's a little bit hazy.
- 9:14SPEAKER_01: You've got proprietary models, you've got open models coming out of China, open weights models, not completely open source in the way I'd like to think of open source.
- 9:22SPEAKER_01: There's a lot of debate and dispute as to what really is open when it comes to models.
- 9:27SPEAKER_01: So that area is a bit murky.
- 9:29SPEAKER_01: Now, the layer on top, the agentic layer, that's the bit that I'm focusing on right now.
- 9:33SPEAKER_01: That's where the agentic AI Foundation is focused on.
- 9:36SPEAKER_01: Think of that as the W3C of the Internet.
- 9:40SPEAKER_01: It's the basket of protocols that makes the web work in the internet uh space.
- 9:44SPEAKER_01: In the agentic space, it's the same basket of protocols that makes agentic AI work.
- 9:49SPEAKER_01: Starting with A2A, with MCP and A2A, which is very similar.
- 9:53SPEAKER_01: We've got other protocols like ACP, we've got agentic payment protocols, agentic identity protocols that we're talking about right now.
- 10:01SPEAKER_01: There's a whole basket of things happening there.
- 10:03SPEAKER_01: Um and then you've got the frameworks and the tools that make those protocols work.
- 10:08SPEAKER_01: So, for example, we've got agent routers and gateways, harnesses and libraries, toolkits, and so on and so forth that make it easy to implement things on that agentic layer, and all of that is increasingly open.
- 10:20SPEAKER_01: And that's what the agentic AI Foundation is focused on.
- 10:23SPEAKER_01: Now you're absolutely right that when it comes to the models themselves, that is murky, there's work to be done there.
- 10:28SPEAKER_01: Um, as a community, yes, we are talking about it.
- 10:30SPEAKER_01: We are seeing more open source, open weights labs come up around the world, um, in addition to China, which I think is a really positive step.
- 10:38SPEAKER_01: I'd love to see even more.
- 10:40SPEAKER_01: Um there's the big challenge there, though, is training a model is still very expensive and very difficult.
- 10:46SPEAKER_01: It's not like open source software.
- 10:48SPEAKER_01: I mean, I can compile the Linux kernel in my laptop.
- 10:51SPEAKER_01: I can't train a model on my laptop very easily.
- 10:53SPEAKER_00: Yeah, you make such a good point.
- 10:55SPEAKER_00: I mean, in in you know, this the agentic layer is more like Linux.
- 10:59SPEAKER_00: It can be adopted very quickly, it can change very quickly.
- 11:02SPEAKER_00: A lot of people can easily contribute to it, but you don't need to spend you know a billion dollars on GPUs to train it, you know, but you do in the model layer.
- 11:09SPEAKER_00: And I think we're seeing that play out in Europe right now.
- 11:12SPEAKER_00: I can't, what is it called?
- 11:13SPEAKER_00: Mistral?
- 11:14SPEAKER_00: Is that the lab in Europe?
- 11:15SPEAKER_00: Mistral, yeah.
- 11:15SPEAKER_00: Yeah.
- 11:16SPEAKER_00: So they they just announced that they raised like, I don't know, a billion dollars to build out a bunch of data centers.
- 11:20SPEAKER_00: And they also announced like, and I haven't seen the data myself, but I heard that the performance of their model was like way behind, you know, and um and and I think they were I think they were angling for open source.
- 11:31SPEAKER_00: And this came up in conversation, um, I think with Max, who I was telling you about from Franklin Templeton, uh because we were talking about how there's gonna be like regional open models and like what does it mean for things to be open?
- 11:44SPEAKER_00: Um and and and and the big thing is like adoption matters in open source just like it matters anywhere else.
- 11:50SPEAKER_00: If no one uses your stuff, then it's kind of irrelevant.
- 11:53SPEAKER_00: And so it's so it begs the question is like how do you catch up?
- 11:56SPEAKER_00: Because these things are exponential.
- 11:58SPEAKER_00: You know, it's just such a much harder problem to solve if you're starting from scratch building an open model right now, unless it's like a different form of AI.
- 12:06SPEAKER_00: That's the that's that for me is like the blast part for the in the industry.
- 12:09SPEAKER_00: I like I know it's coming.
- 12:11SPEAKER_00: I don't know who's gonna make it, but I hope it comes from the open source community, whatever that is.
- 12:15SPEAKER_01: Well, it's two things.
- 12:16SPEAKER_01: I mean, one is it's it's like as you said, a completely different type of AI.
- 12:20SPEAKER_01: But the other also is small models.
- 12:22SPEAKER_01: And small models are starting to become a thing.
- 12:25SPEAKER_01: Um, they are not NVIDIA's got that.
- 12:28SPEAKER_01: There's a bunch of Chinese small models that are coming out.
- 12:31SPEAKER_01: Um IB IBM's got one, or Gramaton.
- 12:35SPEAKER_01: That's not Graviton, Gramiton's something completely different.
- 12:37SPEAKER_01: Um Granite, that's right.
- 12:39SPEAKER_01: Uh IBM has a model called Granite, which is a tiny model.
- 12:42SPEAKER_01: Um some of these things are pretty interesting.
- 12:44SPEAKER_01: The whole goal being you can run them on constrained devices.
- 12:47SPEAKER_01: They run on your phone, they run on your laptop, things like that.
- 12:49SPEAKER_01: And then self-sovereign personal AI starts to become a reality as as that starts to you know get more and more powerful.
- 12:56SPEAKER_00: Yeah, I mean, it's it's crazy to think of how powerful this stuff already is, but we're still in like the dial-up phase, right?
- 13:03SPEAKER_00: Like it we're so early, I feel like, in this whole industry.
- 13:06SPEAKER_00: And so uh these things are only gonna get better.
- 13:09SPEAKER_00: I want to hear a little bit more about your history because it you know, I I had forgotten, but I think I read this on your LinkedIn or something at some point, but you were at Block before this, and that's where Gooze comes from.
- 13:20SPEAKER_00: And for people who don't know Block, you know, that's Jack Jack Dorsey's company.
- 13:25SPEAKER_00: It's got uh it's kind of the intersection of like payments and crypto.
- 13:29SPEAKER_00: But one of the things that I like about them, I've always looked up to Block as like a real uh leader in the open source space.
- 13:35SPEAKER_00: And that's that I mean that was what you led there.
- 13:38SPEAKER_00: Um can you just talk about Block a little bit and your work there and Goose and like kind of how that got you to where you are?
- 13:43SPEAKER_01: Yeah, absolutely.
- 13:44SPEAKER_01: I mean that that that's actually a fun story.
- 13:46SPEAKER_01: My background's always been in open source from from very early days, from I mean, compiled my first Slackware kernel in the late 90s, things like that, off of you know, downloading, downloading stuff off of uh really bad ADSL connections and what have you.
- 14:01SPEAKER_01: Uh my background's been in open source for a really long time, um, really deep with the MySQL community early on, with the PHP community, contributing sort of language features and things back by language.
- 14:11SPEAKER_01: Um and I ended up at Red Hat for many years, where I started a thing called Infinispan that got very popular.
- 14:18SPEAKER_01: Um and I thought that was it.
- 14:20SPEAKER_01: I mean, this is my life.
- 14:21SPEAKER_01: I, you know, I don't want I don't want anything more sort of thing.
- 14:24SPEAKER_01: I'm really enjoying being steeped in the open source community and having a company that could sponsor me to do that.
- 14:30SPEAKER_01: Right.
- 14:30SPEAKER_01: Um and somehow Square at the time got on my radar and he said, Oh, let's just go have a chat with these guys.
- 14:36SPEAKER_01: I know a lot of people there, they're good friends of mine.
- 14:39SPEAKER_01: Um, let's see what's going on.
- 14:40SPEAKER_01: And they convinced me to join.
- 14:42SPEAKER_01: And it was Jack's whole idea of saying is trying to build a company that's steeped in open source uh ethos from the beginning, even though it was not an open source company, it was a commercial company.
- 14:51SPEAKER_01: Um and a lot of the work they were building was proprietary, but his ethos was very open source friendly.
- 14:57SPEAKER_01: And I said, okay, let's go and explore this a little bit.
- 14:59SPEAKER_01: So I joined the company pretty early on.
- 15:01SPEAKER_01: Um, and I just watched it grow from strength to strength from an open source perspective.
- 15:06SPEAKER_01: Jack is very much an open source guy.
- 15:08SPEAKER_01: He himself was a big person in the open source community early on, um, hired a lot of people from the open source community, like myself.
- 15:16SPEAKER_01: And that sort of set the standard for how the company was run, which was really interesting.
- 15:20SPEAKER_01: Because of that, open source begins more open source, right?
- 15:23SPEAKER_01: A lot of the work that people were doing, people wanted to open source them from day one, and we did that early on.
- 15:29SPEAKER_01: Lots of upstream fixes to MySQL came out of block.
- 15:33SPEAKER_01: Um, a lot of core Android libraries came out of block, stuff that is de facto standards today around the world.
- 15:40SPEAKER_01: Um, so we started doing that.
- 15:42SPEAKER_01: And then at some point, I want to say maybe about three years ago, Jack turned to me and said he actually wants to formalize things and actually make block an open source company full stop, open source our processes and ways of doing things and everything else.
- 15:53SPEAKER_01: It's like, wow, that's really interesting.
- 15:55SPEAKER_01: We've done a lot of open source work.
- 15:56SPEAKER_01: We've got like 600 open source repos at that time, mostly small ones like component libraries and iOS or Android and things like that, through to really chunky pieces of work like open source hardware designs for our Bitcoin mining rig or open source chip design.
- 16:13SPEAKER_01: And it's like really impressive stuff that we were doing, and it was all in the open.
- 16:16SPEAKER_01: That's cool, but there was no cohesive sort of command center for open source within block.
- 16:21SPEAKER_01: It was all ground up, it was all grassroots, everyone was doing it because it was our DNA full stop.
- 16:26SPEAKER_01: Um, so it was my job to kind of corral all of this stuff and you know, bring it under a single umbrella and a single brand.
- 16:33SPEAKER_01: We called it block open source, and that became a really important and interesting movement, both within the company as well as in the community.
- 16:40SPEAKER_01: And through that, that's where Goose grew.
- 16:43SPEAKER_01: This was in the early days of AI at this point.
- 16:45SPEAKER_01: LLMs were starting to become a thing.
- 16:48SPEAKER_01: People were using Chat GPT to write code by asking it questions and copying and pasting stuff in between ChatGPT and your IDE.
- 16:55SPEAKER_01: And we were like, hang on a second, there's got to be a better way than this.
- 17:01SPEAKER_01: Well, more like 18.
- 17:04SPEAKER_00: It was still, it's like it's crazy how fast it's it's been that we're sitting here talking about it.
- 17:08SPEAKER_00: It's like not been that long.
- 17:11SPEAKER_01: So this predates stuff like claud code and like clawed code and uh codecs and all this stuff.
- 17:17SPEAKER_01: So, you know, there was no coding agent.
- 17:19SPEAKER_01: There was no such thing as a coding agent.
- 17:20SPEAKER_01: The best you had was the GitHub copilot plugin to VS Code or something.
- 17:25SPEAKER_01: I say, okay, well, we can do better than that.
- 17:28SPEAKER_01: And we created this thing called Goose.
- 17:30SPEAKER_01: Um, worked pretty well internally.
- 17:32SPEAKER_01: We liked a few of us were using it, saying, okay, this there's this promise here.
- 17:35SPEAKER_01: So we opened it up to the company internally first.
- 17:38SPEAKER_01: And block's a big company.
- 17:40SPEAKER_01: At the time, we must have had, gosh, maybe 5,000 engineers or something.
- 17:43SPEAKER_01: So it's a large engineering community to expose this tool to.
- 17:46SPEAKER_01: And they all started using it saying, This is actually really cool stuff.
- 17:49SPEAKER_01: It's really helping us, it's really productive, it's awesome.
- 17:52SPEAKER_01: Cool.
- 17:53SPEAKER_01: Well, we needed to find a way to have Goose talk to internal systems.
- 17:57SPEAKER_01: Um, and we wanted a standardized way of doing that.
- 18:00SPEAKER_01: So we started sketching out this idea.
- 18:02SPEAKER_01: We had friends over at Anthropic.
- 18:04SPEAKER_01: We'd worked pretty closely together on a bunch of things, and that was where the idea for MCP was born.
- 18:09SPEAKER_01: So we're sketching out this thing, this interface, how do you actually talk to internal systems uh for tool calling?
- 18:14SPEAKER_01: And we wanted them to generalize it, so we were using their models a lot.
- 18:17SPEAKER_01: And they're like, okay, tell you what, let's go ahead and make this thing.
- 18:20SPEAKER_01: We'll create the standard.
- 18:21SPEAKER_01: Um, you we want you to be our design partners and our guinea pigs.
- 18:25SPEAKER_01: So we were the guinea pigs for MCP originally.
- 18:28SPEAKER_01: Um, and when that got created, we had Goose and we had MCP, and they both started to work really well.
- 18:32SPEAKER_01: We said, you know what, we need to open source this stuff because I think the world's gonna benefit from this.
- 18:37SPEAKER_01: And because it's pluggable by design through MCP, we want to see what other people will do with it.
- 18:42SPEAKER_01: How are they gonna extend it?
- 18:44SPEAKER_01: How are they gonna make it do more things, give it more arms and legs?
- 18:47SPEAKER_01: So we did that.
- 18:48SPEAKER_01: And all of a sudden, Goose, which is originally a coding agent with access to bash and you know, a C compiler or a Java compiler or something on your laptop, all of a sudden had access to Google Drive and to Figma and to your calendar and to be able to do things that we didn't design it to do.
- 19:04SPEAKER_01: And it became this personal assistant for a lot of people.
- 19:06SPEAKER_01: So that was very cool.
- 19:08SPEAKER_01: And off the back of that, Cloud Code was born, and that was a whole separate uh thread that happened within uh Anthropic.
- 19:14SPEAKER_01: But along the way, we were like, well, this stuff needs to live somewhere.
- 19:18SPEAKER_01: It needs a good home, a vendor neutral home.
- 19:20SPEAKER_01: Um, they were getting very popular in the open source community.
- 19:23SPEAKER_01: I think it's got like 30,000 stars in a few months or something like that.
- 19:26SPEAKER_01: It was getting really popular very fast.
- 19:28SPEAKER_01: Um, but we were getting a lot of community noise saying that people wanted a neutral home.
- 19:34SPEAKER_01: They wanted vendor independence to really trust it and build on top of it and extend it.
- 19:40SPEAKER_01: It's like, okay.
- 19:41SPEAKER_01: And that's when between Anthropic and us, we decided, well, we're going to start our own foundation as a home for this stuff.
- 19:47SPEAKER_01: And that's where the Agentic AI Foundation was born.
- 19:49SPEAKER_01: OpenAI soon got wind of it and said, hey, we want to join too.
- 19:52SPEAKER_01: We've got the stuff we want to bring in as well.
- 19:54SPEAKER_01: And realized, you know, we actually worked pretty well together.
- 19:56SPEAKER_01: We got on pretty well.
- 19:57SPEAKER_01: I was like, all right, let's go start this thing together.
- 19:59SPEAKER_01: And when we launched, um, all of a sudden it was 60 companies.
- 20:02SPEAKER_01: I mean, you know, we had Google behind us, Microsoft, AWS, like every major hyperscaler.
- 20:08SPEAKER_01: Um, and and yeah, it was a big bang.
- 20:10SPEAKER_01: That was just December last year.
- 20:11SPEAKER_01: It's less than a year ago.
- 20:12SPEAKER_00: Man, it's so fascinating to hear that story.
- 20:14SPEAKER_00: I did not know the background on how closely you guys were involved in MCP and working with Anthropic.
- 20:20SPEAKER_00: I mean, for people listening, I mean I think everybody listening knows what MCP is, but this standard is used by uh everyone, right?
- 20:26SPEAKER_00: Like it's had mass adoption.
- 20:28SPEAKER_00: And I mean, for people who work on open source software, that's the dream, right?
- 20:32SPEAKER_00: As many people as possible use this thing and it's important and it helps like move the space forward.
- 20:37SPEAKER_00: And MCP has been a very impactful project.
- 20:40SPEAKER_00: Uh, and so you guys like four billion downloads.
- 20:43SPEAKER_00: Yeah, that's a legacy, right?
- 20:44SPEAKER_00: That everybody who worked on that project can stand on for the rest of their lives and beyond, right?
- 20:49SPEAKER_00: Like it'll outlive them.
- 20:50SPEAKER_00: And you know, it's like when we started our company, it's it was also very cool for me to hear the the story about Jack and like the open source ethos that that he built block around because it's very much how I started our company, Corbett's, you know, around just get a few talented engineers in a room and let them cook on useful open source software and like figure it out.
- 21:10SPEAKER_00: And that's so everything we do is open source.
- 21:12SPEAKER_00: I don't know.
- 21:12SPEAKER_00: I just I just find it very inspiring to hear that story and to hear what like came the results after all that.
- 21:19SPEAKER_00: It's so cool.
- 21:19SPEAKER_01: Yeah.
- 21:20SPEAKER_01: And and and there is business sense to it.
- 21:22SPEAKER_01: I mean, open source is not charity, it's not just giving stuff away.
- 21:26SPEAKER_01: I mean, um, this is stuff that I used to talk about at Block a lot.
- 21:29SPEAKER_01: We'd we'd often have um, you know, folks who work in on the business side who wouldn't quite get open source saying, it's a stupid whim of Jax.
- 21:37SPEAKER_01: I don't know why he's doing it.
- 21:38SPEAKER_01: It's just, you know, a drag on all of us, and it's just gonna, it's just a waste.
- 21:41SPEAKER_01: We're giving stuff away.
- 21:42SPEAKER_01: And like it took a while convincing them, but then actually demonstrating that it's not just giving stuff away.
- 21:48SPEAKER_01: You're absolutely growing the pie.
- 21:49SPEAKER_01: The pie is growing for everyone, and as a result, our slice grows as well.
- 21:53SPEAKER_01: And there's a huge amount of community um momentum that you build as a result, and you get stuff done for you.
- 22:00SPEAKER_01: For free.
- 22:00SPEAKER_01: Like another story I love to tell is the GRPC story, which also happened while I was in Block.
- 22:06SPEAKER_01: Prior to GRPC existing, Block had their own RPC network mechanism.
- 22:12SPEAKER_01: For folks who don't know, GRPC is basically a way that different microservices talk to each other.
- 22:18SPEAKER_01: It's a network-level protocol and a toolkit.
- 22:21SPEAKER_01: And prior to that, everyone sort of did their own thing.
- 22:24SPEAKER_01: It was all very home-grown.
- 22:26SPEAKER_01: And Block had their own thing as well.
- 22:28SPEAKER_01: And that was one of my teams back in the day at Block.
- 22:31SPEAKER_01: And we decided at some point, you know what, this needs to be open sourced.
- 22:34SPEAKER_01: We need to change this.
- 22:35SPEAKER_01: We need to modernize it.
- 22:36SPEAKER_01: And we had all these ideas and plans to modernize it.
- 22:38SPEAKER_01: And then I got to meet a bunch of folks at Google who also had an internal proprietary thing.
- 22:43SPEAKER_01: And they were going to modernize it and open source it too.
- 22:45SPEAKER_01: And we sort of joined forces.
- 22:47SPEAKER_01: And those joint forces was GRPC, basically.
- 22:51SPEAKER_01: And the interesting part about it is prior to GRPC, like my team that maintained that internal version of, you know, the free GRPC thing, was like 10 people, uh full-time, 10 very senior engineers, because it's not simple software, it's really complicated stuff.
- 23:07SPEAKER_01: And today, guess how much Block has to invest in GRPC?
- 23:11SPEAKER_01: Like zero, right?
- 23:12SPEAKER_01: I mean, yeah, we contribute to it, but then we get it for free because the community's taken it over and run with it, and we all benefit from it.
- 23:18SPEAKER_01: So by opening it up and letting it fly, I mean it's it's benefited a lot out of it.
- 23:24SPEAKER_00: Well, and there's so there's so many examples of things like that where you're having to invest in core infrastructure to run your business, but that's not your product.
- 23:35SPEAKER_00: That's not the thing that you sell to customers or make money on at all, and you're still having to spend like enormous resources on it.
- 23:40SPEAKER_00: And I I to my team, we um like two companies ago, we were the first people to build an indexer on Solana, and it was for indexing all the on-chain data, and we did it fully open source.
- 23:52SPEAKER_00: We probably invested close to seven figures in that because there was nothing there, right?
- 23:58SPEAKER_00: Like no one had had that tool.
- 24:00SPEAKER_00: And we built it open source, and then eventually, you know, indexing became a thing on Solano because it was all open source.
- 24:05SPEAKER_00: And um, you know, we worked with a couple different companies in the space on that that are like professional RPC companies, different kind of RPC, but for you know, for on-chain data.
- 24:15SPEAKER_00: And um, I only bring that story up because this was back when it was like early days of the internet, but for Solana, right?
- 24:24SPEAKER_00: Like no tools existed.
- 24:25SPEAKER_00: And we this was not the thing at all that we were monetizing as a business.
- 24:30SPEAKER_00: And and somebody might have said, some business person would be like, oh, but we could do this and sell it.
- 24:35SPEAKER_00: It's like, no, that's not what you do, you know.
- 24:36SPEAKER_00: And even on the most extreme case, there's so many uh ways to bring open source to market that aren't just like, hey, this is like not our core product, but if it is, you know, open source is really an adoption mechanism, it's a it's a distribution mechanism.
- 24:49SPEAKER_00: And and especially if you're an enterprise, people uh like what I'm hearing from enterprises is they want open source agentic tooling because they want to own their intelligent stack.
- 25:01SPEAKER_00: And they don't necessarily that doesn't mean they don't want to pay you for it.
- 25:04SPEAKER_00: They actually really want to pay you for it because they need your help putting all this stuff together.
- 25:09SPEAKER_00: It's the classic Red Hat model, and you know that because you worked at Red Hat, but I just find this way of building a company to be so interesting.
- 25:16SPEAKER_00: It's like way more fun, way more interesting, way more exciting.
- 25:20SPEAKER_00: Uh, I I like literally like my hand, like the hair stands up on my arm, you know.
- 25:24SPEAKER_00: Like it's just just a better, it just is a more fun way to do things.
- 25:27SPEAKER_00: I can't imagine any other way.
- 25:28SPEAKER_00: And I know it's not for everybody.
- 25:30SPEAKER_01: And more than fun.
- 25:30SPEAKER_01: I mean, leverage, the leverage is enormous.
- 25:33SPEAKER_01: You suddenly get like this, everyone in the world wants to help you push it forward.
- 25:36SPEAKER_01: It's like, wow, okay.
- 25:38SPEAKER_01: Yeah, uh, yeah, no, it's incredible.
- 25:40SPEAKER_00: Um, and I mean, it seems to be enormous resources, like these 60 companies, you're at 130 now, right?
- 25:45SPEAKER_00: We're talking about like the biggest companies on the planet.
- 25:47SPEAKER_01: Oh, no, no, we're at 250 now.
- 25:50SPEAKER_00: Oh, wow.
- 25:50SPEAKER_00: So I'm my even even since we made these notes, it's way behind, right?
- 25:53SPEAKER_00: Like 250.
- 25:55SPEAKER_00: So, like, how do you quantify that?
- 25:56SPEAKER_00: I mean, you know, you you were talking about how much block used to invest in the in GRPC or what became GRPC, but when you look at all these companies that are participating, you know, it's like gotta be tens of millions, hundreds of millions of dollars in engineering resources and marketing resources and all this stuff.
- 26:14SPEAKER_00: Like, how do you measure that?
- 26:15SPEAKER_01: Oh, yeah.
- 26:16SPEAKER_01: That's right.
- 26:17SPEAKER_01: That's right.
- 26:17SPEAKER_01: I mean, like the Linux Foundation has tried to quantify this because it's it's their job to be able to, you know, to help people understand why open source, right?
- 26:26SPEAKER_01: Um and there's a lot of material out there.
- 26:28SPEAKER_01: I mean, I'm happy to share some links if that's useful.
- 26:30SPEAKER_01: But the numbers are mind-boggling.
- 26:32SPEAKER_01: Like if open source didn't exist, the cost that every one of us would have to bear is just astronomical to achieve the same level of everything that we have today.
- 26:41SPEAKER_01: You know, absolutely astronomical.
- 26:43SPEAKER_01: So in terms of leverage, it's enormous.
- 26:46SPEAKER_01: Um, and we'll talk about the commercial side of it.
- 26:49SPEAKER_01: I've seen companies really wield this well.
- 26:52SPEAKER_01: Uh, we see a lot of it in the Linux Foundation all over, whether it's in the CNCF or the PyTorch Foundation or now the AAIF and other places.
- 27:00SPEAKER_01: Um, companies that really understand open source well, who will get in, who'll open source everything they have, they'll use the community as a lead generator, as a marketing machine, as a way to speed up their development, to build more secure software, all of this stuff.
- 27:16SPEAKER_01: And as a result, monetize it really well.
- 27:18SPEAKER_01: Because at the end of the day, people want the trust and the transparency of open source, but the practical bits of actually running it, no, they'd happily pay someone to do it.
- 27:27SPEAKER_01: Red Hat made a ton of money doing that.
- 27:29SPEAKER_01: It became a$40 billion company when they sold to IBM,$39 billion or whatever.
- 27:34SPEAKER_00: Every time I hear this um kind of analogy, I always think I get this like Bruce Lee meme in my head of like be like water, my friend.
- 27:40SPEAKER_00: I don't know if you've ever like seen that video of him, but he's like, be like water, my friend.
- 27:44SPEAKER_00: It's like exactly that.
- 27:45SPEAKER_00: Like it's so counterintuitive to like some traditional like NBA, you know, who's like, we gotta have this IP, you know, we gotta protect our IP.
- 27:52SPEAKER_00: It's like, no, you have to give your IP away as much as possible, and then all the money will come your way if you just do it right, you know, it'll it'll all come to you.
- 28:00SPEAKER_01: Um and I'm seeing newer companies who understand it as well, like folks in the agentic AI space who are building agentic infrastructure, new companies, startups who are you know um under a year old and stuff, and they get it.
- 28:13SPEAKER_01: They are using open source well, they're using it as lead gen, they're using it to grow their businesses quickly because they know the demand is there, and they're using it as a platform for the world to see what they're doing and to trust what they're doing.
- 28:25SPEAKER_00: So we're getting towards the top of the show, and I always ask this question what have I not asked you that I should have asked or that you wanted to talk about?
- 28:35SPEAKER_01: Um that's a tricky one.
- 28:37SPEAKER_01: Um well, this is with with only a few minutes to go, I'm gonna try and narrow this down just a little bit.
- 28:43SPEAKER_01: What I am gonna do is I'm gonna talk about community.
- 28:47SPEAKER_01: It's a big part of open source.
- 28:49SPEAKER_01: I think it's a big part of agentic AI right now, because there's so much stuff happening in AI, happening at the speed of AI, which is which means that, like, you know, traditionally we we meet online, we do things on the internet, like any open source community.
- 29:03SPEAKER_01: I think there's a ton of value in doing things in person just because of the fidelity.
- 29:08SPEAKER_01: You get so much more throughput.
- 29:10SPEAKER_01: Uh, and as a result, and just given the pace of how things are changing and moving, that throughput's kind of important.
- 29:17SPEAKER_01: Um, so what I have seen is an explosion in in-person events around AI.
- 29:21SPEAKER_01: And that's great.
- 29:22SPEAKER_01: I love that.
- 29:23SPEAKER_01: So I'm going to a lot of those.
- 29:24SPEAKER_01: I encourage people to join a lot of those.
- 29:27SPEAKER_01: We are running a lot ourselves within the foundation.
- 29:29SPEAKER_01: We do have our flagship events around the world.
- 29:32SPEAKER_01: We also have local chapters and local meetup groups everywhere.
- 29:35SPEAKER_01: Um, please reach out if you want to start one.
- 29:37SPEAKER_01: We have ambassadors around the world for the Agentic AI Foundation.
- 29:41SPEAKER_01: Um, if there isn't one in your city or your location, reach out to me.
- 29:45SPEAKER_01: I'm happy to connect you.
- 29:46SPEAKER_01: I hope we'll create one, right?
- 29:48SPEAKER_01: Um, there's so much demand for this stuff.
- 29:50SPEAKER_01: And I'm seeing so much energy in the room when people meet in person.
- 29:54SPEAKER_01: Great ideas sparking off.
- 29:55SPEAKER_01: Um, people quitting their jobs because of an interesting conversation and starting a company because that's a cool idea.
- 30:01SPEAKER_01: They want to go do it now.
- 30:02SPEAKER_01: And now you can.
- 30:03SPEAKER_01: The barriers to entry are so low.
- 30:05SPEAKER_01: And I just love the cut the space that we are in right now.
- 30:08SPEAKER_01: Um, another plug will be a little bit of a shameless plug, if you don't mind, for AgentCon, which is our flagship conference.
- 30:14SPEAKER_01: It's happening in San Jose in a couple of weeks.
- 30:17SPEAKER_01: Um, I would love for folks to come there.
- 30:19SPEAKER_01: We've got three and a half thousand people attending, the people from Anthropic, OpenAI, Google, um, lots of really great content, um, lots of workshops, lots of breakouts happening, just great chances to meet people in the in the community, in the industry and see what everyone's up to.
- 30:33SPEAKER_00: Look, I really appreciate you taking the time to come on today and just love the work that you're doing.
- 30:39SPEAKER_00: And I look forward to collaborating uh more in the future.
- 30:43SPEAKER_00: Thanks for taking the time.
- 30:44SPEAKER_00: Likewise.
- 30:44SPEAKER_01: Thank you for hosting.