#361: Foxglove CEO Adrian Macneil on the Infrastructure Layer Powering Physical AI

So prior to Foxglove, was everyone building their own platforms internally?

Pretty much.

Yeah, exactly.

This was my sort of thesis going into it was that we built this entire platform at Cruise, but we brute forced it, right? We raised $10 billion and we had thousands of people.

And I think about this as the infrastructure problem is the bottom of the iceberg in any autonomy company.

Any Cruiser, Waymo, Tesla, you have a small number of highly skilled, highly paid researchers that are coming up with novel models and things and then novel architectures.

And then you have this enormous corpus of people that are dealing with data infrastructure and fleet infrastructure and assigning rides to people and simulation infrastructure and ML training infrastructure.

and so all of this whole kind of platform is the the machine that builds the machine and so you know that was what we had to build sort of from scratch at cruise figuring out from first principles because there wasn't really a playbook um and you know towards the what really fed my thesis at foxgov was going out and talking to people at waymo at tesla at zoox at aurora you know talk to all the other av companies and you compare notes and you're like oh, wow, you guys built a multimodal data visualization tool.

That's really interesting.

And it looks almost exact, you know, without any, you know, compare notes like two, three years into it, you're like, hey, we built the exact same thing.

That kind of tells me that there's a product here.

And, you know, sure, maybe the top five AV companies can afford to raise billions of dollars and build it themselves internally.

But the industry as a whole is not going to be successful if the like bar to entry is raising billions of dollars.

But also, it's not just the replicating the same work where the inefficiencies or where the opportunity is created, right? It's also, you know, all of these companies, whether it's Waymo or Crude, like they were able – they had the type of resources, investment, capital to put into these things that – when you're talking about people who are operating real robots in real functioning businesses aren't going to have that.

And so creating, not just eliminating the replication, but creating an efficient system is sort of key to this.

Is this where MCAP comes in?

MCAP, yeah.

Can you just tell

folks a little bit about that? Because it feels like that's sort of where the rubber hits the road on this

model.

Yeah, MCAP's a fun one.

So MCAP is actually a logging file format that we created in the early days of FoxGlove.

So I... we left with this grand vision of building developer infrastructure and building the data systems.

And I got out there and started talking to more companies.

And we realized that most of the companies we talked to had actually all of their logging was like, hey, guys, logging data.