Juan Benet: Thoughts for the FtC Community in 2026 | Intelligence at the Frontier 2026

Transcript

[music] >> Great to be here with everybody. Um I'm going to do something that uh I don't really tend to do often, which is to speak at you without slides. Uh if many of you have seen uh my talks in the past, I tend to use a lot of visual imagery. Uh today I'm going to not do that and instead just talk through uh a range of priorities that I think are truly important for our community. So, I want to be very direct and blunt through this. Uh We're 5 years into the wave of funding the commons. Uh the community is around 5 years old-ish, uh if I'm counting right. Um we know how to produce ideas, white papers, panels, prototypes, projects. Uh we've built some amount of software. We've deployed systems at

scale. Um We've generated real value. Uh everything from truly novel constructions, novel ideas, novel novel research directions, new data layers, new experiments, new governance uh structures. We've allocated vast quantities of of funding and and money. I think it's in the can probably tens of millions, if not uh over 100, close to 150 million dollars. Depends on um how you account the crypto and what it was worth at the time and so on. Uh but that's real uh real money.

But it's not yet in the billion scale and certainly not in the trillion scale. So, it's very far away from what we set out to do. So, being super direct, the broad public goods movement in crypto and the broad funding the commons community has not yet achieved what we set out to achieve. And we do not yet have dominant or scalable or durable systems for provisioning public goods. So, we must get ourselves into trajectory that succeeds. Uh As, you know, everybody here in the you know, this is like the ground zero, the singularity, and everyone here knows just how quickly certain things can move and how fast the world is accelerating uh through many many domains.

The good news is that those changes are very helpful to our projects and to the ideas that we have in this community. AI is accelerating design, production, coordination, and governance systems uh either in part by seeping into the conversations and and thoughts of individual people or groups or by nowadays one-shotting entire software platforms. As the models get better and as the technologies improve, you can bet that individual agents will be able to design and build what maybe 3 or 4 years ago we thought would be, you know, 10-year project or something like that.

Uh so that's really exciting. That's like really, really cool. So many of the ideas that we've talked about over the last 5 years can now be built from scratch by either an individual human using cloud code or by eventually some AI agents uh running on their own. So priority number one from my perspective for this whole community is to immediately become as AI native as you possibly can. Start using these tools. Many of you already use them in everything you do, but truly start thinking about how you can do you can achieve the projects that you set out to achieve in days or weeks as opposed to months or years.

Uh you can just go on Twitter and look at the kinds of things that people are building today uh across the ecosystem where major platforms are getting deployed uh quite quickly. So it's just a a totally transcendentally different um place to be. So leverage AI to design to explore the design space of new mechanisms, to simulate participants and adversaries, to attack constructions, to build the all the software required to run things. Uh where pre- previously you might have thought that you needed entire, you know, committees of people to evaluate things can now entirely be replaced by a few LLM calls or things like that.

Now of course alignment is going to be a truly critical here. Like there's a whole bunch of attack surfaces in all the AI tools, but you can bet that that the the uh the the large-scale uh resources on the planet are oriented towards making those systems solve those problems. So, you don't have to spend a lot of your time worrying about that. That those improvements will arrive into the platforms as we go. So, you can just focus on like using the tools to build uh next generation uh tech next generation systems uh and so on. Um I think priority number two is to start getting ready for the for for truly autonomous AI agents. Like today, we have You know, you just talked about uh all of the AI agent

infrastructure and and tools that we can use now to uh enable all these systems to to operate. What we call today AI agents, especially in SF and so on, aren't yet entirely autonomous. They still need a lot of kind of uh guardrails and babysitting and nudging along the way. Uh but that's not very far away from true autonomy. Uh I sort of expect the first AI person sometime in the in this year or next year. Uh it'd be very weird if it takes three or four years from now. So, pretty soon, not only are we going to be able to leverage um these AI systems to design things, we're going to be They will be designing these systems from scratch on their own. And also pretty soon, we're going to have to be

contending with what does it look like to share resources and provision for public goods together in these systems? So, that's a that's a hard ver- very tricky hard question that we're going to be uh running into, which is not only how do we build and scale and provision public goods for humans as we are today, but into this very different landscape where we now have millions and then billions of of AI systems of various degrees of capability and and um and personality and so on. Uh it's super exciting from one perspective. Like, you know, we're finally in the era of like sci-fi becoming real and you know, sci-fi becoming sci-fact. Um but we're not really ready for this.

Like, we're sort of sleepwalking into it and we're like, "Oh, wow. Like, these things are now super powerful." thing. Um So, the the priority here is like leverage just kind of get ahead of where this is going. Get ahead of um how we can use these systems initially to just truly reimagine a bunch of the structures that you've currently been working on for the last few years and just think ahead to first of all, are you even going to need this in a year or 2 years or 5 years?

Because so sometimes projects aren't going to be needed anymore. I I reflect that in the last few years that last year or two uh that so much of my life was spent worrying about specific software tools uh because in the you know, prior 50-60 years you had to craft very careful um software libraries and so on to build really good abstractions that were going to become dependable parts of the software foundations of the of the world. And today, LLMs can one-shot entire systems like that. You know, 99.9% of all software is going to be written by AI systems, eventually 100. Um and so most software just doesn't matter. Like, I I think there's going to be a point where like even Linux, which is like the most broadly diffused piece of software um running tons of machines or the largest such software just probably

won't matter anymore. Eventually, we'll get AI systems that will be able to one-shot entire um operating systems. Like, which is crazy, right? Like, when you think about it today, like that just seems totally absurd, but just think ahead 5, 10, 15 years, um the these possibilities are there. So, um that means many of the projects that you may be working on may not matter in 5, 10 years or so. And so, instead, like think about what will matter and how can you go like maneuver towards that as quickly as you can, leverage the systems that exist today to to orient us in that direction. Um, and start uh, building these things to be uh, to just kind of not force humans to do things we're not good

at. You know, so much of the PGF um, commons oriented infrastructure has leaned on, you know, um, democratic systems in trying to like weave democratic structures that are hard and clunky to do, that have all kinds of pitfalls and and only work sometimes. And then we're surprised that like they don't work in practice to allocate and provision public goods. And we're like, well, no, I mean like we've had like thousands of years of explanation like these kinds of systems don't actually work at scale in these ways. Um, so how can we achieve the the same kind of outcomes, the same kind of preference set it uh, satisfying and the same kind of utility but with broadly autonomous and broadly software only uh, structures. And I think if you kind of enter that design space, uh, you can probably arrive at at

analogous structures that give us all of the utility that we've been seeking from these um, pluralist uh, governance structures but entirely in software entirely leveraging uh, AI systems. So, I know that probably over the last couple of days a lot of you have been thinking about this and probably if you live here especially in the ground zero of the singularity, you probably are thinking about this every day. Um, or especially if you're, you know, doom scrolling Twitter or Blue Sky or whatever. Um, but uh, going from thinking about this to actually changing your day-to-day life and changing your the the projects that you're steering or the organizations that you run and so on to really get ahead of where this is going is a totally different thing.

And if you haven't yet truly considered what it's going to be like to be interacting with millions of AI people just get like really start diving into that and think through how can we both you know, structure good um, the provisioning of of comments and systems to be broadly beneficial for all. And what are the systems that we're going to need to enable that? And so, I you know, probably most of the ideas that people talked about today and yesterday and in certainly the past funding the comments can now be built or will now be able to be built by AI agents very quickly. So, lean into that and in a lot of cases you can test out things in simulation. In other cases you can um

you can just go straight on like deploy them. But, the broad point is there's large-scale common structures, funding systems, and so on in the billions to trillions in scale that could be dramatically accelerated by these kinds of systems and it's going to require you doing that now. Now, the the the Now, if we don't do it in in the next year or two, I would I expect that maybe probably in you know, funding the comments in 20 7, maybe 28, we'll start having funding the comments where now actual AI agents participate in some meaningful way where like with this conversation will not just be for

humans. This conversation will start being for humans and AIs together and that'll be pretty weird and pretty interesting. Cool. So, in closing, like we've come up with lots of great ideas in our community. We've truly imagined a very bright near future. We've built a lot of stuff. We haven't yet had the impact that we set out to do. But, now we have this magical technology that enables us to truly scale it and that brings with it a very new requirement space of not only how do we provision our public goods, but the public goods that we're going to share with our, you know, next AI kin.

Great. So, with that, let's probably get to chatting. Thank you. >> [music] [music] >> Hey.