We believe the next generation of companies won't just use AI — they'll be built for AI.
These aren't programs or mandates — they're the questions we're most excited about right now. Open invitations to explore what the agent-native era makes possible. If you're chasing one of them, we want to talk.
Token-native economies
What becomes possible when value, ownership, incentives, and coordination are all programmable?
Every internet-native network eventually grows its own economic layer. We want the teams treating tokens not as a feature but as the settlement rail for machine-to-machine value — agents that hold, earn, and spend; networks that coordinate without a company in the middle.
Software built for agents
What does software look like when the reader is a machine, not a person?
The next billion users won't have hands. We want products legible to agents by default — machine-readable interfaces, live documentation, execution over UI — and the picks-and-shovels beneath them: inference silicon tuned for agentic loops, memory that persists across a thousand steps, the identity and payment rails an autonomous workforce runs on.
Self-improving systems
What compounds when every execution loop is also a training loop?
The most valuable companies won't just automate work — they'll get measurably smarter each time they run. Closed-loop discovery that proposes its own experiments. An operating layer that makes a whole company queryable and rewrites its own workflows. Show us the system where doing the work is how it learns.
Robotics & the physical stack
Why can AI reason about the world but still can't load a dishwasher?
We may reach recursively self-improving software before a robot that clears a table. The gap is infrastructure, not intelligence. We're excited by the layer that closes it — world models that transfer to messy environments, sensors and APIs that give software hands, the hardware supply chain and safety rails that let capable machines ship into the real world.
The next billion users
may not be people.