The question I keep returning to is how large distributed autonomous systems — just like civilizations — collaborate. It’s a problem I’ve been working on for three decades, building infrastructure at scale, from the cluster systems that trained early transformer architectures at Google, to the Internet of Agents and the entanglement-based Quantum Internet at Outshift by Cisco. The rest of my time is split between photographing the world, reading Durant (it’s a tome), and awkwardly generating noises from my guitars.
Distributed Superintelligence
Internet of Cognition // link
70,000 years ago, the invention of language ushered in collective intelligence for humans, allowing us to share intent, share knowledge, reason together. We are at a similar inflection point for multi-agent systems. Agents today can connect with each other, but they aren’t able to think together, yet. They cannot share intent, coordinate to solve problems or create emergent behaviors. The Internet of Cognition is the platform that allows agents to think together.
Internet of Agents // link
The current state of the art, allowing agents to connect to each other: discover, get access, connect, across vendors, frameworks, clouds. Co-founded AGNTCY — now at the Linux Foundation with 80+ partners including Google, Dell, Oracle and RedHat, and formative partners for the A2A project and Agentic AI Foundation in LF.
Quantum Computing
Quantum Internet · Security · Networks // link
Distributed quantum computing via networked QPUs, quantum-safe cryptography, quantum data center interconnects, and post-quantum security. Overseeing Cisco’s entanglement chip (200M photons/sec) and Santa Monica quantum lab
Coherent Cognition / Substack
Essays on interesting problems in scaling-out infrastructure for non-deterministic systems // LINK
More recent is not the same as more true. More frequent definitely isn’t. Agents need to understand what our collaboration systems barely represent: authority and state.
As agentic systems shrink teams toward one, the coordination surface explodes toward many. And most operating models aren’t even thinking about this problem.
Specialized intelligence should be deployed as a society. Yet most multi-agent systems still call themselves swarms — because ants, apparently, have better branding than employees.
Superintelligence will not arrive as one mind that knows everything. It will arrive as a distributed set of superhuman specialists that need shared intent, shared context, and shared reasoning to act as one. Yann LeCun’s paper builds half the case for this and stops there.
ON THE EAR
🎼 Under Darkening Skies, Monolink, 2020 Embassy One
📢 Vikings, Ragnar, Berserkers, Valhalla & the Warriors of the Viking Age / Lars Brownworth, Lex Fridman Podcast
ON THE ROAD
ON THE SCREEN
AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society
A large-scale LLM-driven social simulation platform that models thousands of agents and millions of interactions to replicate and experimentally study complex social phenomena such as polarization, misinformation spread, policy impacts, and external shocks.