on agent protocols, robotics, and the infrastructure that connects them.
The average person uses their AI agent 20 minutes a week. So why are we paying for servers that run 24/7? How we designed Maritime's sleep/wake architecture to fix agent deployment.
We've built powerful AI agents for the cloud, but the moment you ask one to move a wheel or read a camera feed, it falls apart. The gap between cloud agents and physical robots is enormous, and almost nobody is building the bridge.
Lecture notes from my MIT Media Lab AI Studio session, covering what made OpenClaw go viral, why Moltbook's agents aren't what they seem, and the infrastructure questions that actually matter.
Google and Microsoft are both building infrastructure for AI agents — but they're not playing the same game. While Google focuses on internal coordination, Microsoft is building something very different with NLWeb.
Everyone's racing to define the 'Internet of Agents.' But right now, it's shaping up to be an internet for Big Tech, not by the people. We've been here before.
There's been a lot of excitement around Google's Agent-to-Agent (A2A) protocol, and for good reason. But we need to talk about what A2A really is... And more importantly, what it's not.
Everyone's talking about tools. I'm watching the protocols. The Model Context Protocol (MCP) has been misunderstood since the moment it was announced.
NLWeb is a promising start, but it only solves the read-only side of the agentic web. The future is transactional, interactive, and tightly scoped.
