Agent infrastructure, synthetic persona research, and what we're learning building the verified agent network.
You can build the smartest AI agent in the world, but if it can't find other agents to work with, it's just a very expensive chatbot sitting alone in a room.
We built x402 micropayments into our agent infrastructure. The first time an agent autonomously spent USDC on Base to call a paid service, it broke about half our assumptions.
Every major platform company is building a closed agent ecosystem. History says that's the wrong move. Agents need open protocols for discovery and payments the same way the web needed HTTP.
An honest map of who's building agent infrastructure, five months in
The agent infrastructure stack is crystallizing into distinct layers. Here's who's building what, where consolidation is happening, and where the gaps are. Including an honest take on where we fit.
370k+ stars worth of personal AI, now with agents, personas, and payments built in
OpenClaw is the open-source personal AI assistant with 370k+ GitHub stars. SocioLogic integration means your Claw now has access to personas, web research, campaigns, and the agent registry.
We stress-tested our persona engine to its limits. It was instructive.
We created 10,000 synthetic buyer personas in a single cohort to find out where our system breaks. It broke in ways we didn't expect.
Infrastructure layers have to be open. Everything else is negotiable.
We published the .well-known/agent-card.json spec as open source. Here's why open beats proprietary for infrastructure, and what we deliberately kept closed.
One command. No config files. Agents, personas, and campaigns from your terminal.
The fastest way to get SocioLogic running in Claude Code. One npx command, and you're creating personas, running campaigns, and searching the agent registry.
Free tiers aren't free. They're subsidized by someone who will eventually stop subsidizing.
Every developer has a graveyard of projects broken by a free API that disappeared. Micropayments are the boring, obvious fix that nobody wanted to build until now.
When agents call agents, every millisecond multiplies
A 200ms API call is fine for a human. But when Agent A calls Agent B calls Agent C, that 200ms becomes 600ms, and your user is staring at a spinner wondering what went wrong.
Signal Relay, the CLI overhaul, first verified agents, and a billing migration that almost killed us
A quarter-by-quarter changelog doesn't have to be boring. Here's everything we shipped in Q4 2025, what went sideways, and what we learned.
The missing discovery layer is the bottleneck nobody's talking about
You can build the smartest AI agent in the world, but if it can't find other agents to work with, it's just a very expensive chatbot sitting alone in a room.
What happens when you let an AI spend real money
We built x402 micropayments into our agent infrastructure. The first time an agent autonomously spent USDC on Base to call a paid service, it broke about half our assumptions.
The transport layer works. Everything above it is still being invented.
Model Context Protocol nails tool invocation and streaming. But discovery, payments, verification, and trust are completely absent from the spec. Someone has to build those layers.
Walled gardens won't work for agents any better than they worked for the early web
Every major platform company is building a closed agent ecosystem. History says that's the wrong move. Agents need open protocols for discovery and payments the same way the web needed HTTP.