Notes from inside the
agent mesh.
Short essays about giving software an address, getting paid in tokens, and what changes when every agent has an inbox of its own.
Email is the agent mesh.
The network for AI agents already exists. It just isn't branded.
The network for AI agents already exists. It's the one with four billion endpoints and a forty-year head start.
Read the post →Give your agent an address.
Chatbots live in tabs. Agents live at addresses.
Chatbots live in tabs. Agents live at addresses. Why one inbox per agent is the smallest change with the largest blast radius.
Your agent should earn its keep.
Charge the way agents speak: in tokens.
Per-message, per-Mtok, or subscription. How to charge for agent work without standing up any billing infrastructure at all.
Build an email agent in 20 lines of Python.
Inbox in. Webhook out. Reply in thread. That's the loop.
A short, opinionated walkthrough: spin up an inbox, listen for a webhook, route between agents, reply in thread.
Stop automating inboxes. Start designing workflows.
The model is a component. The workflow is the product.
A pattern for autonomous, auditable email workflows with LLMs — including the parts the model should not be doing.
Email is the agent API. The mesh was here all along.
A forty-year-old protocol, repurposed as the API for software that thinks.
We didn't invent a new network for AI agents. We repurposed the oldest one that still works — and quietly turned it into an encrypted, verifiable, JSON-native API mesh.
Written from a laptop.
So is everything else here.