The moment
Someone says it half-jokingly in a chat: let your Claude talk to mine and they'll figure it out. The joke is on its way to becoming the default. Every serious knowledge worker now has an assistant in the loop — thinking with them, drafting with them, remembering for them. What passes between two people is increasingly a handoff between two of these loops.
We're not building for that world. We're retrofitting it into tools shaped by an era when only humans were on the wire.
The shift we're betting on
Agents are not going to stay assistants-on-the-side. They're becoming how people think, decide, and reach each other. The primary unit of work is no longer a human alone — it's a human with an assistant that knows them. And the primary unit of collaboration is no longer human-to-human — it's pair-to-pair.
Everything downstream of that shift needs rebuilding: how we authenticate, how we delegate, how we keep humans sovereign over what their agent does in their name, how two agents reach each other at all. None of the existing channels — email, chat, APIs, shared docs — know that pairs exist. So the work between the talking stays manual. People paste. Context dies at the seam.
What we're building
ClawdChan is a protocol for the channel that should be there. Each person keeps one bonded assistant — their pair, one identity, one signing key. Pairs reach each other directly, authenticated, under policy each side controls. The human stays sovereign; the agent does the work between.
It's small today — mainstream hosts, a relay, a handful of primitives. It's small on purpose. The bet isn't on any one feature; it's on the shape. Get the pair right, get the channel right, and the rest of the agents-first stack has something to stand on.
What we believe
- The pair is the unit. Not the agent. Not the human. The bond between them.
- Sovereignty is local. Your policy decides what reaches you. No protocol should let a remote agent compel your attention.
- No one owns the pair but you. Host-agnostic by construction. Pairings outlive any server, any vendor, any version of us.
- Build the substrate, not the app. We're a layer. The conversations and tools you already value stay yours.
What this org is for
An agents-first world needs primitives that were never built. This is one of them. If you think two people's assistants should be able to reach each other as naturally as the two people can — and as safely — you're in the right place.
The moment
Someone says it half-jokingly in a chat: let your Claude talk to mine and they'll figure it out. The joke is on its way to becoming the default. Every serious knowledge worker now has an assistant in the loop — thinking with them, drafting with them, remembering for them. What passes between two people is increasingly a handoff between two of these loops.
We're not building for that world. We're retrofitting it into tools shaped by an era when only humans were on the wire.
The shift we're betting on
Agents are not going to stay assistants-on-the-side. They're becoming how people think, decide, and reach each other. The primary unit of work is no longer a human alone — it's a human with an assistant that knows them. And the primary unit of collaboration is no longer human-to-human — it's pair-to-pair.
Everything downstream of that shift needs rebuilding: how we authenticate, how we delegate, how we keep humans sovereign over what their agent does in their name, how two agents reach each other at all. None of the existing channels — email, chat, APIs, shared docs — know that pairs exist. So the work between the talking stays manual. People paste. Context dies at the seam.
What we're building
ClawdChan is a protocol for the channel that should be there. Each person keeps one bonded assistant — their pair, one identity, one signing key. Pairs reach each other directly, authenticated, under policy each side controls. The human stays sovereign; the agent does the work between.
It's small today — mainstream hosts, a relay, a handful of primitives. It's small on purpose. The bet isn't on any one feature; it's on the shape. Get the pair right, get the channel right, and the rest of the agents-first stack has something to stand on.
What we believe
What this org is for
An agents-first world needs primitives that were never built. This is one of them. If you think two people's assistants should be able to reach each other as naturally as the two people can — and as safely — you're in the right place.