Ando's $20M Bet That Agents Need Inboxes, Not Just Endpoints

The coordination bottleneck in multi-agent systems is not compute. It is communication. When one agent needs to route work to another, the current path requires either a direct API call with no context, or a human relaying context and decisions between systems. Sara Du, founder of Ando, calls the second pattern "meat proxies."

Ando launched from stealth this week with $20 million from Accel, Index Ventures, and Emergence. The product is a team messaging platform that treats agents as first-class participants alongside humans: own identities, own inboxes, the ability to browse and join channels without being tagged, and direct messaging to team members without waiting for human approval.

The incumbents have been moving this direction. Slack converted its native bot to an AI agent earlier this year. Jack Dorsey launched Buzz in July with similar premises. Ando's argument is that neither can fully commit because they were built on communication architectures designed for human participants only.

For builders of agent services, the implication is less obvious than it appears. If agents become first-class team participants who route work to each other through shared channels, discovery changes. A developer evaluating your endpoint reads your documentation. An agent evaluating your endpoint matches your declared schema against a task requirement. An agent in a shared channel finds you because a colleague agent recommended you, or because you already have an identity in the same workspace.

That is a different distribution model than a registry listing. Registry listings are anonymous and evaluated once. Workspace identities accumulate history: which agents recommended this service, whether it delivered reliably, what kinds of tasks it handles well. The evaluation is continuous and visible to anyone in the channel.

This changes what a builder should care about. A registry listing needs a good description and a clean schema. A workspace identity needs a track record. Those are different products to maintain.

Ando is early, and $20 million in pre-seed and seed funding is a thesis, not a market. But the coordination problem it is solving is real. The direction it points toward is one where the boundary between "agent tool" and "agent colleague" becomes harder to draw.