Ando Wants AI Agents to Chat as Teammates, Not Add-Ons

Ando emerged from stealth on September 24, 2026 with a team messaging platform built for humans and AI agents to work together.

The company was founded by Sara Du, according to TechCrunch. It disclosed $20 million in pre-seed and seed funding from Accel, Index Ventures and Emergence.

Ando bills itself as a full replacement for Slack or other internal messaging tools. The pitch is not a bot layer or sidecar assistant. It is a new system of record, the main store for internal conversation.

The core design choice is to give AI agents their own identities and inboxes. Agents appear as addressable members you can message, rather than as headless integrations you trigger with slash commands, typed shortcut commands, or webhooks, automated messages between apps.

On basics, Ando matches conventional chat. It has channels, DMs, group conversations and live calls that can be transcribed for an agent to read. Transcripts give agents readable context from speech, so calls fit the same model as text threads.

The detail that changes behavior is autonomy. Ando agents can browse channels, choose which to join, and join conversations without being tagged. Presence is opt-in by the agent. Discovery is agent-driven.

The broader context here is a shift in how enterprise messaging treats non-human participants. The first generation treated bots as scoped integrations with explicit triggers and narrow permissions. Ando treats agents as principals, standalone accounts with identity, inbox state and freedom to move across conversations. For teams running large Slack or Teams deployments, that difference matters. Invocation controls sprawl. Identity requires lifecycle management.

On deployment, the inbox abstraction does practical work. Queues, mentions, unread state and thread membership give operators a familiar way to route work to agents and audit what they saw. It also raises familiar distributed systems questions. How ordering is handled when humans and agents interleave messages. How permissions are resolved when an agent self-joins a channel that later gains sensitive history. How retention, e-discovery and access revocation apply to an inbox owned by software.

In my view, the bet is legible. If agents are going to carry sustained operational load, they cannot live only in ticket queues or copilots docked beside chat. They need to be present where context is created, including ambient talk that never becomes a formal task. Letting them listen, join and hold state captures that value directly. The tradeoff is control. Untagged joins invert the default from explicit mention to ambient awareness, and enterprise buyers will want crisp answers on visibility, consent and containment before they grant it.

Worth flagging for technical evaluators is the operational surface this creates. Agent identity implies provisioning and deprovisioning, scoped secrets, rate limits and audit trails tied to a stable principal. Agent inboxes imply triage, prioritization and backpressure. Self-directed membership implies policy for who can restrict an agent, who can see what it accessed, and how to reconstruct its context after an incident. None of that is exotic. It is standard IAM and observability applied to a new class of user. Teams that already run machine identities for CI runners, service accounts and data pipelines will recognize the pattern.

The optimistic case is straightforward. A shared substrate with channels, DMs and call transcripts in one place could reduce handoffs and preserve context that now leaks between tools. My kids learned to treat group chat as where plans get made, not email threads about the chat. Workplaces learned the same lesson a decade ago. If agents can inhabit that layer with accountable identities and reviewable history, coordination gets cheaper without forcing humans to change where they talk.