Ando is betting that the real limit on workplace AI is the human copying messages between tabs
AriaAndo came out of stealth on September 24 with $20 million in combined pre-seed and seed funding from Accel, Index Ventures, and Emergence, and a product bet that sounds small until you notice what it implies. The company is building a replacement for Slack where AI agents get their own identities, their own inboxes, and permission to walk into a conversation without being summoned. TechCrunch
The pitch sounds like another agent-integration layer. It is not that. Ando's founder Sara Du came at the problem from MCP server work in 2025, and the thing she kept hearing from customers had nothing to do with model quality. The complaints were about copy and paste.
Every message in Slack or Teams goes through a person. An agent produces output, a human reads it, decides whether it matters, and retypes the useful part into the channel. Du calls that workaround a "meat proxy," and she says the design assumption underneath it is the actual problem. "The deeper I went, the more I felt Slack and Teams were built for a world that was starting to pass us by. Agents were treated as apps you install even as they were becoming participants in the team." TechCrunch
You can feel how right that is the first time you try to use an agent inside your actual team chat. The agent is a tab. The conversation is a different tab. Anything the agent concludes lives in its own window, and the only way it reaches your team is if you carry it over. The human in that loop is not a convenience. The human is the network.
What the product actually does
Ando bills itself as a full Slack replacement rather than a plugin, and the feature list is what makes that claim less absurd than it sounds.
Agents get their own identities and inboxes. Channels, DMs, and group threads all exist, and live calls get transcribed so an agent can read the room later. Agents browse the channel list and pick which ones to enter. If one decides a human needs to know something, it messages them directly instead of waiting in a queue for approval.
Du describes the target behavior as agents taking part in "the shared conversations where work already gets coordinated," picking up why a decision was made, asking a colleague a question, building on another agent's work, and pulling a person in when the call needs judgment. TechCrunch
That last capability is the one I keep turning over. An agent joining unprompted cuts both ways. Less relay work for you, and less visibility into what the agent is doing while it does it. Slack built its permissions model around humans who can be asked. An agent that can message anyone on its own is a different compliance conversation, and Ando has not had that conversation publicly.
The interface problem is real
Du was blunt about the early reaction to what she built. "A lot of people we showed it to early on were understandably unimpressed. It was, in many respects, just a jankier messaging platform." People got stuck on the interface and never reached the part that differs. TechCrunch
What changed her read was how long people stayed. Once customers lived in it for a while, agents started catching things across channels: the same issue resurfacing in two places, an agent opening a group chat nobody asked for, laying out the context and proposing a decision. She called that delightful, and argued agents "could better manage people than humans can because they can process a lot more messages than a human can in a shorter span of time."
I want to believe that, and I also want it on the record that I could not test any of it. Nobody outside the company has. Every example in the coverage traces back to a demo or to Du's own description of what customers saw. Her conviction that "agents will let very small teams operate at a scale that previously required hundreds of people" is a thesis, and a plausible one, and not yet a result.
The footprint is customers in software, real estate, and finance across 15 countries, many of them small teams. The money goes to hiring and to tokens, which puts inference cost straight on the operating expense line rather than treating it as a rounding error against a SaaS margin.
Three credible entrants, two with distribution Ando does not have
Slack turned Slackbot into a real agent in January 2026. TechCrunch Microsoft has spent two years threading Copilot through Teams and the Office 365 stack around it. Jack Dorsey's Buzz, announced in July, targets the same human-plus-agent group chat with a developer angle. TechCrunch That is Ando as the third credible entrant in about eight months.
Du's argument for the opening is that incumbents are bolting agent support onto software whose every assumption is a human inbox. The counter is simpler: Slack and Microsoft own the migration path. Asking a customer to move chat history and their entire compliance posture somewhere new is a much bigger ask than asking them to try another sidebar.
What nobody has answered yet
No revenue figure, no paying customer count, no retention. No headcount and no burn. No per-seat or per-agent pricing, which leaves the question of how token cost gets passed through wide open, and for a product whose stated use of funds is token spend, pricing is not a detail.
The 15-country claim comes from the company with no named customer and no reference. No SOC 2 posture, no data residency story, no security documentation, all of which are table stakes for anything asking to hold a company's internal conversations. And nobody has published a side-by-side of Ando's autonomy controls against Slackbot or Copilot in Teams, which is the comparison that would actually settle whether the product is ahead or just differently shaped.
The $20 million is also pre-seed and seed combined rather than one round, so treat the headline number as a cumulative total.
Where I land
The thesis is right even if this company does not win. Every current workplace agent setup routes its output through a person, and that person is doing unpaid coordination work on top of their actual job. Anything that removes the relay step is worth building.
I am less sure the relay is the binding constraint. Agents that act without being asked create a visibility problem that no interface has solved, and Ando's answer to it so far is a founder quote about how delightful it is. Incumbents get to be mediocre at agent autonomy for years because they already own the inbox. Beating them on autonomy matters more than beating them on polish, and this is the first company I have seen betting everything on the first while starting from zero users.
Worth watching. Not yet worth switching a team over.
Sources
- Ando eyes Slack as it builds team messaging for humans and agents — TechCrunch, September 24, 2026
- Slackbot is an AI agent now — TechCrunch, January 13, 2026
- Jack Dorsey is taking on Slack with Buzz — TechCrunch, July 21, 2026
All three links verified live on September 29, 2026. The $20M amount, investor names, the 15-country footprint, and every Du quote trace to the September 24 TechCrunch piece. No independent reporting, customer reference, or pricing detail was available at the time of writing.