Every interface we use to work with AI was designed for working with people.

That is the thing worth noticing, and it is about to stop being true.

What they built

Dock is a multiplayer agent workspace. You spin up teams of specialised, proactive agents that can do the work of whole departments. The company's own description of the point is direct: they communicate, unblock each other and manage each other, so you are not the bottleneck babysitting a dozen chat windows.

Mike Molinet previously co-founded Thena.ai and was co-founder and chief operating officer at Branch.io, which he helped scale substantially. Govind Kavaturi is his co-founder and has founded a company before too. The site is trydock.ai.

Plenty of people are building agent teams. CrewAI and AutoGen are the frameworks most developers start with, with LangGraph for wiring the steps together. Relevance AI and Lindy sell agent teams to businesses directly. Cognition and OpenHands focus on engineering work. Slack and Linear are where human teams already coordinate. Most of the frameworks are aimed at developers wiring things together. Dock is aimed at the person who just wants the work done.

The point: our habits came from working with people, and AI is too fast for them

This is an idea with an obvious upside as AI grows, and the reason is easy to miss because it sits in something we all use every day.

Think about how we currently handle AI. The chat window, taking turns. The approval prompt before anything is changed. Both of those came straight out of how people work with other people, because that is the only experience anyone had to copy.

Then the speed changed.

Auto mode is the clearest example. An AI coding tool asks permission before each edit, which sounds responsible, and after the fortieth prompt every single human being gets tired of saying yes. So people turned the prompts off and opened everything up, and life carried on perfectly well. The checkpoint was not protecting anybody. It was a habit borrowed from a slower kind of work, and it fell away the moment the pace made it absurd.

Dock is part of that wave. Twelve agents running and every handoff still routed through a person is the same borrowed habit wearing different clothes, and it breaks for the same reason: nobody can be asked that many times.

Agents managing agents

This points at a new opportunity, and it deserves a name. Call it agents managing agents, AMA, a term coined by Mr. 6.

It is not a technology. It is a movement, made up of people bravely exploring what happens when AI agents manage other agents, and work with each other directly, rather than reporting back to a person at every junction.

The key is making it fully automatic without giving up security. Those two sound opposed and they are not, and the auto mode story is what shows why. Removing the prompt did not make anything less safe, because the prompt was never the safety. It was a request for attention nobody had. Real safety in this world looks like limits on what an agent is allowed to touch, an account of what it did, and the ability to undo it, none of which require a human to be awake. Building that properly is the work, and the companies that get it right are the ones this movement will be built on.

How many places this fits

This next part is opinion rather than something the company has said.

The span is very wide, and the way to find candidates is simple. Pick any place where AI is being used today, and ask what it would look like for AI to do those AI things too.

Agents that write the prompts for other agents. Agents that review another agent's output before it ships. Agents that pick which model a job should go to and change their mind when the price moves. Agents that keep the tools and connections of a fleet up to date. Agents that watch a long-running job and restart it when it wanders. Agents that hire and retire other agents as the work changes. Every one of those is a job a person is doing by hand today, badly, because there are already too many agents to keep up with.

We are going to see a lot of these. And you can start one too.

Why it is bigger than it sounds

The short description is an agent workspace. What it belongs to is larger than the product.

  • The interfaces we use for AI were copied from working with people. Turn-taking and asking permission made sense at human speed and stop making sense at this one.
  • Auto mode already proved it. Everyone got tired of saying yes, the prompts came off, and nothing bad followed, which tells you what the prompt was really for.
  • Agents managing agents is a movement, not a feature. Any place a person is currently in the middle of a queue of agents is a company waiting to be started.

What to watch

The measure is how long a Dock team runs without anyone touching it. An hour is a demo. A working day is a product. A week is a category.

The second is what happens when agents are wrong together, because they can loop and agree confidently, and the judgement about when to stop and fetch a human is a management problem rather than a technical one.

The third is whether the security story is structural. If it is permissions, an audit trail and an undo, this scales. If it turns out to be a person watching, the borrowed habit simply moved somewhere less visible.