What is an autonomous company? How AI agent teams actually run the work (and where humans stay in charge)

What does "autonomous company" mean?

An autonomous company is organized so that AI agents can carry work from start to finish without a person moving every step along. A person still decides what the company is trying to do. Agents, alone or in teams, do the work toward it, hand it on to the next team, and stop to ask when a decision needs a person.

It is useful to think of it as levels rather than a label:

I'd be wary of anyone selling level 4 for a real business. Agents still make confident mistakes, and a company has customers, money and legal obligations that someone has to answer for.

How do AI agent teams actually run the work?

A single agent is a model wrapped in a harness: tools, context, memory, permissions and a loop. LangChain sums it up as "Agent = Model + Harness". An agent team is several of those working toward one outcome, each with a role. A typical loop looks like this:

1. A signal comes in. A metric drops, a customer complains, a goal is set.
2. A team picks it up. For example, a discovery team of a research agent, an analyst agent and a person who owns the result.
3. They do the work. Agents gather evidence, compare options and draft a recommendation.
4. They hand it on. The recommendation goes to the team that will act on it, with the evidence attached, not as a one-line message.
5. A person approves what matters. A pricing change or a public announcement waits for a yes.
6. The result feeds back. What happened becomes part of the company context the next piece of work starts from.

None of the steps is magic. What makes it work, or not, is everything around the agents: whether they know what the company is trying to do, whether handoffs carry the context, and whether the line between "go ahead" and "ask first" is written down.

What does a company need before agents can run work?

• Shared company context. Goals, principles, customers, products and past decisions in one place every agent and person works from. Without it, every run starts from zero.
• Clear ownership. Persistent teams that own outcomes, so work does not disappear after one prompt.
• Structured handoffs. When work moves on, the reasoning and evidence move with it.
• Written boundaries. What runs on its own, what needs approval and what only a person decides, agreed before anything runs.
• A record. Who did what, what was approved, and what it cost, so you can fix the process instead of guessing.

Together these make what we call a company harness: the same idea as an agent harness, at the size of a company.

Where do humans stay in charge?

The useful question is not "how much can we automate?" but "what should each kind of work require?" A simple split:

• Runs automatically: low-risk, reversible, checkable work. Researching competitors, summarizing feedback, preparing a weekly report.
• Requires approval: anything public, costly or hard to undo. Publishing a campaign, changing prices, merging code to production.
• Human only: decisions that define the company or affect people directly. Setting the vision, sensitive people decisions, accepting legal commitments.

Start with more in the approval lane than feels necessary, and move things to "runs automatically" as you see the results. Moving work the other way after a mistake is much more expensive.

What are the risks of an autonomous company?

• Errors compound. One agent passes a wrong assumption to the next, and by the end nobody can see where it started. Handoffs with evidence attached make this visible.
• Accountability blurs. "The agent did it" is not an answer a customer or a regulator accepts. Every outcome needs a person who owns it.
• Costs creep. Agents that loop or redo work spend real money. You need to see AI cost next to the work it produced.
• Context drifts. If the company context is out of date, agents confidently work toward last quarter's goals.

How do you start?

1. Write down your company context: goals, customers, principles and recent decisions.
2. Pick one outcome a team of agents could own, such as a weekly competitor and customer feedback brief.
3. Decide the three lanes for that outcome before anything runs.
4. Review the results every week and update the context with what you learned.
5. Add the next outcome only when the first one runs without you chasing it.

If you're a small business deciding what to hand off first, we wrote a more practical guide: AI agents for small business.

Where does Koru Squads fit?

Koru Squads is the company harness for humans and AI that we're building now. It is in development and not available yet. The idea: start with a goal, then build the squads to run it.

A squad is a persistent operating group made up of whatever the outcome needs: humans, agents, or both. Squads work from shared, evolving company context, own outcomes, hand work to each other with the context retained, and come to you at clear human checkpoints. We are designing controls that separate what runs automatically, what requires approval and what is reserved for people, and a view called Pulse to show progress, handoffs, approvals, bottlenecks and AI cost across squads.

So we'd describe the goal less as an autonomous company and more as autonomy where you want it and control where you need it. You are not managing a collection of AI bots. You are building a team.

If that's how you want to run your company, join the Koru Squads early access waitlist. We're bringing early users in gradually, and joining early is a way to help shape how it works.