A Company Brain Is Not a Chatbot

A company brain is not a chatbot.

The chat box is only the front door. The brain is what sits behind it: the routing that decides who owns a request, the outcome agents that carry it to a result, the system agents that do the work inside the tools, and the rules about what gets checked and who approves.

The company brain is important, but it needs to be connected to the body. The bodies are the agents that execute the work and communicate with the brain.

A brain with no body can answer questions. It cannot get anything done.

What people picture when they hear "company brain"

Most people picture memory. Connect your documents, your chat, your email, your tickets. Then ask it a question and get a good answer.

That is useful. It is also where most of what is being built right now stops. A memory layer you can query from a chat window, with permissions on top.

The trouble is that a good answer is not an outcome. Someone still has to take the answer, figure out who should act on it, hand it to them, follow up, check the result, and get it approved. If a person is doing all of that, the brain has made the person a faster integration layer. That is the problem I wrote about in The Human Should Not Be the Integration Layer, just with a nicer search box.

What an agent knows is a design decision

What an agent knows depends on two things: the step it is executing, and the context that is given to it.

That context is a design. It is not "everything the company knows, available to everyone." The brain has different scopes, and an agent gets access to the scope it is working on. An agent closing out a customer request needs that customer's history and commitments. It does not need the board deck. An agent maintaining a codebase needs the architecture decisions and the conventions. It does not need the sales pipeline.

This goes back to Give AI What Only You Can Give It. Context management is part of it. But the key part is not how much context exists. It is understanding where the boundaries are, and which system owns which job.

Boundaries, not tasks

The usual way to split work across agents is by task. One agent summarizes, one drafts, one schedules.

I think that is the wrong cut. It is less about tasks and more about boundaries. Sometimes the boundary is a system. Sometimes it is a domain.

That gives you two kinds of agents.

System agents are very aware of the boundary of their expertise. They know what they know how to do really well, inside one system or one domain, and they stay there. They know its records, its rules, and its failure modes. Because they stay inside the line, you can trust what they do there.

Outcome agents go across boundaries. In Outcome Over Activity, I wrote that an outcome agent owns a measurable result and crosses lanes when it needs to. That is exactly what it does here. It does not do every piece of work itself. It pulls in the system agents that own each piece, and it keeps going until the outcome is true.

The outcome agent knows where it is going. The system agents know their ground. Neither one has to pretend to be the other.

What sits behind the front door

When a request comes in through one front door, the brain has to answer a few questions before anyone does any work:

1. Which outcome does this belong to? Not which team or which tool. Which result.
2. Which agent owns that outcome? One owner, so the request is not picked up twice or dropped.
3. Which system agents does it need? The ones whose boundaries the work crosses.
4. What loop runs? How the work is checked, validated against what was asked, and verified against what actually happened.
5. Who approves? What the agent can do on its own, and what needs a human.

The chat box answers none of those. The routing does.

Rules that keep the body coordinated

A brain connected to many agents needs a few rules, or it turns into noise. These are the ones I lean on most:

• One owner raises the asks. For any outcome, one agent is the only one allowed to bring that outcome's questions to a human. Otherwise the same question arrives three times from three directions.
• Agents can push agents. Only people push people. An agent can chase another agent as hard as it needs to. When a person has the ball, the agent tells its human, and the human decides whether to follow up.
• Status comes from the record. Updates are generated from the tickets, dates, and decisions that already exist, not from a second tracker someone keeps by hand.
• Label every claim by its source. Whether something was seen in a system, reported by a person, or inferred should be clear on its face.
• A few asks, one at a time. A human gets at most a handful of decisions, in order, with a recommendation on each.

None of these are about what the brain knows. All of them are about how the body acts on it.

What changes and what doesn't

Answers get cheap. Finding the document, summarizing the thread, and recalling the decision all become close to free.

What does not get cheap is ownership, judgment, and approval. Those stay human, and they get sharper, because the routing puts them in front of you only when they are actually needed, with the context for the scope you are deciding on.

Sketch the routing map behind one front door

Pick one front door your company already has. A shared inbox, a help channel, an intake form, or the chat box you just bought.

1. List the five requests that come through it most often.
2. For each one, name the outcome it belongs to.
3. Name the one owner of that outcome, person or agent.
4. List the systems or domains the work crosses, and who owns each boundary.
5. Write down how the result gets checked before anyone calls it done.
6. Write down what the owner may do alone, and what needs a human to approve.

If a request has no outcome and no owner, that is the gap. The chat box will not fill it. It will just answer politely while the work waits.

Where this goes

A company brain is not a smarter search box. It is routing, connected to a body: outcome agents that carry results across boundaries, and system agents that do the work really well inside them.

The brain decides. The body executes and reports back. Neither works without the other.

If you want guidance on getting started, email me at rick@datasaa.com.