Workers, managers and teams: how Korps organises AI

What this is

Most AI tools give you one assistant and one conversation. You type, it answers, and when you close the window the work is gone. That is fine for a question. It is not how a business runs.

Korps starts from a different idea: an AI agent is a worker. A worker has a name, a job title, a manager, a place on a team, and a record of everything it has done. You can hire several, give them different jobs, and put a manager over them. The result looks like an organisation chart, because it is one.

Three words carry most of the weight:

• Worker. One AI agent with a defined role, such as "Sourcer" or "Reporting". It keeps its identity, its memory and its history even if you later change the AI model or the runtime behind it.
• Manager. A worker whose job is to coordinate other workers: break an objective into tasks, hand them out, check the results, and escalate only the decisions that are yours.
• Team. A manager and the workers reporting to it, usually named after a business function: Recruiting, Sales, Operations, Research.

Who it helps

A business owner or team lead who wants AI to take on recurring work without having to sit in the chat window. If you have ever thought "I would delegate this if I had someone", this is the model for it.

A concrete example

Say you want the month closed and anything odd flagged. On your own, that means a spreadsheet, a stack of invoices, and an evening. With a team in Korps:

1. You give the Operations Manager one objective: "Close September. Reconcile the accounts, chase what is unpaid, and tell me what looks wrong."
2. The manager splits it: Finance Ops reconciles, Vendor Ops chases invoices and renewals, Reporting writes the summary.
3. Each worker reports progress as a state you can see at a glance: working, waiting, blocked, done.
4. Vendor Ops hits a supplier who has not answered. It is marked blocked and goes to the manager, not to you.
5. The manager sends you one thing that needs you: "Approve paying the disputed invoice?" Everything else it handles or reports.
6. You get the report, with a line for every system a worker touched.

The team above is illustrative. The roles are ones you define; the faces are fictional.

Why the manager matters

Without a manager, every worker talks to you, and you become the bottleneck you were trying to remove. With one, the manager absorbs the coordination and you deal only with what genuinely needs a human. Korps shows this as a ladder: one worker, you talk to it; a team of five, you talk to the lead; a whole department, you answer what needs you.

What the record gives you

Every task, message, approval and tool call is written to Activity, which names the system the worker acted in. That is how a worker explains itself, and how you can check what happened without trusting a summary.

Requirements and availability

• Available in Korps. Workers, managers, teams, reporting lines, tasks and Activity are the core of the product.
• A free Korps account is open now. The Korps app for Windows and Android, where you hire workers and organise them, opens at launch.
• A worker needs two things before it can think: a runtime (Hermes Agent comes bundled) and an AI model account you connect. See choosing a model.
• Today workers run on a machine you pair with the app: your PC, or a server you rent. See where the work runs.

Limitations and approval points

• A manager is an AI too. It coordinates; it does not carry legal or financial responsibility. Decisions with consequences should be approval points, which you set. See approvals and staying in control.
• Quality depends on the assignment, the tools you connect and the model you choose. A vague objective produces vague work, exactly as with a person.
• Korps keeps the organisation and the record. It does not make a model more accurate than it is.