Where does control sit?

A simplified, scripted illustration of two coordination patterns, not a benchmark or a live swarm. The message counts cover only what crosses a group boundary or comes from the controller. They are not a score, because they count different things.

"Federated swarm" and "hybrid" get used as if they mean the same thing. They don't. The difference is where control sits, and how much of it you are willing to let go of.

Both are groups of agents. Both can look busy and impressive from the outside. But one is a partnership between equals, and the other is a frame with freedom inside it.

On this page, federated means peer-to-peer: groups keep their own ground and share conclusions sideways, with no central coordinator. (In federated learning the usual set-up has a central server that combines updates. The peer-to-peer version is the one we mean here.) Hybrid means one controller sets the boundary and the agents coordinate inside it. Some writers use 'hybrid' for any mix of central and peer coordination. We are using the narrower sense.

Same 24 sites, two kinds of authority

The demo runs the same job both ways: map 24 sites. In the federated version, three groups of agents each take 8 sites (the split is agreed between them up front, not handed down), and the only thing crossing between them is what they learned. In the hybrid version, one controller sets the area, the budget, the hours and the escalation path, then stops talking. The agents inside the boundary do the mapping.

The interesting part is what both have in common. Neither sends step-by-step instructions to individual agents. Federated coordination is 12 messages between peers. Hybrid coordination is 4 rules from one frame. If your agent system needs a human to approve every next move, you do not have either of these. You have a queue.

Where control actually sits

• Federated: with the groups. They agree on what to share, and each stays responsible for its own ground. There is no single place to escalate to. Anything that needs agreeing goes through the protocol the groups set up between themselves.
• Hybrid: with the frame. The controller decides what is in bounds, what it costs, when it runs and what happens when something falls outside. Inside that, the agents are on their own.

Federated is trust between equals. Hybrid is a frame with freedom inside it. That is why federated systems are hard to sell to a business that has never run one: there is no single name on the hook for the behaviour of the whole.

Federated systems still need shared rules of the road: a message format, what counts as a conclusion, how to settle a disagreement. 'No centre' means no one in charge, not no structure.

Which one does a business actually want?

Most businesses that think they want a swarm actually want a hybrid. They want the efficiency of agents sorting it out themselves, but they still want a name on the hook when something goes wrong. A boundary gives them that name without turning the agents into remote-controlled toys.

Federated suits independent teams, partners or suppliers who already trust each other and already own their own operations. It is coordination between organisations, not within one.

How to tell which one you are looking at

• Ask what crosses between the groups. Conclusions and findings? Federated. Rules and constraints? Hybrid. Task assignments? That is a fleet.
• Ask who can change the boundary. If one node can, it is hybrid. If the groups would have to agree, it is federated.
• Ask what happens when something fails. Federated: the other groups carry on, but the failed group's ground stays unmapped until the peers agree to cover it. Hybrid: the boundary still holds and agents keep working inside it, but anything that needs an escalation waits until the controller (or a named deputy) is back. That single controller is the thing to design around.
• Count the orders. If a person or a controller is choosing each agent's next step, none of these words apply.

Want the vocabulary underneath this? Single agents, fleets and swarms covers the three basics. Why agents get expensive covers what it costs to keep them talking.