Multi-Agent Systems

Multi-Agent Systems, MAS, are groups of autonomous software entities that share an environment and interact to reach goals.

Each agent has its own perception, decision logic, and actions. Overall behavior comes from many exchanges. Agents may cooperate, compete, or negotiate. In traffic, each vehicle can share speed and position so flow organizes without a single dispatcher. In smart grids, homes and generators negotiate power use in real time. In markets, algorithmic traders react to prices and add liquidity.

Robot swarms use simple local rules to search hazardous sites or run rescue.

Two properties matter. Scale: adding agents rarely requires redesigning the whole system. Failure tolerance: when some agents die, others can cover the work. As IoT grows and vehicles drive themselves, MAS is how billions of devices coordinate.

There is no single brain. Agents sense, decide, and act, then react to each other. Traffic agents share speed and position. Grid agents bid for power. Trading agents answer prices and supply liquidity. Swarm robots follow local rules in places people cannot go. Adding agents is how the system grows, rather than rewriting a monolith.

If some agents fail, others still run, which is the failure-tolerance argument. IoT and autonomous vehicles push the agent count toward billions, so coordination has to be this distributed style. Multi-agent systems are older than LLMs. The field studies agents that cooperate or compete in a shared environment.