What Is a Multi-Agent System? Three Core Working Modes Interviewers Expect

Multi-agent systems are commonly organized into three working modes that cover about 90 % of real-world use cases.

Sequential execution : multiple agents run one after another, passing the result forward like a relay; suitable when tasks have clear predecessor-successor dependencies.

Parallel execution : agents run concurrently when sub-tasks are independent; results are later aggregated by a coordinating agent, fitting scenarios with relatively independent tasks.

Evaluator-optimizer loop : one agent generates an output, a second agent evaluates it against explicit criteria, and if the output fails to meet the standard it is sent back for revision; the cycle repeats until the criteria are satisfied or a maximum number of iterations is reached, making it ideal for tasks with clear evaluation metrics that can be improved through iterative feedback.

For interview answers, mentioning these three patterns signals practical experience with agents and can earn extra points.

Additional guidance: design agents as simply as possible—if a single agent can accomplish the task well, avoid adding extra agents; when using multiple agents, default to the sequential pattern; consider parallel execution only when tasks are truly independent and yield measurable benefits; adopt the evaluator-optimizer pattern only when the result can be explicitly evaluated.