Agents receive a research goal and divide tasks (literature review, hypothesis generation, experiment design and execution, analysis). They communicate and coordinate through shared state or an orchestration layer, exchange partial results and iteratively narrow the solution space. Human oversight and verification mechanisms limit errors and risk.
A single agent is a bottleneck in complex research. Swarms parallelize and scale scientific discovery by dividing work and coordinating results.
