As tasks grow, a single agent grinding through them sequentially hits limits: context fills up, and independent subtasks wait in line. Multi-agent systems address this by splitting work across a coordinator and specialized sub-agents. Done well, it's powerful; done reflexively, it's overhead.
The shape
A coordinator agent owns the goal and delegates pieces to sub-agents — a researcher, a coder, a reviewer — each with its own context and tools. Sub-agents work in parallel, report back, and the coordinator integrates. The win is twofold: independent work happens concurrently, and each sub-agent keeps a focused context instead of one agent juggling everything.
One agent is a worker. A multi-agent system is a small team — and teams need coordination, not just headcount.
When it helps
Multi-agent structure pays off when a task genuinely fans out: many files to analyze, many sources to check, several independent sub-problems. Parallelism turns a long serial run into a short concurrent one, and specialization lets each agent stay sharp on its slice.
When it doesn't
For sequential work — where each step depends on the last — delegation adds communication cost without parallelism benefit. And sub-agents don't share context automatically; the coordinator must pass along what each one needs, or they'll flail on missing information. Spawning agents for work a single one could do directly is a common and expensive mistake.
The rule of thumb: reach for multiple agents when the work is genuinely parallel or benefits from focused specialists. Otherwise, one well-run agent with good memory beats a committee.
