1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
51 small wins to finish your pathNext question →
Single agent or multi-agent: how do you decide?
30-second answerSay your answer out loud first, then reveal.
Reasons to go multi-agent
- Context isolation: a sub-agent can read 50 documents and return a 1-page summary, keeping the main agent's context clean.
- Parallelism: research across 5 independent topics at once, reducing wall-clock time.
- Specialisation: different prompts, tools or even models per role, e.g. a SQL agent and a charts agent.
- Too many tools: split 60 tools among 4 specialised agents instead of one confused agent.
- Security boundaries: the agent that reads untrusted web content has no access to send-email tools (see Q40).
Reasons not to
- Token cost: multi-agent research systems can use several times more tokens than a single chat. Anthropic reported roughly 15x more than chat for their research system.
- Coordination failures: agents duplicate work, contradict each other, or lose information at hand-offs.
- Debugging: failures spread across multiple traces.
- Tightly coupled tasks (like most coding edits) parallelise poorly, because every part depends on the others.
Good answer. "I'd start with one agent and measure. If I see context overflow, tool confusion, or tasks that can obviously run in parallel, I'd split out sub-agents with narrow responsibilities and clear input/output contracts."
Follow-ups to expect
- Name multi-agent topologies. (See Q21.)
Related
Every expert started right here.