Single agents work alone. One LLM. One system prompt. One set of tools. They handle everything themselves.
This works great when problems are well-defined and you don’t need collaboration. Simpler to build. Easier to debug. More predictable.
Multi-agent systems use multiple agents working together. Each has a specific role. They communicate. They coordinate. Like a team where everyone has their specialty.
You need multi-agent when tasks require different expertise or multiple execution paths. Example: One agent screens leads, another drafts emails, a third updates the CRM.
The tradeoff is complexity. Multi-agent systems need good communication protocols. Clear task division. Sometimes a leader agent to coordinate.
Here’s the key insight: Start single-agent until you can’t. Most problems don’t need the complexity of multiple agents.