What is a multi-agent setup?
A multi-agent setup uses more than one AI agent, each with a narrower job, passing work between them. A subagent is a helper one agent calls. A swarm is a loose name for many agents at once. The owner still needs one place where a person can see the result.
Also called: subagent, agent swarm
In plain English
One agent drafts the tenant text. Another only checks that the date and the amount came from the sheet. That is already multi-agent. It is not five bots arguing in a tool you cannot log. More agents means more handoffs and more ways to lose the thread. Two agents means two places for a mistake to hide. Start with one draft and one checker, which can be a person. Add a second model only when you can point at a job the first one keeps failing for a clear reason, such as pulling the wrong record. If you cannot draw who speaks to the customer, you have a crowd, not a process.
How an owner would use it
Add a second agent only when a single one keeps mixing two jobs, such as writing and verifying. Give them different instructions and different tools. If you cannot say which agent is allowed to send, you have too many.
Common mistakes
Starting with a swarm because a demo had five avatars is how a small office loses the plot. Subagents that can all send mail will double-message. Do not give every agent the same full inbox access. Do not call a single chatbot with moods a multi-agent system. If the log cannot show which agent did what, take one of them away.
A worked example
A landlord with a dozen units
One agent writes, one agent checks the ledger line
This is an illustration, not a customer result. The drafting agent prepares a late-rent note from fields on the sheet. A second agent may only flag if the amount in the draft is not on that row. Neither one emails the tenant. The owner sends the note after the flag is clear.
Related terms
Prompts and guides
Questions owners ask
What is a multi-agent system?
It is more than one agent dividing a job, such as a drafter and a checker, instead of one agent doing every step with every tool.
How do I use more than one agent without making a mess?
Split by permission. The checker should not be able to send. Log both. Add a third agent only after the first two have been dull for a while.
Is a subagent a different product?
A subagent is a helper an agent calls for a side task. It is still your responsibility. If the helper can spend money or email, it needs the same approval rule as the main agent.
Product guides are on AI apps for business.