Both subagents and agent teams involve your main Claude delegating work to other Claudes. But they solve different problems, and reaching for the wrong one wastes time.
The one-line version
- A subagent is one helper you send on a side-quest. It comes back with an answer.
- An agent team (workflow) is many helpers working in parallel on pieces of a bigger job. They come back with a combined result.
Subagent — the research assistant
A subagent handles a single, self-contained task that would otherwise bury your main thread:
Use a subagent to find every place we send email and summarize how each one works.One helper goes off, does the digging, returns a clean summary. Your context stays focused.
Reach for a subagent when:
- You have a side-quest (research, reading lots of files, answering a tangential question)
- You care about the answer, not the process
- The task is one job, not many parallel jobs
Agent team — the parallel workforce
An agent team splits a bigger job across multiple Claudes running at the same time:
Review every page in this site for broken links — run them in parallel and give me one combined report.Instead of one Claude checking pages one by one, a team of Claudes each take a batch and work simultaneously. The results are gathered into a single output.
Reach for an agent team when:
- The work splits cleanly into independent pieces (one per file, per module, per question)
- Sequential work would take too long
- You want a combined result from many parallel investigations
Side by side
| Subagent | Agent team (Workflow) | |
|---|---|---|
| How many helpers | One | Many, in parallel |
| Task shape | One side-quest | Many independent pieces |
| Best for | Research, reading, tangential questions | Refactors, reviews, migrations |
| Returns | One answer | A combined report |
| Speed gain | Keeps your context clean | Multiplies throughput |
| When it’s wrong | When you need many parallel workers | When the task doesn’t split into independent pieces |
The mental model
Think of a subagent as sending one person to the library to research something for you.
Think of an agent team as handing out folders to a room of people who each review their folder and report back.
Both keep your main thread clean. The difference is whether you’re delegating one job or dividing one big job into many.
Composing them
These aren’t either/or. A workflow fans out across agents, and each of those agents might spin up its own subagent for a research sub-task within its piece. The levels compose naturally:
- You give a goal to your main agent
- The main agent creates a workflow that fans out to team agents
- Each team agent might use a subagent for a deep-dive within its piece
- Everything rolls back up into your conversation as a clean result