Claude Runs 4 Tasks at Once Now (Sub Agents Explained)

Give Claude one complex task and it splits it across four agents working at the same time. Not one assistant grinding through a checklist. A team.

They're called sub agents, and they're parallel workers. Instead of Claude doing everything in sequence (step one, step two, step three), it spins up multiple agents that all work simultaneously. (You know, the thing you've been trying to get actual humans to do in meetings for years.)

Here's the 40-second version:

How It Actually Works

Say you tell Claude: "Prep everything for my investor pitch." Here's what happens.

Agent one does market research. Agent two pulls your financials. Agent three writes the narrative. Agent four builds the slide deck. All at the same time. What used to be an afternoon of sequential back-and-forth happens in one pass.

And here's the part that makes it powerful: each agent can use different skills you've already built. Your research skill, your financial analysis skill, your deck builder skill. They all fire together. The output comes back as one unified deliverable, not four loose files you have to stitch together yourself.

Sub agents don't replace the other Cowork features. They orchestrate them.

Why This Is the Final Boss Feature

Sub agents are number 20 of 20 in my Cowork series for a reason. This is where all the other concepts come together. Skills define the workflows. Connectors reach your apps. The bash tool runs the code. Artifacts package the output. Sub agents conduct the whole orchestra at once.

If you've been building skills one at a time, this is the payoff. Every skill you've created is now a worker Claude can deploy in parallel.

Want the full picture? I broke down all 20 Cowork concepts in one post, from workspace folders to this.

Now go give Claude a task big enough to need a team. That's the whole point.