The Subagent: A Helper With Its Own Desk

In short

A sub-agent is a helper that works in another room, with its own desk.
It comes back and tells you only the result: the mess stays with it.
It's there to keep your table clean — and so to get you better answers for longer.
Some already exist ready-made, and you can create your own.

The problem: some jobs make a mess

There are tasks that, to get done, require looking at a huge amount of stuff and then throwing it away. "Find which document I wrote that thing in": to answer, it has to open twenty files, and nineteen had nothing to do with it.

You already know the catch from the last chapter: those nineteen files stay on the table. You spent half the space to get to a one-line answer, and from then on the rest of the session works worse.

The solution: send it off to do it elsewhere

A sub-agent is a helper you send into another room. It has its own desk, it opens the twenty files itself, makes its own mess, and then comes back and tells you one line: "it's in the March 14th document."

All the clutter stayed over there. Only the answer arrived on your table.

It's not a different model and it's not smarter: it's the same tool, with a separate context. It has its own instructions, the tools you choose to give it, and its own permissions.

The limit, which is also the strength

The empty desk cuts both ways: it doesn't carry your clutter along, but it also doesn't know what you were talking about.

So a delegated task needs to be explained in full. If you send "check if it adds up," it has no idea what's supposed to add up. It's the same job you do when you hand work to a colleague who wasn't in the meeting.

The ones you already have without knowing it

Some helpers already exist, and Claude uses them on its own when needed. The two you'll run into most often:

Explore — goes to search and understand how something is built, without changing it. It's the case of the messy search from before.
Plan — prepares a plan, like in the chapter on permissions, but in a separate room.

If you've seen those names pop up while it worked and didn't know what they were, now you do: helpers sent off to another room.

Creating your own

Once you notice you're always sending the same kind of helper with the same instructions, it's worth defining it once and for all: a helper with a name, a task, a character, and the tools you choose.

The practical way is the same as always: ask Claude — "create me a helper that does this and this, and can only read" — and let it write the file. Then you reread it, same as with skills.

And now

Put that way, it sounds like it's always worth delegating. It isn't, and the reason is the bill: a helper starting from zero has to rebuild everything, and that costs. In the next lesson, when it's worth it and when it isn't.

Did you get it?

Two minutes, no grade. Trying to answer — even getting it wrong — cements the concepts far more than rereading. Not in the mood? Skip it: the lesson still counts as complete.

What comes back into your conversation when a sub-agent is done?

Does the sub-agent see your conversation?

What is the built-in helper called Explore for?

1
1. 1
2. 2
3. 3
2
1. 4
2. 5
3. 6
4. 7
3
1. 8
2. 9
3. 10
4. 11
5. 12
4
1. 13
2. 14
3. 15
4. 16
5. 17
6. 18
5
1. 19
2. 20
3. 21
4. 22
5. 23
6. 24
6
1. 25
2. 26
3. 27
4. 28
5. 29
6. 30
7
1. 31
2. 32
3. 33
4. 34
8
1. 35
2. 36
3. 37
4. 38
9
1. 39
2. 40
3. 41
10
1. 42
2. 43
3. 44
11
1. 45
2. 46