Tal EckerlingAI Specialistעברית
Building with Claude CodeLesson 16 of 16 · 6 min

Agent teams

In this lesson you'll learn

  • A standing team and a one-off team
  • The preparation that decides everything
  • How to run a team in Claude Code
  • When it's worth it and what it costs

Listen to this lesson

8 min · 7 parts · AI narration

From a team of one to a whole team

After a while with Claude, a certain kind of frustration builds up. It doesn't come from Claude not being good. It comes from starting over every time: who you are, what the business does, what tone to write in. You get a good result, close the chat, and the next day you explain everything again.

That's the gap between using AI and having AI that works for you. In this lesson we'll talk about two kinds of team: a permanent team that knows your business, and an on-the-spot team that Claude Code puts together for one big task.

A permanent team: experts, a manager and one business profile

An agent isn't a long prompt. It's a professional character with one specialty, its own working rules, and access to the shared business profile. A team is several agents like that, with a manager who routes work to them: a pricing question goes to the pricing specialist, a request for a post goes to whoever writes posts. You don't need to know who to ask.

The topic A regular chat An agent team
Context You explain again every time One business profile that everyone reads from
Specialty The same Claude for everything An expert for each area, with its own rules
The output Depends on the quality of the prompt Consistent, because the base is always there

In a good team, the first run is a short interview where the team gets to know you and builds the business profile. That's the difference between generic output and a team that sounds like you.

In practice, a team like this comes as a plugin: one bundle of agents, skills and commands. That's also how the AI teams I build are made, and they work in English or Hebrew, both in the Claude app and in Claude Code. You upload the plugin once to your Claude account, it's saved there, and it syncs to Claude Code on its own. In Claude Code you can also install directly with /plugin. And one detail from the official support table: in a short chat with no task, only the plugin's skills load, and the full team, with the agents, works in Claude's tasks and in Claude Code.

The prep that decides everything

I'll say it openly, because it's the thing that matters most and gets skipped most: don't start before your materials are ready. I have a client who spent a week and a half preparing materials before we started, and it wasn't a waste of time. It was the whole job.

What to gather into one folder:

  • Your website, or your business page if you don't have a site.
  • Content you wrote yourself: posts, emails, articles. That's what teaches the team your voice.
  • Presentations, a company profile, product documents and prices.
  • Client testimonials, media mentions and awards.
  • Everything you've already written about your audience and your competitors.

It doesn't need to be neat. It needs to be in one place. And on day one, tell the team three things: what the goal is right now ("I want you to get to know me before we start"), where the materials are, and what your sensitive points are. Everything's scattered between Drive and Notion and there's no presentation? Say so. A team that knows adapts instead of assuming.

An on-the-spot team: agent teams in Claude Code

As of October 2026, Claude Code has a feature called agent teams, and it's still experimental and off by default. One conversation acts as the team lead: it hands out tasks, coordinates and summarizes. The teammates are separate Claude Code conversations, each with its own context. They share a task list, and they talk to each other directly. You can also talk to any of them directly.

How to turn it on: add the CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS variable with the value 1 to your settings, and open a new conversation. You don't have to touch any files yourself: ask Claude to add it to your personal settings. From there, describe the task and the teammates you want in your own words, and the team lead puts them together. They show up in a list under the input box: arrows to select, Enter to view one and talk to it directly.

Teammates start in the team lead's permission mode, and their approval requests come to you in the main conversation. And one side effect from the docs: when the feature is on, Claude may put a team together even when you didn't ask. Not for you? Set the value back to 0.

When it's worth it, and what it costs

The official docs list four strong uses: research and review from several angles, building separate parts that don't step on each other, testing several hypotheses in parallel when it's not clear what the problem is, and changes that cut across several layers of a project.

The question Subagents An agent team
Who coordinates The main conversation The teammates, through a task list and messages
Good for A focused task where only the result matters Work that needs discussion, debate and coordination
Cost Lower Higher, every teammate is a separate Claude

And in money terms: according to Claude Code's cost management page, an agent team uses about 7 times the tokens of a regular conversation when the teammates work in plan mode, and the spend grows with every teammate. For a routine task, one conversation is cheaper.

The official recommendations: start with three to five teammates, run them on Sonnet, shut each one down when it's done, and start with research and review rather than building. And two more rules: two teammates editing the same file will overwrite each other, and going back to an old conversation doesn't bring the teammates back.

Try it now

  1. Ask Claude Code to turn on agent teams in your personal settings, and exit with /exit.
  2. Open a new Claude Code conversation in the folder of the landing page you built.
  3. Paste:
Ready prompt

"Put together an agent team of three to review the landing page in this folder, without changing any file: one as a client seeing the page for the first time, one checking the language and wording, and one checking mobile and accessibility. Have them challenge each other's findings, and at the end give me five notes in order of importance."

  1. Pick one teammate from the list and ask them a follow-up question.
  2. At the end, check in /usage how much it cost, and compare it with the single-subagent review from the previous lesson.

What to take with you

A permanent team saves you explaining it all again, and an on-the-spot team multiplies the brainpower for a big task, and the cost too. The difference isn't in the number of prompts. It's in the shared memory and the specialty. And there's no magic here: a team takes the grunt work off your plate, but it doesn't think for you.

Want to keep your progress?

Sign up once, at no cost: the lessons you finish are marked in your library, and the shelf of recipes and tools opens up.

Join the academy
Agent teams | The Claude Academy | Tal Eckerling