This lesson is being recorded
Every video of this course is being uploaded to my YouTube channel. Subscribe so you do not miss it.
Bruno Ashford @brunoashfordSubscribeModule 7 · Lesson 7.1 · 10 min
One Agent per Role
Understand subagents as specialists with a clean context, and why one do-everything assistant breaks down.
Use ← → to move. The marker highlights text and the pencil draws on the slide; navigation locks while you annotate. Undo removes the last mark, Esc exits.
1Know what a subagent actually is
A subagent is a Markdown file with a short header (name, description, optional tools and model) and a body that works as its job description. Project subagents live in .claude/agents/ inside your project folder. Personal ones, available in all your projects, live in ~/.claude/agents/.
2Understand the clean context advantage
Each subagent starts fresh, with its own instructions and your CLAUDE.md, not the whole messy history of your chat. It can read 40 web pages or a 200-row spreadsheet and hand back one page. The detail stays in the subagent; only the conclusion comes back to you.
3Use the three-question test
A task deserves a specialist when the answer is yes to at least two: Does it repeat every week? Does it need a different mindset from the main task (writing versus checking)? Does it produce a lot of noise you do not need to see (research, data crunching)? If not, a skill or a single prompt is enough.
4See what you already have
Claude Code ships with some built-in helpers, and you may already have subagents from plugins. The /agents command shows them all and is where you create, edit and delete your own.
/agents5Map your first candidates
Ask Claude to look at your business context and propose which roles deserve a specialist. You are looking for three to five, not twenty.
Read CLAUDE.md and my skills folder. Suggest 3 to 5 subagents for my business. For each, give: name, the job it does, how often it runs, which tools it needs (read files, web search, write files, connectors) and which tools it must NOT have. Explain why each one is better as a subagent than as a single prompt.Picture hiring one person and asking them to prospect, write your ads, reconcile the books and proofread contracts, all in the same meeting, without ever leaving the room. By the afternoon they would be mixing up client names and forgetting what you said at 9am. That is exactly what happens when you run a whole business day through one long Claude conversation.
Subagents fix this. A subagent is a specialist Claude Code can hand work to. It starts with a fresh context, reads its own job description, uses only the tools you allow, does the work, and hands back a summary. Your main conversation stays clean, and each specialist stays focused.
Key points
- A subagent is a specialist with its own instructions, context and tools.
- Project subagents live in .claude/agents/, personal ones in ~/.claude/agents/.
- Clean context means less confusion and less noise in your main chat.
- Create a specialist only when the task repeats, needs a different mindset or creates noise.
Avoid
- Creating twenty agents on day one, none of them used twice.
- Using a subagent for a one-line task a simple prompt would handle.
- Expecting a subagent to remember your chat history; it starts fresh.
A 12-person agency ran all client work through one Claude session per day. By mid-afternoon, drafts started mixing one client's tone with another's. They split the work into four subagents: researcher, copywriter, data analyst and reviewer. Each one starts clean for every client, so tone mix-ups stopped, and the account manager's main session stayed short enough to use all day.