Claude Code subagents: 5 ready-made business examples

Claude Code lets you add subagents: small markdown files in ~/.claude/agents/ that give Claude a specific job, a method and rules. Most examples online are for coding. Here are five for running a business, plus the structure that makes them reliable.

The structure that works

Every good business agent has 5 parts:

  1. One job. "You are the quality gate." Not "you help with marketing".
  2. A "not your job" table. Who does what instead. This stops agents from doing everything badly.
  3. Reads first. The files it must read before working (brand, goal, log).
  4. A numbered method. Steps, not vibes.
  5. An output template. So every answer looks the same and is easy to scan.

Plus one safety line: drafts only, never post, send or spend.

1. The Gate (harsh critic)

Judges every draft like a tired stranger. Keeps a claim ledger: every number, name or "best" in the text is VERIFIED, HELD or BANNED. Anything below 9/10 on any axis fails, with an exact fix. This one agent improves everything else the team makes.

2. The Scout (dated trends only)

Searches what is working in your niche in the last 30 days, and writes down the post, the date and the view count. No date or no number = not a trend. Old advice is the most common AI mistake.

3. The Modeler (decision sims)

Turns "should I hire?" into a small simulation with ranges and runs 100,000 futures. Returns one line: who wins, how sure, the bad case and the biggest lever.

4. The Analyst (kill or double)

Reads your exported numbers and says what to double and what to kill, always with the sample size. "3 posts" is not a pattern. "12 posts, all above median" is.

5. The Manager (2-minute stand-up)

Reads the goal and the board, then writes: pace vs goal, what's ready, the one bottleneck, and today's 3 jobs. Routes work to the other agents.

Copy-paste starter (the Gate)

---
name: gate
description: 'Independent harsh critic. Use for: is this good, check before I post.'
tools: Read, Write, Glob, Grep
---
You did not make this. Agreeing with the maker is a failure.
1. List every piece (hook, body, caption, cover).
2. Claim ledger: each fact = VERIFIED (source) / HELD / BANNED.
3. Score 6 axes 1-10. Any score 7 or less = FAIL.
4. For every score under 9, write the exact fix and who fixes it.
Never post, send or spend.

Ready-made crews

If you'd rather install a full, tested team: Solo Creator Team (10 agents for any creator), Sim Your Decisions (decision simulator + 3 agents) and Restaurant Agency Crew (10 agents to run a local food social media agency).

Get the ready-made kitSee all AI Crew Kits →Sim Your Decisions · $79 →Solo Creator Team · $79 →Restaurant Agency Crew · $79 →One-time purchase · installs in 3 minutes · 30-day refund

More guides · AI Crew Kits for Claude Code. Not affiliated with Anthropic.