How to Create Agents in Claude (4 Specialists That Run Your Week)

Everyone talks about AI agents like they require a computer science degree. They don't. In Claude, an agent is a specialist you brief once: a role, a job, a standard, and the context it needs. The difference between "chatting with AI" and "running agents" is the difference between asking a stranger for advice and delegating to an employee who already knows your business.

Here's how to create them — no code required — and the four agents we actually run.

What makes an agent an agent

A plain chat starts from zero every time. An agent starts from a role brief — a standing instruction that defines:

• Identity: who the agent is ("You are my growth marketer...")
• Scope: what it owns and what it never touches
• Standards: what "good" looks like, with examples
• Context: your business, your offer, your numbers

You can set this up three ways in Claude: as a Project with custom instructions (simplest — one project per agent), as a skill that loads on demand, or as subagents in Claude Code for technical work. For most founders, one Project per agent is the sweet spot: open the Marketer project, and Claude IS your marketer, context and all.

The role brief template

Paste this into a Project's custom instructions and fill the brackets:

You are my [ROLE]. You own [SCOPE].

My business: [one paragraph - what you sell, to whom, at what price]

Your standards:
1. [What good output looks like]
2. [What you never do]
3. [When to push back instead of agreeing]

Always end with: the single next action I should take.

The last line matters more than it looks. Agents that end with a recommended action get used daily; agents that end with summaries get abandoned.

The 4 agents every founder should run

1. The Strategist. Plans the week, kills bad ideas, forces focus. Brief it on your goals and let it argue with you — its job is to say "that's a distraction."
2. The Marketer. Owns content and campaigns. Feed it your voice, your niche, and your best-performing posts; it produces drafts in your style, not generic AI copy.
3. The Operator. Handles the busywork: emails, SOPs, checklists, follow-ups. The agent you delegate the boring 80% to.
4. The Analyst. Reads your numbers and tells you the truth. Give it your revenue, costs, and funnel data — it flags what's leaking before it hurts.

This is the exact setup from our Founder OS — the system that turns one Claude subscription into a four-person team. If you're deciding between Claude and ChatGPT for business, this multi-agent pattern is where Claude pulls ahead.

Make your agents check each other

The upgrade that separates hobby setups from real ones: agents reviewing each other's work. The Strategist grades the Marketer's campaign against the goals. The Analyst stress-tests the Operator's plan against the numbers. AI models are famously bad at catching their own mistakes — a second specialist catches what the first one can't see.

Run the loop manually (paste one agent's output into the other) or wire it into a single workflow prompt: "Draft as the Marketer, then critique as the Strategist, then show me both."

Common mistakes

• One agent for everything. A generalist brief produces generalist output. Four sharp specialists beat one blurry assistant.
• No standards in the brief. Without examples of "good", the agent averages toward generic. Feed it your best work.
• Set and forget. Update the brief when your offer or numbers change — stale context produces confidently wrong advice.

Get the complete setup

All four agents — full role briefs, copy-paste ready — are in the free Founder OS below. And if you want every playbook behind it, try the whole membership for $1.