Every working agent has the same four parts.
Whatever platform you build it in, an agent is only as good as four decisions behind it. Get these right and the specific tool matters far less than people assume.
01 · INSTRUCTIONS
Its job description
The rules that define its role, its boundaries, and what “done” looks like. This is the single highest-leverage thing you'll write.
02 · INPUTS
Its eyes
The specific data and tools it's allowed to read — files, exports, connected apps, APIs. Nothing it wasn't explicitly given.
03 · JUDGMENT
Its rubric
The literal, specific criteria it uses to decide what's worth flagging versus noise. Vague instructions produce vague output.
04 · OUTPUT & LEASH
Its voice, and its limit
The exact format it reports back in, and the hard limit on what it's allowed to do without a human. Both matter equally.
You don't need to write code to have all four. A well-configured Claude Project or custom GPT with your rubric pasted into its instructions, a no-code flow (n8n, Zapier, Make) that pulls data on a schedule, or a short script an engineer points at your tools' APIs — all of these count as “an agent” as long as they have all four parts above. Pick whichever is fastest for you to stand up this week. Upgrade the plumbing later.
PART TWO — BUILDING THIS AGENT
The rest of this one is for members — the full framework, plus the working file to run it on your own company. → Subscribe to keep reading
Subscribe to CLUB to read the rest.
Become a paying subscriber of CLUB to get access to this post and other subscriber-only content.
UpgradeA subscription gets you:
- Everything included in Premium.
- Early access to new tools and the specialized AI and product-market fit newsletter.
- One-on-one feedback on your business, with actionable guidance on strategy, positioning, and growth
- Masterclasses with Pablo Fernandez Alvarez