The curriculum

The operating model: why an LLM in a loop is not a business

The smallest amount of structure that turns a model running in a loop into something you can actually leave running. The four parts, and the two roles.

Read the lesson →

Constraints as data, not vibes

A committed scope, budget, and acceptance file the agent re-reads every run, reviewed like code. Why a prompt is not a guardrail, and why deny beats allow.

Read the lesson →

A token budget read from its own ledger, with a hard cap

Log every model call to a ledger, enforce a real daily cost cap from recorded usage, and stop instead of running up a bill overnight.

Read the lesson →

CI as the acceptance gate

A protected branch plus a required check, so neither you nor the agent can ever merge something broken. A change is only done when CI is green.

Read the lesson →

The CEO-agent oversight loop

A second agent that grades each session on traction, not motion, and tells the worker to stop rabbit holes and pivot when the numbers do not move.

Read the lesson →

Shipping products, payments, and a funnel the agent maintains

Wiring real Stripe checkouts, landing pages, and a content and SEO funnel that the agent keeps current by itself.

Read the lesson →