The Loop — AI Operator Course #3
“I put 20 good tasks somewhere. Tomorrow morning there are three PRs waiting for me.”
Three, not twenty — the queue holds the other seventeen until you've cleared these.
Before / After
Before
Developer → AI → code. You are the operator. You load the context every time, for every task, from scratch.
After
Backlog → readiness check → AI developer → PR → human review. You are the reviewer.
The Four Levels
- L0 — No AI in the loop. Every line typed by hand. AI is a search engine you occasionally paste into.
- L1 — AI as a chat window. You open a session, paste context, wait, review, paste the next thing. One task at a time, and you're in the room for all of it.
- L2 — AI as a queue (you are here after this course). You're here after this course. A tracker item clears a context readiness gate, an agent picks it up unattended, and a PR is waiting when you next check.
- L3 — Multi-agent orchestration. Agents managing agents. You don't need this yet, and most teams never will. Learn Level 2 first.
You don't need Level 3. Learn Level 2 first.
The Readiness Gate at 85%
An agent working from a half-specified task guesses, and a guess wrong at 2am is a PR you have to unwind at 9am. The Readiness Gate scores every tracker item before an agent is allowed to touch it — nothing below 85% context completion gets picked up.
This is the edge the course exists to teach: not a bigger model, a gate in front of it.
Cadence: One an Hour
The loop fires once an hour, because that's one human's review capacity. Faster doesn't help — a queue of unreviewed PRs is just a backlog wearing a different hat.
Running Right Now
1,118 runs
382 hours of unattended agent work, on the same loop this course teaches. No dollar figure — this is a workflow, not a savings calculator.
What You Get
- The zip: unzip into ~/.claude/, open Claude Code, and say "install the loop." Claude does the rest.
- The /install-the-loop session: about 45 minutes, walks your machine through config, skills and the hourly job, then fires once so you watch it work.
- Five Claude Code skills that run the loop end to end.
- Eleven reference pages for what to read before something breaks, and after.
- The Automation Accountability Check, so the loop tells on itself when it's stuck instead of silently stalling.
- Branch-janitor, as a bonus: audits and cleans up stale branches, no extra setup.
Who It's For
Qualifies: an AI-assisted (L1) developer already using an agent day-to-day, comfortable with git and PRs, who wants a queue instead of a chat window.
Doesn't yet qualify, but welcome: a determined non-technical solopreneur with no dev workflow to plug a queue into. This course assumes one exists — start with Get Stuff Done first if that's you.
Price
Founding — current
€29
5 seats left
Next 5
€49
5 seats left
Standard
€99
Open
Founding price is base only — a testimonial is required within 14 days of your first merged agent-authored PR. One-time payment, lifetime access, and the loop is yours to keep either way.
Add-ons, never bundled at founding tiers: Install With Peter (€450, for companies who'd rather hand the whole session over) and Keep My Loop Alive (€19/mo — my updates as I make them, not a licence to keep working).
The Guarantee
Your first agent-authored PR within 7 days, or a full refund and you keep the skills. One condition: complete the ~45-minute install session. No other conditions.