Playbook

7 Ways to Actually Use Loops

Everyone talks about loops. Nobody says where you'd point one. Here's seven — ready to paste.

The Idea

Everyone's talking about loops right now — but no one tells you what to actually point one at. So here are seven practical ones you could set up today.

Quick reminder: a loop is where you give Claude a goal and it works on its own until it hits it — no babysitting. Each one below is a paste-ready starter: drop it into Claude Code and it'll help you set the whole loop up — goal, checks, stop condition and all — and walk you through the plan before it runs.

The Seven

The Speed Loop

Optimises every page until it loads under fifty milliseconds — measuring after each change so it knows when it's done.

Loop 1 — paste into Claude

I'd like to set up a "speed loop" in Claude Code, and I want your help getting it right. The goal: get every page in my app loading fast — ideally under 50ms. Before you start, help me turn this into a proper loop: - Agree how we'll measure page-load time consistently across every page, so the result is real and not guesswork. - Pin down what "done" looks like, plus a max number of rounds so it can't run forever. - Each round: make one improvement, re-measure, and commit only if it got faster. - Keep progress in /tmp/speed-loop.md so it survives a fresh start. Walk me through the plan first. Once I'm happy, kick it off.

The Overnight Docs Sweep

Updates your docs to match the day's changes and opens a PR. Wake up to documentation that's actually current.

Loop 2 — paste into Claude

I'd like to set up an "overnight docs sweep" loop in Claude Code, and I'd like your help setting it up. The goal: bring my documentation back in line with what the code actually does, then open a PR — so I wake up to docs that are current. Help me set it up properly first: - How you'll work through the codebase, and what "up to date" means here. - A clear finish line and a stop condition. - Each round: update a section, check it against the real code, commit. - Track what's changed in /tmp/docs-loop.md, and open a PR at the end. Talk me through the plan before you start.

Refactor Till Happy

Cleans up the architecture until it's genuinely tidy — testing and committing after each step.

Loop 3 — paste into Claude

I'd like to set up a "refactor" loop in Claude Code, and I want your help setting it up. The goal: clean up the architecture until it's genuinely tidy — without breaking anything. First, help me make it safe and finite: - Define what "clean enough" looks like, so the loop has a real finish line. - Set a max number of rounds. - Each round: make one improvement, run the tests, do a quick self-review, and commit only if it's green. - Track progress in /tmp/refactor-loop.md. Show me the plan first. Once I'm happy, run it.

The Logging Loop

Adds logging until every important path is covered — so next time something breaks, you can see what happened.

Loop 4 — paste into Claude

I'd like to set up a "logging" loop in Claude Code, and I'd like your help setting it up. The goal: add logging so every important path in the app is covered — so next time something breaks, I can actually see what happened. Help me set it up first: - Let's agree what counts as an "important path" and what "covered" means, so it has a clear finish. - Set a stop condition. - Each round: add logging to one path, run the app and confirm the new logs actually fire, commit on green. - Track which paths are done in /tmp/logging-loop.md. Walk me through the plan, then start.

The Nightly Error Sweep

Reads your production logs, fixes what's broken, and opens a PR. Triage while you sleep.

Loop 5 — paste into Claude

I'd like to set up a "nightly error sweep" loop in Claude Code, and I need your help setting it up safely. The goal: each night, read my production error logs, fix what's broken, and open a PR — so triage happens while I sleep. Help me set it up properly first: - How you'll pull the logs, and how you'll prioritise (most frequent / most damaging first). - A limit on how much to take on per run, so it stays sane. - Each round: fix one error, verify the fix actually works, commit. - Open a PR at the end with a summary, and track what you fixed in /tmp/error-loop.md. Show me the plan before running anything.

The SEO + GEO Audit

Improves how you show up in search engines (SEO) and AI answer engines (GEO) — fixing gaps and re-checking until nothing critical is left.

Loop 6 — paste into Claude

I'd like to set up an "SEO + GEO" loop in Claude Code, and I want your help getting it right. The goal: improve how my site shows up both in search engines (SEO) and in AI answer engines like ChatGPT, Claude and Google's AI answers (GEO) — fixing the biggest gaps and re-checking until nothing critical is left. Help me set it up first: - Agree the checks we'll run for both SEO and GEO, and what counts as a "critical" issue. - Set a stop condition. - Each round: audit, fix the highest-impact gaps, then re-crawl / re-check. - Track each pass and what changed in /tmp/seo-geo-loop.md. Talk me through the plan. Once I'm happy, run it.

The Eval Loop

Runs a hundred real scenarios through your app and fixes everything that fails — until they all pass.

Loop 7 — paste into Claude

I'd like to set up an "eval" loop in Claude Code, and I'd like your help setting it up. The goal: throw around a hundred real-world scenarios at my app and fix everything that fails — until it all passes. First help me set it up properly: - How we'll generate realistic scenarios, and how each one is judged pass or fail. - A stop condition. - Each round: run the scenarios, fix the underlying bug behind each failure, re-run. - Commit on green and track results in /tmp/eval-loop.md. Walk me through the plan before you start.

Two Things Before You Run One

The whole trick is giving it a goal it can actually check — "loads under 50ms," "tests all pass," "no critical SEO or GEO issues left." That's why every starter above asks Claude to nail down the checks before it runs. A loop with nothing to verify against just runs in circles agreeing with itself. (That's the entire point of the Verify Loop.)

And be straight with yourself about cost: loops burn through tokens, and some of these run for hours. That's fine for an overnight docs sweep or a weekend refactor — just don't kick one off and wander away from a metered account expecting it to be cheap. Set a stop condition and let it work.

Inspired by a video from Matthew Berman. He's also collecting these in his Loop Library — worth a look if you want more than seven.

Want to get better at this?

Let's set up your agent ecosystem.

This page is one job. The whole thing is a set of them running off the same foundation, the folder your AI works out of, the brief on how you like things done, the accounts it can reach and the skills that run off all of it, so your admin gets done the way you would do it. It is what I teach one to one, on your own machine and against your own work, and the quickest way to find out what yours would look like is a proper chat about it.

Book a call Or read about the course first

Not ready for a call? Start with the free agent series, what an AI agent actually is, and build up from there.