Welcome to Prompt, Tinker, Innovate—my AI playground. Each edition gives you a hands-on experiment that shows how AI can sharpen your thinking, streamline your process, and power up your creative work.
This week's playground: AI loops
An AI loop is a set of instructions that tells AI to do a task, check its own result, fix what failed, and check again. It repeats until the result passes or it hits a limit you set.
Most of us use AI one pass at a time. We give it a task and read the output. Then we decide what to fix and ask again. You are the checker in that setup, and you are the slowest part of the cycle.
A loop moves the checking to AI. You define what "done" looks like. AI does the work, tests it against your definition, and revises until it passes. You review the version that passed.
Every loop has four parts:
- The task: what AI produces
- The check: a test with a pass or fail answer
- The fix: AI changes only what failed
- The stop rule: when the loop quits, on success or after a set number of rounds
Loops fit any task with a check you can write down. A loop for code runs the tests and fixes the failures. A loop for a report compares the totals to the source and fixes the mismatches. This week's example is content, because a draft is easy to check against written rules.
Use case spotlight: a content loop with a stop rule
For content, the check is a rubric: a list of rules the draft must pass. A loop only works if it knows when to stop. I use three rules.
First, every check is pass or fail. "Make it punchy" is not a check. "No sentence over 25 words" is. Writing checks like that forces you to say what "done" means.
Second, the loop rewrites only what failed. If you let it rewrite everything each round, the voice drifts toward average.
Third, it stops after three rounds. If a draft still fails after three, the problem is the idea, not the wording. That is a signal for you to step in.
I built a scorer for this edition—a simplified version of it is below. It counts banned words, sentence length, passive voice, and missing sections. Then I ran this newsletter through it. You can see the score for each round at the end of the post.
Your AI experiment: try this prompt
👉 Time to tinker: Pick a piece of content you have not sent yet. An email, a post, a one-pager. Write five to eight rules it must follow. Make each rule something a stranger could check. Paste both into your AI tool of choice.
📝 Prompt:
I am going to give you a draft piece of content and a list of rules. Run this loop. Round 1: Score the draft against each rule. Mark each rule PASS or FAIL. For every FAIL, quote the exact sentence that broke the rule. Then rewrite only the sentences that failed. Do not touch anything that passed. Keep my voice: direct, short sentences, no filler. Round 2 and Round 3: Score the new version the same way. Rewrite only what still fails. Stop early if every rule passes. Stop after Round 3 no matter what. At the end, show me a table with one row per rule and one column per round. Then give me the final draft. If any rule still fails after Round 3, tell me which one and why you think it keeps failing. My rules: [paste your rules here] My draft: [paste your draft here]
💡 Pro tips: three ways to get more from the loop
- Add one rule that guards against bland output. Something like: "Keep at least one specific detail or opinion in every section." Without it, loops tend to sand the edges off until the draft says nothing.
- You can also split the job. Ask one chat to write the draft and a second one to score it. Even better, give the draft to a second AI tool entirely. A scorer that did not write the draft is harder on it.
- Save this loop as a Skill to use over and over again. Prompt your AI tool to turn this loop into a Skill. Once complete, save it and call it next time you need the loop run.
What did you discover?
Which of your rules failed most often? That tells you where your first drafts are weakest, and which habits are worth fixing at the source.
Did any rule turn out to be impossible to check? Rewrite it until a stranger could score it. Then share the version that worked.
The scorecard for this edition
Here are the simplified rules I put in place for the editing of this newsletter:
📝 My rules:
1. No banned words: revolutionary, unlock, leverage, embrace, delve, elevate, "game changer," "cutting-edge," "in conclusion." 2. No sentence longer than 25 words. 3. No passive voice. Write "we ran the test," not "the test was run." 4. No exclamation marks. 5. Every header uses sentence case. Capitalize only the first word and proper nouns. 6. No more than 3 em dashes. 7. No spaces around em dashes. "Tinker—my" passes. "Tinker — my" fails. 8. The draft runs between 400 and 800 words.
How we fared:
- Round 1: 9 of 12 rules passed on the first run. It failed on sentence length, passive voice, and sentence case.
- Round 2: 12 of 12. The loop stopped there.
Until next time—keep tinkering, keep prompting, keep innovating.
📩 Not subscribed yet? Hit the button at the top. You won't want to miss what's next.
