One prompt turns your 30-minute chat into a repeatable process

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: Turn your best AI session into a repeatable skill

You know that feeling when you finally get an AI conversation to produce exactly what you wanted? You went back and forth, corrected the tone, fixed the format, told it what to cut. Twenty minutes of refinement, and now the output is perfect.

Then you close the chat. And all that work disappears.

Next week, same task, you start from zero.

This week's experiment fixes that. At the end of a great working session, you ask one question: "Can you create a skill that follows this process every time?" The AI packages everything you just taught it—the steps, the corrections, the quality bar—into a reusable instruction set you can run again and again.

Why this matters

Most people treat AI conversations as disposable. Every chat starts from scratch, which means you re-explain your context, re-correct the same mistakes, and re-teach your preferences every single time. That's not a tool problem. It's a workflow problem.

The professionals getting the most out of AI right now have made one mental shift: they've moved from chatting to building. A single good conversation is a win. A conversation you can rerun forever is an asset.

And here's the part people miss: you don't need to know how to write a skill. You just need to have done the work once. The AI watched the whole process. It knows what you corrected, what you rejected, and what "good" looked like when you finally approved it. It can write its own instructions better than you can.

Use case spotlight: The process was the hard part

This works for almost anything you do on repeat. Here's what it looks like in practice:

• Inbox triage. You already know what a "must-answer-today" email looks like versus something that can wait. Most people run that judgment call fifty times a day, from scratch, every time. So work through it once with the AI: show it a stretch of your inbox, tell it why one email jumps the queue and another gets ignored, correct it when it flags the wrong thing. Once it's reading your inbox the way you do, skill it. Now that triage runs itself every morning, and you're just reviewing the calls instead of making them from zero.
• Competitive intel. Say a client asks how they stack up against a competitor. You dig in: pricing, positioning, where they're weak, where they're strong, what the story is. That's real work, and it's easy to think it only applies to this one client, this one competitor. But the process you just built—gather sources, compare on the same axes, call out the gap—is the same process next time a different client asks about a different competitor. Skill it once and you're not rebuilding the analysis from scratch every time someone asks "how do we compare."
• Client deliverables. Say you're producing a blog post for a client. You pull insights from a source document, shape an outline, draft in a specific voice, run it past a persona review, edit until it's tight, then check it against brand guidelines before it goes out. That's six stages, each with its own round of corrections. Work through it once, start to finish, and you haven't just written a blog post. You've mapped a pipeline. Bundle it, and every new source document runs the same path without you rebuilding it each time.

The pattern is always the same. The hard part was never writing instructions. The hard part was figuring out the process. And you do that naturally, just by working through the task once with the AI.

Your AI experiment: Try this prompt

👉 Time to tinker: Pick a task you do repeatedly—a weekly report, a content review, meeting prep, formatting data, drafting a certain type of email. Open Claude (Chat or Cowork) or ChatGPT and work through it the way you normally would. Don't rush. Correct the AI when it gets things wrong. Push until the output is genuinely what you'd use.

Then, in that same conversation, run this:

Now save (Claude) or install (ChatGPT) the results into your instance so you can use it whenever you do this task. You can also download the .md (Claude) or .zip (ChatGPT) file(s) to use with another AI tool.

💡 Pro tip: Make sure it holds up past the first run.

A skill you use once is nice. One you still trust three months from now is the actual point. Here's how to get there.

• Give it a name you'll remember. "Review process" won't mean anything in six months. "Monday-inbox-triage" will. In Claude, a clear name means you can just ask for your Monday triage, or type "/monday-inbox-triage" and skip straight to it. No digging through old chats.
• Test it cold. Open a brand-new conversation, load your skill with a slash command, and run it on a fresh input. Whatever breaks tells you what the AI absorbed from context that never made it into the instructions. Fix the skill, not the output.
• Treat it as version one. Your process will improve. When you catch yourself making the same correction twice while running the skill, don't rebuild it from scratch. Just tell the AI what changed and ask it to update the skill. Skills are living documents, not stone tablets.

What did you discover?

What task did you turn into a skill? And here's the more interesting question: how much of your "process" turned out to be corrections you'd been making manually for months without noticing?

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.