The part of AI you're already good at

Hi {{first_name|friend}},

This week I want to show you the idea behind the workshop I'm teaching Wednesday, because it's useful whether or not you sign up. It's about getting AI to do the same task well every time and why you're better set up for that than you might think.

My Take

Many people use AI one chat at a time. You type out what you want, explain the background, fix the first answer and end up with something decent. Next week you do it all again for the same task, because the chat doesn't remember how you like it done.

A Skill fixes that. It's a one page document, a recipe your AI follows the same way every time. You write it once: what the task is, what you'll hand it each time, the steps and what the result should look like. Then you give it new input and it runs. It's a document you own, so it works in ChatGPT, Claude, Gemini or whatever comes next.

I run on these myself. One I built last fall keeps tabs on the competition in my advising business. It's installed in Claude, so I just type "Run weekly market intelligence." It checks the companies I watch for news from the past week or so, compares what it finds against its notes from earlier runs and tells me what's actually new, including companies I wasn't watching yet. Then it suggests what to add to those notes for next time. It replaced my manual searching and one off chats. Swap my competitors for yours and it works in any field, like a realtor watching new listings nearby or a shop owner keeping an eye on the store across town.

The part most people skip is the last line of the recipe, the quality check. That's where you write down what "done well" looks like and what a miss looks like. The Skill checks its own work against it before you see it. That line is hard to write if you've never done the work and easy if you've done it for 20 or 30 years. You already know what a good client recap sounds like and which number in a quote looks off. Writing that down is most of the job, and it's the part AI can't do for you.

In the workshop I use a made up furniture shop with three product lines. They're stretched thin and asking whether to focus on one. A regular chatbot told them to make custom chairs their growth priority, since chairs brought in the most revenue. A Decision Brief Skill, given the shop's details and the owner's own criteria, said to wind the chairs down: the outsourcing was eating the margin and the line was eating the owner's week. It was the same question both times. What changed the answer was the Skills recipe and what it knew about the shop.

The most useful thing you bring to AI is something you already have, knowing what good looks like in your own work. Put it in writing once and your AI can follow it every time.

(More details about the Skills Workshop and the sign up link are below.)

Brett

Try This Today

Build a small Skill in about 10 minutes. Copy this into any AI, then paste a letter, bill or policy under it. Cover any account numbers first.

Skill: Explain This

What it does: I paste a letter, bill, contract or policy. You tell me in plain English what it says, what I need to do and by when.

Steps: 1) Sum it up in 3 sentences. 2) List anything I have to do, with dates. 3) Flag anything unusual or worth a phone call.

Quality check: Before you answer, make sure every action has a date or says "no deadline," and nothing is left in jargon.

If it works, save it in a doc named "Skill - Explain This." The next time a confusing letter shows up, paste the doc and the letter together and it runs the same way.

Watch: why I agree with Sam Altman's warning about handing AI your judgment

@aiover50

I agree with Sam Altman on this one: handing your judgment over to AI is a real safety issue. It happens in small steps. My check: ask it ... See more

Worth Knowing

Workers are exaggerating how much AI they use

A new report finds a wave of employees are padding how much they lean on AI at work, telling the boss they use it more than they really do, because looking fluent has become its own workplace signal. Plenty have barely touched the tools they claim to run.

The pressure to look current is real at our age, but faking it skips the part that builds actual skill: using the tools with your eyes open and learning where they fall short. Read more

Scammers used AI to clone a new book about AI on Amazon

Days after journalist Kevin Roose released his buzzy new book about the AI industry, counterfeiters used AI to spin up lookalike versions and sell them on Amazon next to the real one. The fakes were up within days of launch.

Run the same check here you would on a shady email: look at the seller, the publish date and the reviews before you buy. Read more

Can AI now read your mind?

Researchers built a system that can figure out what a person is thinking about by reading a brain scan. It is early and narrow, not the full mind reading the headline suggests, but it works, and it pushes the privacy question into new territory.

Nothing for us to act on today but worth keeping an eye on as you decide how much of yourself any tool gets to see. Read more

Tell me

What's the first task you'd hand to a Skill? One line is fine, I read every reply.