If you’ve rewritten a prompt three times because ChatGPT kept missing the point, you’ve been practicing a skill you also need with people.
That’s the argument Mike Taylor makes in a recent column for Every. Mike runs AI evaluations at Every, a publication about AI and business. Before that he built a 50-person marketing agency. He’s managed both people and AI, and he says the skills are the same.
Why AI is good practice
Getting good work out of anyone, human or machine, depends on the same things: clear direction, enough context, and a well-defined task.
A benefit of practicing on AI is that it doesn’t keep score. You can ask it to redo something 15 times, give it fuzzy instructions, and change your mind halfway through. It won’t get annoyed, and it won’t gossip about how disorganized you are.
Mike points to some history here. In the 1880s, Frederick Winslow Taylor stood over factory workers with a stopwatch and broke their jobs into small, measurable tasks. His workers went on strike. AI doesn’t resent being asked to do the same task 50 times.
Five habits that work on both
Mike developed five principles for prompting. Here they are in my words, aimed at delegation:
Give clear direction. Describe what you want in detail. The more specific you are, the better the result, whether you’re briefing Claude or a paralegal.
Specify format. If structure matters, say what it should look like.
Provide examples. Show what a good finished version looks like.
Give feedback. Say what worked and what didn’t, as soon as you can.
Divide the work. Break a big job into smaller pieces and hand off each one separately.
If you already delegate well, this list may look obvious. It’s also how you get great results from AI instead of mediocre ones.
Where people still have the edge
Mike’s second point is about what AI still can’t do well:
The big picture. AI is good at this. Give it a pile of research and it can propose a reasonable strategy.
Precise execution. Also good. Clear instructions get carried out reliably.
The messy middle. This is where it struggles: deciding what to do next when the information is incomplete and the situation calls for judgment.
He offers an interesting example. When researchers had Claude play Pokémon, it understood the goal and could execute any single move. It still got stuck in the game’s first cave for more than 48 hours.
That middle ground is where a lawyer’s judgment lives. Deciding what to tell a client when you know you don’t have all the facts isn’t something to hand off to a chatbot. In many situations, careful human thinking beats fast machine prediction.
What to do with this
The next time AI gives you a mediocre result, look at what you asked for before blaming the tool.
Notice what fixed it. More context? An example? A clearer format?
Try the same fix the next time you hand work to a person.
Mike’s view is that many people are getting a crash course in management without realizing it. They think they’re just getting better at AI.
;-)
Ernie
P.S. In the Inner Circle, lawyers share what’s working with AI, so you don’t have to figure it out alone
→
https://ernietheattorney.net/