01The shift
The moment you use AI for real work, you become a manager.
Not a people manager. A manager of fast, capable, slightly overconfident systems that can do incredible work if you know how to direct them. And there’s actual proof this matters.
Ethan Mollick at Wharton ran a study with the Boston Consulting Group. Hundreds of consultants, real tasks. The ones who got 40% better results were not the most technical. They were the best at directing the AI. Delegation and direction, not prompting tricks. And it works on everything — code, campaigns, hiring. The context changes. The management part doesn’t.
02Practice 1
Treat AI like it works for you.
A lot of us use AI like a search engine. Ask a question, get an answer, move on. Fine for simple stuff. Here’s the upgrade: treat AI like it works for you — a really fast team member who just handed in a first draft. Because that’s literally what every response is.
How most people ask
Generic question. Generic answer.
- Prompt“What features should I add to the app?”
- ResultA list that sounds reasonable but could apply to any app.
- QualityNot bad. Just not managed.
How to direct
Context, constraint, pushback.
- Prompt“Here’s what my app does. Here’s who uses it. Here are the three things they complain about most. What are the highest-leverage things to focus on next?”
- Followup“And what would you push back on if I told you I want to build all three at once?”
- QualityCompounds. Every follow-up makes the next answer sharper.
Jim’s take
I learned this the hard way. I kept getting mediocre responses from Claude on a project. Reddit was full of “Claude got terrible.” Sometimes that’s real. Most of the time it’s not.
I went back and looked at my prompts. I was asking vague questions and expecting specific answers. The moment I gave it context and asked it to push back, the quality jumped immediately. Same model. Totally different results.
04Practice 2
Synthesize and challenge what comes back.
AI doesn’t give you one answer. It gives you options, drafts, competing ideas, half-right analysis. The skill that matters most isn’t generating. It’s making sense of what comes back.
Ask for five positioning angles for your product. You’ll get five options, all sounding pretty good. That’s the dangerous part. So instead of picking one, pull them apart. Where do they agree? Where do they differ? What keeps showing up across multiple options? What feels thrown in to fill space? That’s synthesis — one of the coolest parts of working with AI. You start seeing patterns the model doesn’t know it’s showing you.
AI is very good at sounding right. That doesn’t mean it is right.
Jim’s take
Push on it. “Critique this. What’s the weakest part? What would someone with ten years of experience disagree with? Give me a counterargument. What am I going to regret about this approach in three months?”
Then take the best parts, refine, and ask for the stronger version. You’re not accepting output. You’re shaping it. That’s where the real quality jump happens.
05Practice 3
For the bigger stuff: break it down, build it up.
A lot of people lose the plot here. They hand AI one giant request and wonder why the result came back mid. It’s like trying to install the kitchen cabinets before the foundation is poured.
Break the work into stages
Don’t try for the whole thing in one shot.
- Research“Most common approaches to converting free to paid in mobile apps. Range + pros/cons.”
- Synthesis“Best three approaches for a small team with 100k users.”
- Draft“Write the strategy for approach two. Keep it one page.”
- Review“Critique. What’s the weakest assumption? What falls apart if conversion is lower?”
Build it up in layers
Don’t polish something not built right.
- Layer 1Shape. Does this have the right pieces? Direction only.
- Layer 2Structure. Does the logic hold? Pieces in the right order?
- Layer 3Polish. Tighten language, fill gaps, apply your taste.
Jim’s take
I know this sounds slow. It’s actually faster. One-pass attempts almost always mean redoing the whole thing. The few extra minutes up front save you from a full rewrite later.
And it’s more satisfying. You can feel the work getting better at each step. You’re having a conversation with the AI. It’s helping you. You’re helping it. And you’re building something great.
