Hi {{first_name|friend}},

Thursday I taught a free live lesson on how to actually talk to AI. Thirty minutes with a sharp, slightly skeptical group of you. The part that stuck wasn't anything I'd prepped. It was watching two AIs disagree in front of the whole room. Here's what happened and the four habits I want you to take from it.

Here's the moment. I had ChatGPT build a growth plan for a semi retired consultant looking for a few more clients (via voice dictation). It came back with something reasonable. Looked solid to me.

Then I copied that plan into Claude, a different AI, and asked one thing: be critical, tell me what's weak. Claude caught the first plan contradicting itself, pointed out there wasn't a real number anywhere in it and flagged that the 90 day timeline didn't match how that kind of business actually sells. The second AI found the holes.

That's the lesson in one snapshot. The first answer you get is almost never the best one. It's the fast, agreeable one that runs a little generic. The value is in what you do next.

I gave the room four habits for exactly that. None of them are settings you switch on. They're things you do inside the conversation (ie human habits).

What makes this stick is running it on real work. One person who took my first lesson came back Thursday and mentioned they'd already reorganized their whole AI setup around what they learned. This is the way! Small habits, run often, on the work you already do.

Someone also asked a great question: what about giving AI a role, telling it to “act as an expert”? Honest answer, that trick has mostly stopped working. Two years ago we built elaborate personas to get good output. The models got good enough that it barely changes the answer now. What moves it is the conversation you lead.

And that part you already own. You've spent thirty years leading capable people through messy problems, asking the pointed question, pushing back when something feels off and not signing off until it's actually good. AI is just the newest person in the room and you already know how to run that conversation.

Brett

Next time you ask AI something that actually matters, don't take the first answer. Before it responds, paste this:

Before you answer, interview me. Ask what you need to know to give me an answer that fits my situation, one question at a time.

Answer its questions, then let it respond. You'll feel the difference right away.

Watch: do you even still need to be good at prompting?

Gallup's Q2 numbers just crossed a line: 47% of US employees say their organization has folded AI into how it operates, up six points in a single quarter, and 52% of workers now use it in their actual role. The biggest gains showed up in coding, analytics and everyday automation.

When half of companies have formally adopted AI, sitting it out stops being invisible, it becomes a position other people notice. Read more

A Nikkei Asia investigation estimates that Alphabet, Microsoft, Amazon, Meta and Oracle are carrying roughly $1.65 trillion in off balance sheet debt tied to AI data centers, more than the debt they report openly. Analysts are already using the Enron comparison and its special purpose vehicles.

If you lived through Enron or the dot com bust, you know this pattern in your gut. It's a good nudge to bet on the skills you're building, not on any single company's promises. Read more

UC Berkeley researchers built a benchmark they nicknamed the “Agents’ Last Exam” and put frontier AI models through real workplace tasks. Despite something like $1.6 trillion poured into the industry, the models still failed the large majority of them at any reasonable bar.

This is the calm counterweight to the replacement panic. AI handles tasks, not whole jobs, and it still botches most of those tasks unless someone who knows the work is checking. That's this week's lesson with real life data. Read more

Which of the four habits will you try first? One word is fine, I read every reply.