You’ve had AI at work for a while now. It answers fast. Some days it’s right. Some days it’s confidently wrong, and you can’t always tell which day it is.

Nobody sat you down and showed you how to check it. Or how to build something with it that does real work. Or how to hand it a whole job and still stay in charge of the result. You picked it up the way most of us did: one question at a time, between meetings.

That gap is what this newsletter is about. Welcome to the AI Operator Dispatch, from AI Command School™. Every Thursday: one argument, one framework, and one thing to try before Friday.

The skill didn’t arrive with the software

AI showed up at work faster than anyone could learn to run it. Most people now have it on their screen. Far fewer feel ready to rely on it. Survey after survey this year finds the same gap: far more people have the tools than feel equipped to work with them.

So most of the people around you are doing what you’re doing. Asking a question, reading the answer, and hoping.

And the training isn’t coming from the top. Employer-backed learning has been flat or shrinking while the expectations keep climbing. Your manager expects faster work because the tools exist. Nobody has budgeted the hours to teach you how to get that speed without the mistakes.

So you were handed a powerful tool, a deadline, and no manual.

What an operator does

There’s a difference between someone who types into AI and someone who operates it. The second person does four things.

Check. You look hard at what it produces before your name goes on it. Facts, figures, sources, tone. You know which parts it tends to get wrong, and you look there first.

Build. You stop asking the same question every Monday. You build a small tool that does a repeatable job the same way each time, with your rules written into it.

Deploy. You hand an AI agent a whole task, inside limits you set. What it can touch, what it can’t, and when it has to stop and ask you.

Govern. You answer for the result: deciding what goes out, keeping a record of how it was made, and owning the call when it’s wrong.

The short version:

❝

An operator directs the AI, checks what it produces, and answers for the result.

Typing a good question is step one. There are a lot of steps after it.

Your experience is the asset

The scarce skill right now is judgment about AI output. Microsoft’s 2026 Work Trend Index puts quality control of AI output at the top of the human skills growing in importance, with critical thinking right behind it.

Judgment is exactly what 10 or 15 years in a role builds. You already know what a good forecast looks like in your business. You know when a customer email sounds wrong. You know which number in a report would make your boss’s eyebrows go up.

That’s the hard part. The rest can be taught.

The people who’ll do well with AI at work will be the ones who pair it with real experience and a method for checking it. Everyone gets caught by a polished, confident, wrong answer sooner or later. No shame in it. The skill is catching it before it leaves your desk.

Why a newsletter

We’d rather write to you here every Thursday than hope a feed shows you a post.

Your inbox is the one place a feed can’t filter. So this is where the thinking goes first: the argument for the week, the framework behind it, and one thing you can try at your own desk before Friday.

Who’s writing

Two of us. Peter runs the business and the systems behind it. Andrew does the teaching.

Peter has spent thirty years managing teams from ten people to over three hundred, across Wall Street, Silicon Valley and global financial firms, leading major business and AI programs at asset managers overseeing more than a trillion dollars. He runs AI Command School on AI systems he built himself.

Andrew spent fifteen years as a British military instructor, then twenty-five in tier-one banks. Every session runs on EDIP, the British military method used since the 1950s, which he adapted for AI.

We’ve taught this inside a multinational corporate, in an IT function of more than 2,800 people. Now we’re opening it up to people who want to learn it on their own.

One thing to try before Friday

Take a question you asked AI this week, one whose answer you actually used.

Ask it again in a new chat, with memory off, in slightly different words.

Put the two answers side by side.

Sometimes they match. Often they don’t. A different number, a different recommendation, a name that has moved. Both were written with exactly the same confidence.

That is the whole problem, in two minutes. Not that AI is wrong, but that nothing on the surface of an answer tells you which kind of day it is having. Checking is how you find out. It is also the first thing that separates someone who uses AI from someone who operates it.

Come watch a live check

Operator Briefing 1: Check before you trust. Tuesday 10 November, 12:00 ET. Free, 45 minutes, live on Zoom with Andrew.

Andrew takes a real-looking AI answer and checks it step by step, start to finish, so you can see what an operator does before trusting it.

Hit reply with CHECK and we’ll send you the joining link.

The next cohort of the AI Operator Program starts Saturday 21 November. Applications close Monday 16 November.

Hit reply and tell us one thing AI got wrong for you this month. We read every one, and the best questions shape what we write next.

Peter & Andrew

❝

Peter Badger has spent thirty years managing teams from ten people to over three hundred, across Wall Street, Silicon Valley and global financial firms, leading major business and AI programs at asset managers overseeing more than a trillion dollars. He runs AI Command School™ on AI systems he built himself.

Andrew Badger spent fifteen years as a British military instructor, then twenty-five in tier-one banks. He teaches the AI Operator Program with EDIP, the British military method used since the 1950s, which he adapted for AI.