What not to hand to AI in your business, and what you can give it freely

How to read the split

Look at the table row by row. The left column is real work, and a lot of it. AI is very good at all of it, and the book says to hand it over.

The right column is smaller. Each item is a moment where a person has to stand up between the AI's output and the action. Picking. Explaining. Checking the facts that matter. Naming what you will not do.

Notice that both columns appear in every row. You are not choosing between AI tasks and human tasks. Inside almost every task, there is a part the AI does and a part that has to stay with you.

The last row is the odd one. It is not about any one job. It is about keeping your own skill alive, the way a pilot flies by hand now and then so the autopilot is not the only one who knows how.

Why "keep the important stuff" is not enough

The usual advice says let AI do the routine work and keep the important work. It sounds right. It does not hold up.

The trouble is that the handover is a slope, not a step. A tool does one thing better than you, so you let it. Then it does the next thing, right next to the first. Each handoff makes sense. None feels like giving up. One day you notice you only approve.

There is a second trap. The AI's sure tone does not change when it moves past what it knows. In a study of 758 consultants, people using AI on a task just past its range were about 19% less likely to get it right than people without it (Dell'Acqua et al., Organization Science). So you cannot sort tasks once and be done. The line moves.

That is why the split above is about parts of the job, not whole jobs.

The Thursday the work was great and felt empty

In the book, Jonah sends a client a sharp, clean deck. The AI built most of it in under an hour. By every number, it is a good day.

Then he sits in the quiet between two calls and thinks, "Whose work is this, and why am I the one doing it?" He has no answer.

Nothing was wrong. The numbers were up. That is the point. When you hand over the why, you do not slow down. You speed up, toward whatever is easiest to start, because there is nothing left to rank the options by.

So he writes one page with four lines. What the business does. Who it is for. What he refuses to do for money. Why it matters to him. The third line made him put the pen down twice, because he could think of three times he had broken it.

After that, he used the AI for more than ever. He just held its suggestions up against those four lines first.

A quick test before you hand over a task

The table splits each job into parts. When a new task shows up and you are not sure, four questions help. They match the checklist in an OpenClassrooms course on working with AI, and they fit the book's line well.

1. Can I tell if the result is good? If you have no way to check it, keep the final say. That covers a topic you do not know well enough to judge.
2. Is a mistake cheap and easy to fix? A draft social post, yes. A price to a big client, a contract term or anything you cannot take back, no.
3. Does it involve private data? Client details, staff records, health or money information. Leave it out unless you truly need it, and check what your AI tool and your client contract allow.
4. Is a person on the other end who needs you? Letting someone go, a hard talk with a client, an apology. AI can help you plan what to say. The saying, and the care in it, stay yours.

Four yes answers in the safe direction mean hand it over and check the result. Any no means AI drafts and you decide.

The hand-off rule that holds up on a busy day

Here is the rule in one sentence: give the AI the work, keep the deciding.

In practice, that is one small habit before anything important goes out. Ask yourself, "Do I actually agree, or am I just tired and it sounds sure?" If you cannot answer, it does not go yet. I use AI every day and trust it with a great deal. That question is the part I do not hand over.

The skill side matters too. A 2025 study of 666 people found heavy AI use went along with weaker critical thinking (Gerlich, Societies). It is a link, not proof of cause. The book's answer is not less AI. It is staying the one who decides while you use it.

The book that draws this line

This is the problem AI Won't Run Your Business. You Will. was written for.

The Decision: AI Offers; You Choose, and You Live With It explains why the choice cannot be handed over: the one who chooses carries the result, and the AI carries nothing.

The Why: AI Does the What and the How; Only You Hold the Why gives you the Why Anchor, the four-line page Jonah wrote.

The Person You Become: Don't Disappear Into the Tools has the Agency Audit, which shows where you have slipped from deciding into just approving.

The free Judgment Workbook has all of these as printable sheets. For the trust side of the question, see can you trust ChatGPT for business decisions.