AI in Management: The Tool Computes, the Owner Decides

Thursday afternoon, an email from your main supplier: prices go up next month.

You run a small online shop. You paste the email into an AI chat and ask for help. Three paragraphs later it has raised your prices, swapped your supplier and cut your range.

You asked for help. You got a new owner.

Nobody planned it that way. The assistant answered the question it was given, as fully as it could. And a tired owner at the end of a long day is glad to let someone else hold the wheel for a minute.

That minute is where businesses drift.

AI in management: what a tool means on a Thursday

A drill does not decide where the shelf goes. A calculator does not decide what to sell. Nobody needs reminding of that.

With AI, people forget. The tool writes in full sentences. It sounds calm. It answers faster than any adviser you could hire. So its proposals start to feel like decisions.

AI is a tool. So the useful question is not how smart it is. The useful question is who decides what. Get that split right, and the assistant becomes the best instrument on your desk. Get it wrong, and it quietly steers the business by its own enthusiasm.

The who-decides table

Here is the split, row by row. Fill in the right column for your own business tonight. The middle column shows the online shop, as an illustration.

Read the table once more and look at the pattern. The AI owns the arithmetic. You own every row that carries a value judgement or a door that only opens one way.

Three checks before you act on an AI answer

Run these on any reply before you move money, people or stock.

1. Is every number mine or shown arithmetic? If a figure appears that you never gave and cannot trace, stop. Ask where it came from.
2. Is the order of goals mine? Say the plan puts growth first, and you put cash first. Then the plan is for someone else's business.
3. Can I undo this step? If yes, try it and measure. If no, decide it yourself, with your own numbers, after a night's sleep.

Three checks. Two minutes. They keep the instrument an instrument. Call it the smallest decision making framework for AI.

What a tool at the wheel costs

Hand the wheel to the tool for six months and nothing explodes. That is the trouble.

You raise prices because the reply said so. You drop a product because the reply called it weak. Each step feels checked, because a confident voice agreed. None was checked against what you would give up last.

Then a Monday comes when the stock order and the tax bill land in the same week. You find out the order of your goals was never yours. It was the order of the last reply.

The tool did not take the wheel. You let go of it, one helpful paragraph at a time.

"But the AI knows more than I do"

It knows more facts. It does not know your business.

It does not know that the shop pays your daughter's tuition, or that you promised your one employee a stable winter. It does not know what you would defend to the end. That knowledge is the steering. The facts are the engine.

A good AI knows this about itself, but only if you give it rules that say so.

A system that keeps the split

The split in the table is not something you must remember every time. You can hand it to the assistant as rules.

The Jirgenson Model has two plain-text files for this, free on GitHub, with setup steps on the AI assistant page. The canon is the rulebook of the model. The operating prompt tells the assistant how to work with you.

With both loaded, the assistant keeps to its side of the table:

• It refuses to start without the canon.
• It helps first: what your numbers show, then the first safe action.
• Its rules forbid invented data; assumptions are marked.
• It proposes an order of your goals and lets you confirm it.
• It names the most dangerous thing.
• It gives two forecasts with ranges.

Three modes match the moment: emergency for one action today, quick for this week, full for a complete analysis in rounds. It works in ChatGPT, Claude and Gemini.

We tested it blind against a good generic consultant prompt; the details are on the AI assistant page.

Next Thursday, the supplier email arrives again. You paste it in. The reply opens with your new margin, asks one question about your goals, and leaves the decision on your side of the desk.

Where the split does not reach

The table fixes a symptom: the tool drifting into the owner's chair. The rules guide the assistant; they do not make it perfect. Check its numbers against yours.

The deeper cause sits in your own column. Two rows there decide how good the split can be, and the assistant cannot fill them for you.

• An honest state, one source per number. If your figures disagree with each other, the best instrument measures noise. Chapter 2.
• When to change the plan and when to change the goal. A bad month is a reason to change the plan. It is rarely a reason to change the goal. Chapter 10 holds the exact rule.

Without them, you keep the wheel but drive on a foggy map.

Two choices at the wheel

You have two choices. Keep asking an assistant that agrees with everything and slowly takes the wheel. Or set up the system tonight: the canon and operating prompt are free.

The files keep the tool a tool. An honest state and a clear rule for changing goals keep you the owner.

Test the logic for free: Chapter 1, with the refusal test. Or get the full control system on Kindle, on Amazon. It comes as a Kindle book, with the canon and the prompt in its appendices.

Published research

It is scientifically proven that the Jirgenson Model works. The published experiment, its record and its data:

• Zenodo (DOI): 10.5281/zenodo.23964934
• Hugging Face: E5 Jirgenson Model blind test
• Kaggle: E5 blind test dataset
• GitHub: experiment folder
• Wikidata: dataset Q142022866, author Q141981336
• ORCID: 0009-0001-3034-3552