Thinking and Deciding with AI

An assistant never gets tired of generating options, objections and scenarios, which makes it a useful thinking partner. It also tends to agree with you, to sound sure of itself and to make a bad idea look polished. In this lesson you will learn techniques for brainstorming, challenging your own ideas and comparing options, with one rule above all: the AI proposes, you decide.

We have used AI to find information, organize work and handle data. Now to the hardest part: using it for reflection and decisions, where the risks are subtler.

Marc's van decision

Marc, the plumber, is thinking about replacing his old van, with three options: a used van, a new electric vehicle, or a leasing contract. He could ask an assistant, "Which should I choose?" and receive a confident answer. But the answer would depend on many things the assistant does not know: his mileage, his savings, his local charging options, his tolerance for risk. A better use is to structure his thinking and keep the choice for himself.

Technique 1: brainstorming

Brainstorming is where AI shines: quantity and variety. Ask for many ideas, then filter. A few tips:

• Ask for quantity first: "Give me 20 ideas", then select and refine.
• Force variety: "Give me 5 conventional ideas, 5 unusual ones and 5 that cost nothing."
• Provide constraints: budget, time, audience, what you have already tried.
• Build on your own ideas: write yours first, then ask for more. Otherwise the AI's suggestions anchor your thinking.

A ready-to-use prompt: structured brainstorming

I am looking for ideas to [objective, e.g. reduce the number of late payments by clients].
Context: [your situation in 3 to 4 sentences, no confidential details].
Constraints: [budget, time, tools, things already tried].
Give me 15 ideas in three groups: 5 simple and quick, 5 more ambitious, 5 unconventional. For each, one sentence on how it would work and the main downside.
Do not evaluate yet and do not invent data. At the end, ask me 3 questions that would help to choose.

Technique 2: devil's advocate

The best use of AI in decisions may be to ask it to disagree with you. Present your plan and request the strongest objections. This works because the assistant does not mind upsetting you, and because it can take any viewpoint on demand.

Here is my plan: [plan in a few sentences].
Act as a skeptical but fair advisor. Give me the 5 strongest objections to this plan, the assumptions it depends on, and the evidence that would convince you it is a mistake.
Then tell me what would need to be true for the plan to work. Do not flatter me and do not rewrite my plan.

Technique 3: SWOT and pre-mortem

A SWOT analysis lists strengths, weaknesses, opportunities and threats. An assistant can produce a first version, which you then correct with your own knowledge; a generic SWOT is rarely useful, so give specifics.

A pre-mortem is more powerful. You imagine that the project has failed a year from now and ask why. It is easier to find problems when you assume they happened. Julien uses it before changing the way his firm handles client documents: "Imagine this change failed. What are the ten most likely reasons?" Several answers (clients ignoring the new portal, partners reverting to email) are specific enough to prevent.

Imagine that it is [12 months] from now and the following project has clearly failed: [project].
Write the 10 most likely reasons it failed, from the point of view of different people: [roles, e.g. the team, the clients, the management].
For each reason give an early warning sign and a preventive action. Then rank the reasons by likelihood and by impact.

Technique 4: compare options with criteria

When you compare options, the quality of the criteria matters more than the quality of the prose. Define your criteria and weights before asking for an evaluation, and keep your own values in the process.

1. List your options.
2. Define 4 to 6 criteria that matter to you and give each a weight.
3. Ask the assistant to build the comparison table and to point out information that is missing.
4. Fill in the facts yourself, from verified sources.
5. Look at the result, then test it: does the winner feel right? If not, which criterion is missing?

Criterion (Marc)	Weight	Used van	Electric van	Leasing
Total cost over 5 years	High	To fill in with real quotes	To fill in	To fill in
Fit with daily routes	High	To check	To check	To check
Cash flow constraints	Medium	To fill in	To fill in	To fill in
Risk of breakdown	Medium	To check	To check	To check

The "to fill in" cells are deliberate: the assistant should not guess prices or ranges. Marc collects real quotes and enters them.

Biases: yours and the tool's

Two biases deserve particular attention in this lesson.

• Confirmation bias: the tendency to look for arguments that support what we already believe. Assistants tend to agree with the user's framing, so a question such as "Why is the electric van the best choice?" will produce arguments for it. Always ask the opposite question as well, and use neutral wording ("compare the advantages and disadvantages of each option").
• Overconfidence: a well-written, structured analysis looks more reliable than it is. The polish is not evidence. Ask for assumptions, uncertainties and what would change the conclusion.

Also be careful about anchoring: the first option or number the assistant gives tends to stick in your head. Generate your own view first when you can.

Bad use versus good use

Bad use	Good use
"Which option should I choose?" and follow the answer.	Define criteria and weights, collect facts, use AI to structure and challenge, then decide.
Asking only for arguments in favor of your preferred plan.	Asking for the strongest objections and what would prove you wrong.
Using generated numbers in a comparison.	Filling facts from real quotes and sources; leaving unknowns visible.

Your turn: challenge a real decision

Exercise. Take a decision you must make in the next weeks. (1) Write your own list of options and criteria first. (2) Use the brainstorming prompt to find options you had not considered. (3) Use the devil's advocate or the pre-mortem prompt on your preferred option. (4) Build the comparison table with weights, and fill in the facts yourself. (5) Write your decision and the three reasons for it, in your own words.

See a solution

Marc's result: his first preference was the used van. The devil's advocate listed repair costs, the risk of downtime during busy winter weeks and uncertainty about future access restrictions in some cities. The pre-mortem for the electric option highlighted charging at home and on long trips. The comparison table, once filled with real quotes, showed that the leasing option was the best on cash flow but the worst on total cost.

He decided to keep the used van for one more year, set aside savings, and revisit with updated quotes. The decision was his, and he could explain each reason. The AI contributed structure and challenge; the facts and the values came from him. That division of roles is the model to keep.