Using AI to Decide What to Work On

Overwhelm is rarely a shortage of ideas. It is the opposite: three or four plausible directions, all defensible, none obviously best. So you carry them all, make partial progress on each, and end the quarter having moved nothing far.

AI is genuinely useful here — not because it decides, but because it is patient enough to help you do the boring part properly.

Get it all out first

Start by emptying your head. Every direction you are considering, including the ones you have half-dismissed. Write them badly. Completeness beats polish here, because the option you have not written down is the one that will keep nagging.

Then ask AI to group them. You will usually find your four directions are really two, wearing different clothes.

Decide the criteria before you look at the options

This is the step that removes most of the paralysis, and almost nobody does it in this order.

Write down what "good" means before evaluating anything. What matters — revenue this quarter, or a foundation for next year? Work you enjoy? Something that does not need you personally? Something you could stop doing?

Rank those. Three or four is plenty.

Doing this second, after you have looked at the options, means you will unconsciously choose criteria that favour the one you already wanted.

Use AI to argue the other side

Now bring in the model, and use it against yourself. Give it your options and your criteria, and ask it to make the strongest case for the option you are not leaning towards.

This is the single most useful thing AI does in a decision. You already know your favourite's strengths. What you have not done is genuinely steelman the alternative — it is hard to do for yourself, and easy to fake.

Surface the assumptions you stopped noticing

Ask it: "What am I assuming here that I have not stated?"

The answers are often uncomfortable and usually right. "You are assuming demand will hold." "You are assuming you will still want to do this in a year." "You are assuming you can hire for this."

Then, for each one, ask what would cheaply test it this week. Most assumptions can be checked with a conversation or an afternoon, and it is remarkable how rarely anyone checks.

Make the call, and write down why

Decide. Then write a short note: what you chose, what you rejected, and what would have to be true for you to change your mind.

That last clause is the valuable one. It converts a decision into something you can revisit honestly in three months, instead of re-litigating it every time you have a bad week.

What AI must not do here

It must not choose. A model has no stake in your business, does not know what you can stand doing on a wet Tuesday in February, and will happily produce a confident recommendation based on nothing.

Use it as a thinking partner. The judgement is the part you cannot outsource.