Name the one task you would hand off

The situation

It is Thursday afternoon. You are assembling the same report you assembled last Thursday: four numbers out of one system, three out of another, pasted into a document whose structure has not changed in a year, plus the two paragraphs of commentary that are the only part anyone reads. It takes ninety minutes. It took ninety minutes last week. Nothing about it is hard, and nothing about it is interesting.

Ask most people where AI should go first and they answer with the largest problem they have. That is the wrong end to start from. The right answer is smaller and duller — the thing you did last Thursday, will do again next Thursday, and could explain to a new hire in five minutes.

The principle

The first task worth handing to AI is one you repeat on a schedule, can describe end to end, and could afford to get wrong once.

Four tests, and the task has to pass all four. Recurring — it happens weekly or more, so the effort you put in comes back every week instead of once. Bounded — you can name where the inputs come from and what the finished thing looks like, without saying "it depends". Low-stakes if wrong — a bad draft costs you ten minutes of correction, not a customer or a compliance finding. Yours — you own the output, so you can change how it gets made without three other people signing off.

Tasks that fail a test are not bad tasks. They are later tasks. Board-level analysis fails "low-stakes". The quarterly planning cycle fails "recurring", because four attempts a year teaches you almost nothing. The report your director signs fails "yours", and you will spend more time negotiating than building.

Here is what the rest of this course asks of you: about three hours of reading and one hour of building, and one artifact per unit. Not slides. Every unit ends with something you keep — a working daily brief, a configured workspace, a one-page tool map, a build on your own role's work, a written ground rule for where your data may go. The capstone is an unaided hour on the task you are about to name, and it is graded on what you built, not on how much you read.

Try it

Give this five minutes. It is the only part of Unit 0 that matters.

1. List two tasks you did more than once in the last month and would not miss.
2. Score each against the four tests. Cross out anything that fails one.
3. From what survives, take the one that eats the most of your week and write it as a single sentence: "Every [interval] I turn [inputs] into [output] for [who reads it]."

Then write two numbers beside it: how long the task takes you today, and how many times you did it last month. Measure the first one against a clock this week rather than estimating it from memory. That sentence and those two numbers are your baseline, and they are the only things from this unit that anything later depends on.

You are done when you have one sentence and two numbers. Every exercise from here points back at that sentence, and at the capstone you will hold it up against the thing you built and the numbers against a clock. Vague sentences fail that comparison. "Every Monday I turn the regional sales export into a three-bullet status note for my director" survives it.

If you want a wider read on where your whole organisation stands, the readiness assessment is at https://aidolsgroup.com/en/ai-readiness-assessment/ — take it or skip it. The course does not use its output.