Score the job before you hand it over
Five questions, zero to two on each, and a total that tells you what to do with it.
Most people pick their first AI task by what sounds impressive, it goes badly, and they conclude the whole thing is overhyped. The task was wrong, not the tool. Five questions, two points each, and the total tells you which of those three things to do.
What you'll walk away with
- The five questions, and what a zero and a two actually look like on each
- What the total means, and why the middle band is the honest answer for most work
- A prompt that interviews you about your week and scores every job in it
About six minutes to read, ten more if you run the prompt.
Start Here: Paste This
You can do this on paper, and the questions below are worth reading either way. But the slow part is remembering everything you actually do in a week, so this prompt interviews you first, then scores the lot and hands you the table. Paste it into whichever AI you use, nothing to fill in.
Argue with the scores it gives you. The interview only knows what you told it, so the first table is a draft and the useful bit is the conversation where you push back on two or three of them.
The five questions
Take one job you do regularly and score it zero, one or two on each. Two always means hand it over, so a high total is always the same instruction, which is the bit that makes the maths work.
Is it worth it
How often does it happen, and how long does it take? This is the only question about size, and it is first because a job that comes up twice a year is not worth building anything for, however well AI could do it. Frequency matters more than duration here. A ten-minute job you do every morning beats a two-hour job you do each quarter, because the ten-minute one is where the compounding is.
Can you teach it
Could you show someone the whole thing in a ten-minute screen share? If you can, you can write it down, and if you can write it down you can hand it over, because a written instruction is the only thing AI can actually work from. If the job lives entirely in your head and you have never once articulated it, that is a zero until you do. Not forever, just today.
Can you check it
How long does verifying the output take against just doing the job yourself? This is the one everybody skips, and it is the one that decides whether handing the job over actually saves you anything. If checking costs nearly as much as doing, you have not delegated the work, you have just moved it and added a handover.
What if it is wrong and nobody notices
Not what happens if it is wrong, what happens if it is wrong and it goes out anyway. Every other question assumes somebody is paying attention. This one assumes nobody was. An internal first draft scores well because the cost of a bad one is that you write it again. Anything landing in front of a client, a regulator or your accounts scores badly, and it stays badly no matter how good the AI gets, because the exposure is yours.
Are you the reason it is any good
Would a competent stranger produce much the same thing? If yes, there is nothing personal in the output and you should not be spending your week on it. If the whole value is your judgement, your relationships or your name on the bottom, that is the work you are actually paid for, and handing it over is how you become worse at your job rather than faster at it.
What the total means
8 to 10
Deputise
Hand it over. Set it up properly once, spot check the first few, and then leave it alone. If you are still reading every output a fortnight later, one of your scores was generous.
4 to 7
Duet
Split it. Look at which questions scored low and give AI the part they do not apply to. Usually that means it drafts and you decide, or it does the gathering and you do the call.
0 to 3
Defend
Keep it. Either the mistakes are expensive or the person on the other end is expecting you, and neither of those is a problem a better prompt solves.
Most real work lands in the middle, and that is the honest answer rather than a disappointing one. The version of AI that helps most people is not a job disappearing, it is four of your jobs getting a first draft done before you sit down.
The one that catches people out
Question three, every time. Plenty of jobs are quick for AI to do and slow for you to check, and those feel like wins right up until the fortnight where you are proofreading everything it gives you and wondering why you have not got any time back.
Reconciling a spreadsheet against a statement is the clean example on the other side. It takes ages by hand, and checking it is one number matching one number, so verification costs you nothing. Compare that with a summary of a long report, which arrives in seconds and can only be trusted by reading the report.
The question to actually ask
What this will not do
It will not tell you whether AI is capable of the job. That is a separate question and it changes every few months, so anything I wrote here about capability would be wrong by Christmas. This scores whether the job is worth handing over assuming AI can do it, which is the part that stays true.
It is also not precise, and you should not treat a seven and an eight as meaningfully different. The value is in being made to answer question three and question four out loud, because those are the two nobody asks themselves before they start, and they are the two that decide whether it works.
Where this came from
This is Nathaniel Whittemore's, from the AI Daily Brief, where he calls it the deputisation audit. He named the five criteria and the three bands, and I thought it was the tidiest way of thinking about this I have heard, so it is here with credit rather than quietly absorbed.
What is mine is the working detail. He does not spell out what a zero, a one and a two look like on each question, and on two of them a high score has to mean low stakes for the total to add up, so I have written the anchors the way I actually use them. If you hear the original and score something differently, go with yours.
Next
- Replace line items, not job descriptions is the idea underneath this one, that your job is a bundle of tasks and only some of them are going anywhere.
- The two objections that stall AI projects covers the arguments you will hit the moment you try to hand something over at work.
- Three rules for writing with AI is question four as a rule of thumb, for anything you are putting your name to.