what tasks should you give ai first?
most people give ai the wrong job, get a mediocre result, and decide ai does not work. the problem was the pick, not the ai.
the first task to give ai should be high-volume, low-risk, and easy to judge: something you do the same way many times a week, where a mistake is cheap, and where you can instantly tell good output from bad. that combination lets ai pay back the setup fast while you supervise it safely. this is the skill anthropic calls delegation, the first of the four AI fluency skills, and it happens before you ever open the chat.
why the first task decides everything
delegation is deciding what work is yours, what is the ai's, and what you do together. it needs two things: knowing your own goal, and knowing what the tool is actually good at. skip it, and you default to handing ai whatever is on your mind, which is usually the hardest, highest-stakes thing. that is exactly the task ai is worst at unsupervised, so you get burned and quit. the people who think ai does not work almost always started here.
the delegation test
run any candidate task through three questions. a good first task is a yes on all three.
- is it high-volume? you do it many times a week, the same way. repetition is what makes the setup worth it. a once-a-quarter task is not worth teaching ai to do.
- is it low-risk? a mistake is cheap and catchable, not a lost client or a public error. you want the training wheels on for the first one.
- can you judge it fast? you can look at the output and know in seconds whether it is good. if you cannot tell, you cannot supervise, and you cannot trust it yet.
good first tasks vs bad first tasks
| hand ai this first | not this (yet) |
|---|---|
| first-draft replies to common questions | a legal or medical decision |
| turning notes into a formatted summary | your one big pitch to your biggest client |
| repurposing one post into five formats | anything you cannot check for accuracy |
| sorting or tagging a list you review | a one-time task you will never repeat |
then set it up, do not just prompt it
once you have the task, the move is a setup, not a one-off prompt. give ai your voice, your real examples, and the rules you would give a new hire, so it starts on-brand every time. that is the description skill. and keep a human on the send: the ai drafts, you ship. prove it on one task before you expand. i wrote the on-yourself version of this in the ai audit, and why adoption fails without it in the ai adoption gap.
the takeaway
you do not need more ai. you need to point it at the right first job. high-volume, low-risk, judgeable. nail one, and the next task delegates itself.
want help picking the first task?
the free system builder walks you through delegation on your real work, so you set ai up on the right job the first time. built in claude, today.
get the free system builderor just follow along. new field notes most weeks on x, instagram, and tiktok.