Everyone says "you have to use AI." You open ChatGPT, stare at the blank field, then close it. You don't know what to type.
That stuck isn't because you're slow. It's a very real bind: the blank field offers no prompt, so your mind jumps straight to the biggest, most impressive job — write a whole article, draft a whole plan — and then jams, because big jobs are hard to describe, hard to check, and embarrassing to get wrong. You blame yourself for lacking the skill. Really you just picked the wrong place to stand.
And here's the cruel part: that blank field gets scarier the longer you leave it. Every time you open it and close it without typing, you reinforce the belief "this probably isn't for me." It isn't a skill problem. It's a job-selection problem — and job selection is something you can learn in five minutes.
01The first job you think of is usually the worst one to begin with
Instinct steers you toward the most impressive task — because that's the work you hope AI can do. You want to see it handle something hard, to find out how good it really is. But impressive work is usually vague work: no clear right answer, full of constraints only you know, and checking whether it did well costs about as much as doing it yourself. Start there and you're all but guaranteed one disappointment — and then the verdict "AI just isn't for me."
The job worth starting with is the opposite on every count:
✕ The job you mean to start with
✓ The job worth starting with
The right column isn't as impressive as the left — but it's where you learn the craft. Save the left column for when your hands are practiced.
The right column wins not because it's easy, but because it teaches you fast. A small, clear job gives you a near-instant feedback loop: hand it over, see the result, know at once whether it's fine. The big vague job is the opposite — you hand it off, wait a long time, get back something whose correctness is unclear, and lose half a day just deciding whether to trust it. One gives you ten tries in an hour; the other gives you one try you can't even learn from with certainty. The skill of working with an agent is learned through fast repetition, not one big gamble.
02One question to find that job in your week
Don't sit there thinking abstractly about "where to use AI" — that question leads you straight back to the blank field. Ask one concrete question about last week: what did I do over and over that I'd feel slightly embarrassed describing, because it's so mechanical?
The report stitched from a few sheets. The template reply you send again and again. The list you re-sort every morning. The numbers you copy from here and paste over there. Those jobs don't need your mind — they're just occupying its place. That little "embarrassed it's so mechanical" feeling is the signal itself: any job you feel slightly ashamed to admit you do by hand is almost certainly a job to hand off. Pick exactly one. Just one — don't get greedy, because the whole skill is in running one loop all the way through, not opening five jobs at once.
03The three-step loop that makes the first handoff a lesson
Once you've picked the job, don't expect it to be perfect on the first try — and that's normal, not a sign of failure. What you're learning isn't "how to give the perfect order," it's a loop. The loop has only three steps, and it's the whole craft:
Step 2 — Look, against the example: it finishes, you glance for a minute — what matches the example, what did it misread? Don't fix it for it, just note the gaps.
Step 3 — Adjust the brief, hand it back: wherever it drifted, add one line to make that part clear. Two or three rounds and it's in the groove — and that brief is reusable forever.
What you actually build across these three steps isn't a result — it's a honed brief. Next time you just paste that brief back, and the job runs itself.
Notice step three: you don't fix the output, you fix the brief. That's the difference between a beginner and someone with practiced hands. The beginner fixes each result, one at a time, forever. The practiced one fixes the brief once, and that brief then does the work for them from there on. The weekly report, after three rounds, becomes a line you paste and you're done — and you've just won back your Friday afternoon.
04Step through this narrow door, and the road behind widens
The good thing about starting small is that it doesn't stay small. Once you've handed off the weekly report smoothly, you'll spot three other repetitive jobs worth handing next. Then you'll hit the moment you need to brief more carefully, the moment you need it to stop for your approval mid-run, the moment you need to verify before you trust. Each of those is a step up, and you take them one at a time — exactly as you just took one step up through that first report.
That's also why you shouldn't try to learn it all before doing anything. You don't need to know "how AI works" to hand off the first report — the way you don't need to understand an engine to practice driving in an empty lot. Hand off the small one first; the road behind opens on its own, and opens right when you need it.
And when you're ready for the bigger work — real work, with stakes, that has to be done properly — a deliberate method is waiting: four checkpoints for handing over a real job so it doesn't break things mid-run. That's the process for shipping real work with an agent. But don't rush there. Hand off this week's report first — the blank field is only frightening up until you type the first line into it. After that line, it's just a chat box.