x
This, before any job longer than a paragraph:
"Before writing, give me your plan in five bullets — what you'll cover, in what order, and what you'll leave out. Then wait for my go-ahead."
That's it. You get a five-bullet plan back. You correct it in about fifteen seconds — usually one bullet is wrong, and it's the one that would have ruined the whole output — and then you say go. Same as you'd do with any supplier quoting for a job: you look at the quote before they start, not after they've done the wrong work beautifully.
The reason it belongs in a week about metered pricing is that the plan is cheap and the work is billed. You are buying a look at the intention before you pay for the execution.
Because of the size difference, and it's not marginal. A five-bullet plan is a few hundred tokens. A full drafted piece of work — a client report, a long reply, a document rewrite — runs into the thousands, and a typical customer conversation on a metered agent sits at roughly 20,000 to 25,000 tokens, billed at $2 per million on Meta's Business Agent since 1 August 2026.
So the plan costs a small fraction of the run. If it catches a misunderstanding one time in three, it has already paid for itself several times over — and in practice it catches something far more often than one time in three, because most bad AI output isn't a failure of writing. It's a failure of assumption, made in the first sentence and then executed faithfully for six paragraphs.
The tokens are the smaller half of the saving. The larger half is your time: reading a wrong 800-word draft to work out where it went wrong takes several minutes; reading five bullets takes fifteen seconds. That asymmetry is why this habit survives even on tools you pay a flat fee for.
The plan comes back and you're looking for three specific things, not general quality:
A wrong audience assumption. It's written the client update for the client's technical team when it's going to a finance director. Everything downstream would have been wrong in tone and detail.
A missing constraint. It hasn't mentioned the deadline, the budget, or the thing you're deliberately not saying yet. Constraints omitted at plan stage reappear as sentences you have to delete.
A structural mismatch. Five bullets that are all background and no recommendation, when what you needed was the recommendation first. Fixing order at plan stage is a word. Fixing it in a finished draft is a rewrite.
Say what's wrong in one line — "swap bullets 1 and 4, cut the background, it's for a finance director" — and then go. Don't re-explain the whole brief; you're adjusting a quote, not re-tendering.
One thing worth watching for: if the plan comes back and you can't tell whether it's right, that's information about the brief, not the plan. It usually means you haven't decided what you want yet — in which case the model was about to guess on your behalf and you'd have spent the run finding out what it guessed. Two more sentences of brief at that point is the cheapest work in the whole process.
Honestly: for anything short. A quick rephrase, a one-line reply, a spelling check — asking for a plan first is friction with no payoff, and adding ceremony to trivial tasks is how good habits get abandoned.
Use it when the job has length, stakes or an audience. Client-facing work, anything you'll send without heavy editing, anything long enough that you'd be annoyed to redo it. Roughly: if the output would take you more than two minutes to read properly, get the quote first.
It pairs naturally with a standing brief. If you've written the one-pager that doubles every AI answer, the model already knows who you are and how you write, so the plans come back closer to right and the corrections get shorter. The one-pager fixes the defaults; the quote catches the exceptions.
This is the practical end of costing work per outcome rather than per seat. Every re-run is billed now, so the cheapest possible edit is the one you make before the work starts — and the second cheapest is editing what came back rather than saying "try again", which throws away the good parts and re-bills you for reproducing them.
Three vague attempts cost three times one clear brief and usually produce worse work. Asking for the plan is just the version of that you can do in one line, without having to become a better prompt writer first.
Try it on the next long thing you ask for today. The fifteen seconds are the whole investment.
Not financial advice — token rates and usage vary by vendor and by task; check your own billing dashboard.
For the full prompt set we run a one-person business on — free, 5-min read: 7 AI Prompts That Run a One-Person Business.
Related Decodes
AI the Brief (@aithebrief) — AI advice you can act on and trust, for time-starved founders.
TL;DR: Paste one line before any long AI job — "Before writing, give me your plan in five bullets, then wait for my go-ahead" — and you get a cheap, correctable plan instead of an expensive finished draft built on a wrong assumption. A five-bullet plan is a few hundred tokens against roughly 20,000–25,000 for a full run, so catching a misunderstanding at plan stage pays for itself many times over in tokens and far more in your own reading time. Skip it for short tasks; use it whenever the job has length, stakes or an audience.
