The one-page procedure that lets you hand off any task

Every small business has a dozen tasks that exactly one person knows how to do properly: onboarding a new client, putting together the Thursday order, closing out the till, chasing an invoice that's gone past due. They work because that person is around. The day they aren't — holiday, sick leave, or simply buried in something else — the task stalls or goes wrong. The fix isn't a hundred-page manual nobody will read. It's one page per task. And it barely matters whether the person following it started on Monday or is an AI agent, because both need roughly the same thing.

A task that only lives in your head can't be handed over

Try to hand off a task you've never written down and the same thing happens every time. You explain it roughly, they do it their way, you correct it, and by the third round you decide it's faster to do it yourself. It isn't that they're slow or that the task is hard. You gave them 60% of the information; the missing 40% is still in your head, in the form of exceptions.

That 40% sounds like this: long-standing clients don't pay up front; that supplier needs a phone call, not an email; if the order goes over €500, check with me before confirming; in August we don't send reminders on Friday afternoon. None of it is written anywhere. It gets learned by getting it wrong — and the person learning it is you, again, because you end up reviewing every attempt.

Writing that 40% down takes twenty minutes once and saves you explaining it twelve times. It's the difference between having someone who executes and having someone who asks. And with an AI agent it isn't optional: if it isn't written, it doesn't exist.

The six boxes that fit on one page

A useful procedure isn't an essay. It's six short blocks — and if they don't fit on one side of a page, you've squeezed two tasks into one:

1. What sets it off. The exact trigger, not "when needed". A day and a time, an incoming message, an order that gets confirmed.
2. What you need before you start. Access, the template, where the list lives. Half an hour hunting for the right file is half an hour gone.
3. The steps, in order, as instructions. Seven or eight at most. If you end up with fifteen, there are two tasks in there.
4. The decisions, with their rules. Every "it depends" becomes an if-then with a real number or a real name in it.
5. What never happens without you. The boundary, spelled out. This is the box that lets you sleep.
6. How you know it went well. Something you can look at, not a feeling.

Written out, a real procedure takes up about this much:

Procedure: chasing overdue invoices
Trigger: every Monday at 9:00.
You need: access to the invoice list and the "notice-1" template.
Steps:
  1. Pull every invoice more than 7 days past due.
  2. Remove clients on agreed 60-day terms (separate list).
  3. Send notice 1 by email, friendly tone, PDF attached.
  4. No reply after 5 days: send notice 2 and flag it to me.
Decisions:
  - Amount over €3,000 -> send nothing, bring it to me first.
  - Client asks to delay payment -> bring it to me, don't negotiate.
Never without me: extending terms, writing off amounts, changing payment terms.
Done right if: no overdue invoice has gone 12 days without contact.

Notice how unliterary it is. A procedure isn't meant to be read. It's meant to be followed.

How to get it out of your head without losing an afternoon

The classic mistake is sitting down to write it from memory. Memory gives you the idealised version of the task rather than the one you actually do, and the first thing it drops is the exceptions — which are the valuable part.

Do it the other way round. Next time the task comes up, record a voice note on your phone while you work through it, narrating as you go, including the "oh wait, this client is different" moments. Then pour the transcript into the six boxes. Twenty minutes, and it comes out in your own words and your own judgement rather than a consultant's.

Two details that make the difference. First: the procedure gets written by whoever does the task, not by whoever's in charge. If you write the quotes, you write the page; if Marta writes them, she writes the page and you review it. Second: keep them all in one place with one naming habit, one document per task. An excellent procedure nobody can find is a procedure that doesn't exist.

What changes when an agent is the one following it

A person fills gaps with common sense and, if they're really unsure, asks. An AI agent does neither of those as well: it fills gaps with something that sounds plausible. That's the whole practical difference.

So three adjustments when the page is for an agent:

• No ambiguous steps. "Let someone know if necessary" is fine for a colleague and terrible for an agent. Write down who, with what wording, in exactly which case.
• Say where the data comes from. A person knows the real prices are in the newest spreadsheet. An agent will use whatever you point it at — give it two sources that disagree and it will pick one, and you won't know which.
• Box 5 becomes the most important one. Anything touching money, deadlines, personal data or telling a client "no" comes past you before it goes out. You approve, they execute.

The upside: the same page serves both. If it's clear enough for an agent to run without inventing anything, the new hire in September will appreciate it even more.

When writing it down isn't worth it

Not every task deserves a procedure, and writing one isn't free. Three cases where I wouldn't bother:

• Things you do twice a year and never the same way. You don't want a procedure there, you want a short list of the four things you always forget. Different format, different purpose.
• Things that are pure judgement. Pricing an unusual project, deciding whether to take on a client who gives you a bad feeling. Writing it down creates a false sense of process and, worse, invites someone to hand off what shouldn't be handed off.
• Things that don't work today. Documenting a broken process just means the mess repeats with more discipline. Fix it first, write it down second.

If writing it down makes you realise the task shouldn't exist, that's a win. That counts as a result too.

The real test, and where to start

The procedure isn't finished when you're happy with it. It's finished when someone else runs it without you. Hand the page to somebody who has never done the task and don't answer any questions for the first hour. Every question they note down is a hole, and the answer goes into the page, not into the chat. Two rounds of that and the document stands on its own. With an agent the loop is identical: for the first week you review what it proposes before it goes out, and each correction goes into the page instead of being repeated message by message.

To get going, don't pick the task that's easiest to write up. Pick the one that would stall if you disappeared for a week, or the one that interrupts you most often in a day. One page, twenty minutes, and that task stops depending on you being available.

Once you have those pages, handing off the repetitive part is almost mechanical — which is exactly what Yaqbot's operations agent does: it follows your rules, does the prep, and brings you whatever needs your judgement. But the order matters. The page first; the agent after.