Every item on your list goes in exactly one of three columns, and only one item is allowed in Finish. Finish means it would save you time this month and you know the remaining steps. Park means good idea, wrong quarter — with a written reason and a revisit date. Kill means it solves a problem you no longer have, so switch it off properly and delete it. The rule that makes the whole thing work is the cap: a second item in Finish is exactly how the pile started in the first place.

The relief, when it comes, isn't from finishing. It's from having decided.

How do I tell Finish from Park?

Two tests, and both have to pass.

Would it save time this month? Not eventually, not at scale, not once you're bigger. This month. Automations justified by a future volume you don't have yet are the most common thing in the Park column, and they belong there — they're not wrong, they're early.

Do you know the remaining steps? If the honest answer is "I'd have to work out what's left", it isn't ready to finish; it's ready to be scoped, which is a different task and usually a smaller one. Scope it, then decide.

An item that fails either test is a Park, not a Finish. Being generous with the Finish column is how people end up back where they started with a nicer-looking list.

What does Park actually require?

More than closing the tab, which is the whole point of the distinction.

Write down why you stopped. One line. "Needs a client volume we don't have", "waiting on the API to stabilise", "got boring, still probably worth it". This is the single highest-value sentence in the exercise, because in four months you will not remember, and without it you'll either rebuild from scratch or leave it alone out of vague unease.

Give it a date. A specific one — "review 1 Dec" — not "later". A parked item with a date is retrievable. A parked item without one is just forgotten with extra steps.

Then genuinely leave it. Parking is not a soft version of working on it. If it's parked and you keep poking at it, you've actually put two things in Finish and told yourself otherwise.

When is it right to kill something?

When it solves a problem you no longer have — which happens more than people admit, because businesses change faster than their automations do. The invoice chaser you built for a client base you no longer serve. The research workflow for a service you stopped offering. The summariser for a newsletter you unsubscribed from.

Killing is the column people need permission for, so here it is plainly: abandoning something you built is not waste. The waste already happened, and it happened during the build. The only live question is whether you keep paying attention rent on it — and you are paying, in the small tax of every time you notice it and think "I should deal with that".

Running eight unfinished things costs more than deleting eight unfinished things. That's not a productivity aphorism; it's a description of where your attention actually goes.

Kill properly, though. Deleting an automation is not the same as ending one — pause it first, wait a week, see what breaks, revoke its access, export anything you'd miss, cancel what it was paying for and write one line about why it stopped. That's Friday's decode, and skipping it is how a killed project turns into a live one nobody owns.

What this doesn't fix

It doesn't stop you starting things. Nothing does, and honestly nothing should — curiosity is most of why any of this works. What the three columns do is make starting cost something visible: the next new idea has to displace whatever is currently in Finish, or wait.

Revisit the list monthly. Ten minutes. If Finish is empty, put one thing in it. If Finish has two, you already know which one is real.

Related Decodes

• Finish It: What to Do With the Half-Built AI Projects You Started This Summer — this week's hub.
• The Half-Built Pile: A 20-Minute Inventory of Every AI Thing You Started — yesterday: the list this decode sorts.
• The Decision Log: The Five-Minute Habit That Makes Next Quarter Easier — why the written reason beats the remembered one.
• Safe to Automate: The Two-Column List Every Solo Business Should Write Once — the sorting done before the build instead of after.

For the weekend plan that finishes one of them — free, vetted, 5-min read: The Weekend Automation Plan.

AI the Brief (@aithebrief) — AI advice you can act on and trust, for time-starved founders.

TL;DR: Sort every half-built AI project into exactly one of three columns — Finish, Park or Kill — with a hard cap of one item in Finish at any time, because a second Finish is how the pile started. Finish requires both that it would save time this month and that you already know the remaining steps; failing either test makes it a Park, not a Finish. Parking properly means writing one line explaining why you stopped and setting a specific revisit date, since a parked item with a date is retrievable and one without is just forgotten. Kill anything that solves a problem you no longer have: abandoning a build isn't waste, because the waste already happened during the build, and running eight unfinished things costs more attention than deleting eight. Kill properly though — pause, wait a week, revoke access, export, cancel and record why. Revisit the list monthly in ten minutes.