Twenty minutes, one page, four columns. Go through your automations tool, your pinned chats, your "AI" folder and your refuse-to-close browser tabs, and write down every AI thing you started. For each one record what it was meant to do, how done it honestly is, whether it is running right now, and what it can reach. That last column is the one people skip and the one that turns a tidy-up into something worth doing. You cannot decide what to finish until the whole pile is visible in one place, and almost nobody can list it from memory.

Where is the pile actually hiding?

In more places than you'd guess, which is why memory doesn't work for this. Check all six:

Your automations tool. Every draft, not just the live ones. Most platforms hide unpublished work behind a filter, and unpublished is exactly where abandoned things go to sit.

Pinned and starred chats. The conversation where you and a model designed a workflow together and you meant to come back to it. There are usually two or three.

The folder called "AI". Or "Automation", or "Prompts", or whatever optimistic name it got in month one. Open it. Read the file names.

Browser tabs you won't close. A tab kept open for three weeks is an unfinished project wearing a disguise.

Your subscriptions. Anything you're paying for monthly that you built something with, then stopped. The billing is a receipt for a decision you never actually made.

Connected apps and app passwords. In your email, storage and calendar settings there's a list of things you granted access to. Some of them are experiments you've forgotten. This is the least fun list and the most important one.

What exactly do I write down for each item?

Four columns, and they take about ninety seconds each.

1. What it was meant to do. One line, in plain words. If you can't state it, that's your first finding — you built something without a job description, which is usually why it stalled.

2. How done is it, honestly. Resist round numbers. "Works when the input is tidy" is more useful than "80%". You are describing the gap you'd have to close, so describe the gap.

3. Is it running right now? Yes, no, or don't know. "Don't know" is a legitimate answer and a flag to check today rather than a reason to write nothing.

4. What can it reach? Files, inbox, calendar, contacts, a bank feed, a public channel where it can post. Be specific.

Columns three and four are where the inventory stops being about productivity. Unfinished does not mean switched off. An abandoned workflow can hold live access and still be producing output nobody reads, which is a process without an owner — and that is a governance problem wearing the costume of an untidy desk.

What do people usually find?

Three things, fairly reliably.

More items than expected — the count is almost always higher than the guess, because abandoned projects don't announce themselves. At least one or two entirely forgotten — things that were genuinely absent from memory until the folder was opened. And typically at least one still running: sending, syncing, filing or billing, with nobody reading the results.

That last discovery is the payoff. It's the item you'd never have chosen to look for, and it's the one worth handling first.

The honest limit of this exercise: an inventory doesn't fix anything. It's twenty minutes of seeing, and seeing on its own has never finished a project. It just makes the next decision — finish, park or kill — possible rather than theoretical, and that decision is tomorrow's job.

Do it before September. The list is more uncomfortable in October.

Related Decodes

• Finish It: What to Do With the Half-Built AI Projects You Started This Summer — this week's hub and the full two-hour clear-out.
• Finish, Park or Kill: The Three-Column Decision That Clears the Pile — tomorrow: what to do with the list you just made.
• A Half-Built Automation Is a Liability, Not a Head Start — why columns three and four matter.
• The Mid-Year AI Stack Review: Cancel, Keep, Compound — the same instinct applied to what you pay for.

For the ten tools worth finishing with — free, vetted, 5-min read: 10 AI Tools That Save 10 Hours a Week.

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

TL;DR: You cannot decide what to finish until the whole pile is visible, and almost nobody can list their half-built AI projects from memory — so spend twenty minutes writing them onto one page. Look in six places: every draft in your automations tool, pinned and starred chats, the folder named "AI", long-lived browser tabs, monthly subscriptions tied to abandoned builds, and the connected-apps list in your email, storage and calendar settings. For each item record four things: what it was meant to do, how done it honestly is (avoid round numbers — describe the gap), whether it is running right now, and what it can reach. The last two columns matter most, because unfinished does not mean switched off: abandoned automations can still hold live access and still produce output nobody reads. Most people find more items than they expected, one or two entirely forgotten, and at least one still quietly running.