Most action items die in the gap between when they're written down and when someone actually does them. The meeting ends, the bullet list gets pasted into a doc, the doc gets emailed around, and a week later three of the five items have quietly disappeared. Nobody's at fault — the capture itself was the weak point.

This guide is about closing that gap. Not a productivity sermon — a small set of habits and one tooling choice that turns "we should follow up on X" into a thing that gets followed up on. How to write action items that survive a real week, where to put them so you find them again, and how to get the boring transcription work off your plate.

## Why most action items quietly disappear

Take a normal meeting. Six people, an hour, a dozen things mentioned that "someone should look into." Somebody types up notes containing phrases like "look into pricing model" or "follow up with the design team." A week later, none of it has happened. Why?

Because none of those are actually action items. They're topics. An action item has three pieces — a specific task, a specific owner, a specific date — and most meeting notes have none of them. "Look into pricing model" doesn't tell anyone what to do, who's doing it, or when to be done. Nobody does it, and nobody chases it, because there's nothing concrete to chase.

The other failure is mechanical. The action items get captured well, but in a place no one checks again. A Google Doc that closes after the meeting. A Slack message scrolled past by Tuesday. A notebook that lives in your bag. The capture happened; the staging — putting items where they'd re-enter your life — didn't.

Both failures are fixable. You need a clear definition of an action item, a single place those items live, and a way to surface them again at the moments you can act.

## What a real action item looks like

Three pieces. A clear verb-object task, an owner, a date. That's it.

- "Sarah sends the revised pricing deck to Marcus by Friday."
- "Marcus drafts the onboarding email sequence by next Wednesday."
- "I review the Q2 forecast before Monday's leadership sync."

Compare those to what often actually gets written down: "discuss pricing," "onboarding email," "Q2 forecast." Those are agenda topics, not commitments. They evaporate the moment the meeting ends because there's nothing to evaporate from — no commitment was ever made.

The verb matters. "Send," "draft," "review," "decide," "schedule" — these tell the owner what to do. Soft verbs like "look at," "explore," or "think about" mean the team didn't actually agree to anything; they kicked the can. Worth knowing — call it a topic, not an action item, and put it on a separate list.

The owner matters. Not "the team" or "we" — one person. Two-person ownership is one-person ownership in disguise; it's how things slip. If two people need to coordinate, name one as owner and the other as contributor.

The date matters most. An undated action item is a wish. It doesn't have to be tight — "by end of month" works — but it has to exist, and the owner has to know it. Dates make follow-up possible because they create a moment to ask "where are we on this?" without it feeling like nagging.

## The capture problem during the meeting

Even when you know what a real action item is, capturing them in real time is hard. You're trying to listen, contribute, and somehow also write. Most people do one of two things, both with costs.

Option one: take meticulous notes the whole time. Action items get captured cleanly, but you're a worse participant. You miss the look on someone's face when they say "yes" while clearly meaning "I don't think so." You're a transcription machine, not a colleague.

Option two: participate fully and reconstruct action items afterward from memory. Better collaborator in the room, fuzzier output. The exact wording of what someone agreed to is gone. A week later, when there's a disagreement about who said what, the notes can't settle it.

The honest fix is to stop forcing yourself to choose. Let something else handle the verbatim capture so you can be in the room as a person. Then do the synthesis afterward, with the recording in front of you. That's what we built Capy around for the meeting case — it records the audio in your vault and transcribes it with speaker labels, so when you sit down to pull out action items, you have "Marcus said X, Sarah said Y" in plain text instead of a wall of dialogue you have to mentally untangle. The deeper-dive on that mechanic lives at [AI meeting note taker with speaker labels](/blog/ai-meeting-note-taker/), and the broader category read is in [AI meeting notes: what's actually worth using](/blog/ai-meeting-notes/).

## A simple workflow for turning meetings into commitments

Here's the rhythm we recommend. It takes about ten minutes per meeting and reliably produces action items that actually move.

1. **Record the meeting**, with the participants' knowledge. Drop the audio onto a page in your vault as soon as the call ends. Transcription with speaker labels happens automatically — you don't have to babysit it.
2. **Skim the transcript while it's fresh**, ideally within an hour. You're not reading every word; you're scanning for the moments where someone agreed to do something. The speaker labels make this fast — you can jump between contributors without re-listening.
3. **Pull out the commitments** in the action-item format: task, owner, date. If a commitment was vague in the meeting, this is the moment to make it specific, while the context is still in your head. "Sarah will look into pricing" becomes "Sarah sends the three-tier proposal by Wednesday."
4. **Write them somewhere you'll see them again** — and that means a single, durable place, not the meeting notes themselves. We do this with an inline database on a "follow-ups" page in the same vault, but the principle works in any tool: one persistent list, sorted by date, with a status column.
5. **Close the loop on each item.** When something's done, mark it done. When something slips, update the date and note why. The list is only as useful as it is current.

This routine works because it splits the cognitive load. The recording handles capture. The transcript handles recall. You handle the only piece that needs you — turning fuzzy talk into specific commitments.

## Where action items should live (and not live)

Most action-item systems fail at the storage layer. The items get captured cleanly and then put somewhere their author will never look again. A few patterns we've seen, and what tends to happen with each.

**Inside the meeting notes themselves.** The default, and the worst. Meeting notes are written for a single moment in time; nobody opens last Tuesday's notes on Friday. Action items get buried in the doc that produced them.

**In a chat thread.** The items get shared, everyone says "thanks" with a thumbs-up, the message scrolls off by Wednesday. Chat is great for "did you see this?" and terrible for "what was I supposed to do this week?"

**In a project tracker.** Better — persistent and dated. The cost is friction. If transferring an action item from notes to tracker takes more than fifteen seconds, you'll skip it under deadline pressure, and the system breaks.

**In a single living "follow-ups" list inside the same workspace as your notes.** This is what works best, and it's the shape Capy supports natively. The items live in an inline database embedded in a markdown page — same vault as your meeting notes, queryable by status and date. No "open another tool" tax. The action items sit alongside the work that produced them. (For the boardroom variant of this — fiduciary discussions, formal minutes, follow-through across quarterly cycles — see [AI notes for board meetings](/guides/meetings-people/ai-notes-board-meetings/) and [AI notes for advisory board meetings](/guides/meetings-people/advisory-board-meetings/).)

Underneath all four: action items should live where you do your work, not in a separate accountability system you have to remember to check.

## Letting an agent do the boring extraction

Once your meetings are transcripts in your vault, the next time-saver is letting an agent do a first pass on action items. Not a magic "AI summary" button that condenses the meeting into a paragraph — that's a low-trust artifact and you've probably already learned not to rely on it. Something more specific.

Ask Capy: "Read yesterday's product sync transcript. Find every commitment someone made — task, owner, and date if mentioned. Write them as new rows in my follow-ups database on the meetings page." The agent reads the transcript, identifies the moments people agreed to do things, formats them in the three-piece structure, and drops them into the database.

The output is a draft. You skim it, fix anything fuzzy, add the dates the agent didn't have, done. The agent did the boring scan of the transcript; you did the human work of confirming what people actually committed to.

Because the transcript and the database both live in your vault as plain markdown, the agent isn't bridging five tools. It reads one markdown page, writes to a database directive on another, and you can see the chain in plain text afterward.

## The follow-up habit that closes the loop

A list of action items only matters if it gets revisited. Most people who try to build a follow-up habit aim too high — daily reviews, color-coded statuses, a whole personal kanban. That breaks within a week.

A more honest version is two short windows. Friday afternoon, scan the list, mark anything finished, push anything that slipped to a new date. Monday morning, scan again to set the week's intent — what's due, what can be batched, what needs an email today. Ten minutes total.

The list does the heavy lifting. You're not relying on memory to surface what you owe people; the list surfaces it. The discipline is opening the page twice a week.

One step further: ask the agent to do the surfacing for you. "What's overdue on my follow-ups list, and what's due this week?" gets you a focused short list instead of a full database scan. Same data, less friction.

## Try it on your next meeting

If any of this lands, the cheapest place to test it is your next call. Record the meeting, drop the audio in a vault page, let the transcript land, spend ten minutes extracting commitments in the task-owner-date format. Drop them in one persistent list. See whether more of them get done over the next two weeks. If most of your meetings are client-facing, the same capture-and-quote pattern is also useful for [AI notes for client work](/guides/creatives-content/ai-notes-for-client-work/).

Try Docapybara free — [sign up](/accounts/signup/), and your meetings show up in your vault, not in a separate notes app. The agent that handles the boring transcription work is the same one that helps you turn the transcript into an action-item list you'll actually revisit. One person, one vault, one place where the commitments live.