Part of an ongoing series on ordinary workspace friction, and what happens when you point an agent at it.
The meeting ends on a good note. Someone recaps the next steps out loud. "Sarah will send the revised proposal by Thursday. Marcus will ping legal. We'll pick this back up next week." A few people nod. The notes doc already has those lines in it, sitting under a heading called Action Items, in the same Google Doc or Notion page the team has used for every standup this quarter.
Then the call drops, and everyone goes back to the work that already had a ticket.
Thursday arrives. The proposal doesn't. Sarah isn't dodging it. She never saw it as hers in the system she actually works out of. The line lived in the notes. The notes lived in a tab nobody opened again. Next week's meeting starts with "wait, did that go out?" and ten minutes of reconstruction before the real agenda.
That's how commitments die at most companies. Not from a fight. From a document.
The capture is real work, and it never quite finishes
Writing down the action is the easy part. Getting it into the place where work actually happens is the rest of the job: pick an owner who can do it, not just the person who talked the most; give it a due date that isn't "soon"; put it in Linear or Jira or Asana so it shows up next to everything else that person already owes; and make sure the person who wasn't looking at the notes even knows it exists.
That's a few minutes when the meeting is still warm. It's a scavenger hunt two days later, when the phrasing is fuzzy and half the room has a different memory of who volunteered.
So when the day runs long, the transfer is the first thing cut. Understandable: the next meeting is already starting, and the notes "are in the doc." But that trade has a cost nobody tracks. Nobody logs how many action items get written and never become tickets. Nobody logs how often a "we'll do that by Friday" survives as a sentence and dies as a task. Because nobody measures it, nobody treats it as a problem worth fixing. It just quietly keeps happening.
And a lot of what goes into those tickets already exists. The owner is in the notes. The due date is in the sentence. The context is in the paragraph above it, or in the Slack thread that spawned the meeting, or in the transcript sitting in Granola or Fireflies. The person who was supposed to file the tickets isn't inventing work. They're copying work that already happened, from a doc into a tracker, before the room forgets.
The kind of system we'd build
This is the kind of problem we like: not a new meeting culture, just a specific leak between "we agreed" and "it's in the queue." It also passes the test we use before recommending AI agent development at all: it repeats, it spans more than one tool, and a person can describe exactly what a good result looks like.
Here's what an agent for this looks like. It watches the places meeting commitments actually land, not just the tracker: Notion pages, Google Docs, Confluence, the transcript tools that already record the call. After a meeting, it pulls the action items: who, what, by when, and which meeting they came from. It doesn't wait for someone to remember to file them.
It drafts the tickets in the system the team already uses, in the right person's queue, with the source linked so nobody has to go hunting for the original sentence. It can also post a short Slack recap: here's what got captured, here's what still has no owner, here's what looks like a duplicate of a ticket that already exists.
Nobody starts from a blank Linear issue, or from the vague feeling that last Tuesday's notes had something in them. The assembly is already done.
The agent drafts. A person still decides what it means
The agent doesn't create the tickets on its own. That line matters, and we'd draw it hard in the design.
It can tell you a sentence in the notes looks like an action. It can't tell you whether that sentence was a real commitment, a thought out loud, or a joke that got typed down because someone was moving too fast. It can guess an owner from a name in the doc. It can't tell you whether Sarah is actually the right person, or whether she was just in the room. It can draft a due date. It can't tell you whether Thursday is real or a polite fiction.
So someone still reads the draft, drops the items that weren't really actions, reassigns the ones that landed on the wrong person, and decides what gets filed versus what gets left in the notes on purpose. The difference is where they start. Instead of twenty minutes spent re-reading the doc and reconstructing the room, they spend a few minutes checking the list and applying judgment. The commitments show up in the tracker while they're still warm, because the hard part was already done.
A question worth sitting with
Open last week's meeting notes. Count the lines under Action Items that named a person and a date. Then go look in the tracker. How many of those exist as tickets? How many got done?
If the honest answer is "not most of them," that's not a knock on whoever took the notes. It's a sign the miss wasn't discipline. It was handoff. The work was already written down. It just never left the document.
That's the same shape as the Friday status update, the onboarding checklist nobody owns, and the sales follow-up that goes cold: work that already happened, scattered across tools you already own, with a next action that only becomes obvious when someone asks, a week later, whether it shipped.
If you have a workflow like this one and want to talk through whether it is worth automating, that is what our AI agent development work is for.