← Index / All work

11·AI Agent · Operations·2026

Live · try it below

Action Items Agent

Paste a meeting transcript, it pulls out every action item with an owner, due date and priority, plus a two-line summary.

Next.jsTypeScriptAgent designAI-assisted

01 · The opportunity

What it set out to solve

The decisions in a meeting are worthless if the follow-ups get lost. Someone has to listen back, work out who committed to what and by when, and turn it into a task list, and usually nobody does, so things slip.

02 · The approach

How we thought about it

We built it as an agent that reads a transcript the way a good chief-of-staff would: find the commitments, attribute each to an owner, pull out the due date, and rank by urgency. The demo uses verb/name/date heuristics; production runs the same shape on your real call transcripts with an LLM.

03 · What we built

  • 01Paste-in transcript with a realistic prefilled sample
  • 02A visible extraction loop: read → spot commitments → assign → prioritise
  • 03Action items with task, owner, due date and P1-P3 priority
  • 04A two-line summary of who owns what and what's urgent
  • 05One-click copy of the whole list

04 · The result

What changed

A working agent that turns a wall of transcript into an owned, prioritised task list in seconds, the kind of ops automation we wire into a client's meeting and tracker stack.

Live · interactive

Pull the action items.

Paste a meeting transcript (or use the sample) and watch it pull out every action item with an owner, due date and priority, plus a two-line summary. All in your browser.

Meeting transcript

Paste a transcript in Name: … form, or use the sample.

Every commitment, task, owner, due date and priority, plus a two-line summary will appear here.

05 · Decisions

The questions we get asked, and our answers

How does it decide the owner and priority?

Owner comes from who was addressed ("Raj, can you…") or who committed ("I'll…"); priority comes from signal words (urgent, blocker, EOD → P1; 'can wait', 'after launch' → P3). In production the same rules back an LLM so it also catches what the keywords miss.

How would this run for real?

Connected to your meeting tool (Fireflies, Otter, Meet, Zoom): every call is transcribed, the agent extracts the action items, and pushes them straight into your tracker, assigned, dated and prioritised, before anyone's left the room.