OPEN · Q3·26
Projects / Taskie — Meeting Task Logger AI SYSTEMS · 2026
Case · AI Systems

Taskie.

A Notion agent that reads meeting transcripts, pulls out the action items, assigns owners, and pushes them straight into my project management tool.

Year
2026
Status
Shipped, running
Role
Designer · Builder
FIG · 01 [ Hero visual
drop in screenshot or diagram ] PLACEHOLDER
Meetings logged
50+
since launch
Tasks captured
100+
owner-tagged, prioritised
Trigger
AUTO
on new transcript
Build mode
SOLO
designer, builder, sole user

§ 01 — The Context

I sit in a lot of meetings each week with different people and stakeholders. Keeping track of what I'm actually meant to action out of each one is the part that quietly eats time. The notes are there in Notion, but the action items live inside them, buried in the back-and-forth.

Before Taskie, tasks would slip. Not all of them, but enough to be a problem. The activation cost of opening every meeting note, pulling out my action items by hand, and dropping them into my project management tool was just high enough that I'd defer it. Then forget it.

CONTEXT

§ 02 — Role & Approach

Sole designer and builder. Built it to test the new Notion agent functionality and figure out where it actually fits into my work.
  1. P/01Trigger off where the data already lives. Meeting notes are already in Notion. No new capture surface, no new habit to build.
  2. P/02Push results to where the work happens. Tasks land in the project management tool, not back in Notion. Action items don't get done in note form.
  3. P/03Run on autopilot. No manual review step. If it's not automatic, the activation cost doesn't drop.
ROLE

§ 03 — The Process · how the work unfolded

Phase 01
· explore
Test the Notion agent on a real workload.The agent functionality was new. Wanted to use it on something I actually needed, not a toy task. Meeting notes were the obvious candidate: same workspace, same database, no glue code.
Phase 02
· prototype
Extraction first, automation later.Ran the agent against a handful of historical transcripts. Eyeballed extraction quality before wiring any downstream automation. Adjusted the prompt until owner assignment lined up with the meeting attendee list.
Phase 03
· wire
Push tasks to the project management tool.Once extraction was reliable, wired the output to create tasks downstream, owner-tagged. Closed the loop by auto-archiving the source meeting note once tasks were captured, so the inbox stays clean.
Phase 04
· live
Point it at new meetings.Let it run on transcripts as they came in. Watched task quality over the first weeks of live use to catch anything the prompt was missing.
+ ongoing
Priority and urgency matrix added.After ~100 tasks, the list started to feel busy. Updated the flow so it prioritises tasks I can action myself, with a separate bucket for nice-to-haves. Front-of-queue stays workable; the rest doesn't disappear.

§ 04 — Tools & Decisions

The interesting decision was leaning fully on the Notion agent rather than building a separate skill. Notion already holds the transcripts, the attendee lists, and the meeting database in one workspace, so an agent inside Notion has the full context for free. The only external write is the push to the project management tool. Owner assignment runs off the attendee list inside each meeting page. Priority sorting runs on a second pass after tasks are created, so the extraction prompt stays small and focused.
NotionNotion AgentClickUpAuto-archive
TOOLS
§ 05 — The Outcome
100+ tasks
Captured across 50+ meetings since launch. No manual entry of action items, no dropped tasks from the meetings that go through it.

§ 06 — Reflections

"The first version treated every task as equal weight. Once a hundred had landed, the list felt noisy. The fix wasn't smarter extraction, it was a priority bucket that puts work I can actually action at the front."

— Working note · 2026 · what I'd tell next-me
Next Back to projects