The problem this solves
Slack is where decisions actually get made, and Slack is where they go to die. A thread reaches a real conclusion - we are cutting the webinar, Dana owns the pricing page rewrite, ship by the fifteenth - and then the channel scrolls. Nobody was assigned to transcribe the outcome into the project tool, because transcription is nobody's job. Two weeks later the same decision gets made again, slightly differently, by people who each remember a different version of the thread.
The transcription gap exists because the handoff is genuinely tedious. Turning a forty-message thread into clean tasks means rereading it, separating decisions from discussion, inferring who owns what from conversational shorthand, and typing it all into another tool. Done properly it takes twenty minutes of a senior person's attention, immediately after a meeting-like conversation, which is precisely when nobody has twenty minutes.
The work is mechanical once you can read the thread with judgment: identify what was actually decided, capture the owner and date where one was stated, and flag ambiguity honestly instead of guessing. That is a reading and structuring job, and an agent sitting in the thread can do it the moment the conversation ends, while the participants are still there to correct it.
How the mission runs
- Invoke it at the end of the conversation. When the thread winds down, one participant asks the agent to capture it. The agent reads the full thread from the top through Slack, using thread memory rather than a pasted transcript, so nothing depends on someone summarizing accurately before the summary is made.
- Decisions are separated from discussion. The agent distinguishes conclusions from the debate that produced them: proposals that were superseded stay out, the option that won goes in. Where the thread stated an owner or a date, those attach to the item. Where it did not, the item is captured with the gap marked instead of a guessed name.
- Tasks are created in Asana, structured. Each decision becomes an Asana task: an action-verb title, a description quoting the deciding message for provenance, the owner and due date where stated, in the project you designated. The tasks carry enough context that someone opening one cold in three weeks understands why it exists.
- Links post back for the correction window. The agent replies in the thread with the task list and links while everyone involved is still present. This is the review moment: a reply like "task two is actually Priya's" or "add a task for the legal check" gets applied immediately, and the correction costs one sentence instead of a project-tool session.
- Ambiguities surface as questions. Anything the agent could not resolve - an owner never named, a date that conflicts with an earlier message - comes back as a short question in the thread rather than a silently wrong task. The mission finishes when the board matches what the room actually agreed, and the thread records that it does.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A set of structured Asana tasks matching what your thread actually decided: clear titles, owners and due dates where stated, source context quoted in each description, and open questions flagged instead of guessed. Posted back into the thread as links within minutes of the conversation ending, while corrections are still cheap. The thread and the board finally say the same thing.
Make it yours
- Make it standing practice for one channel: any thread where someone reacts with a chosen emoji gets captured to Asana automatically, and the agent replies with the links.
- Run it retroactively on the last month of a decision-heavy channel to recover commitments that were made and never tracked, reviewing the recovered list before any tasks are created.
- Point different threads at different Asana projects by naming the project when you invoke it, so campaign threads and ops threads land on their own boards.
Frequently asked questions
What if the thread is messy, with decisions reversed midway?
Reversals are exactly what the reading pass is for. The agent tracks the thread chronologically, so a decision that was overturned forty messages later is captured in its final form. Where the outcome is genuinely unclear, it asks in the thread rather than committing a guess to the board.
Can it assign tasks to the wrong person?
It only assigns owners the thread actually stated, matched against your Asana members, and the link post-back is the safety net: assignees are visible in the thread immediately, while everyone still has context. Unstated owners are left unassigned and flagged, because a wrong owner is worse than an empty field.
What does it write in Asana, and what does it never touch?
It creates new tasks in the project you designate, and updates only tasks it created when you correct them in the thread. It does not edit existing tasks, restructure sections, or delete anything. The write surface is deliberately narrow so a capture mistake is always additive and reversible.