Module 2 — Conversation to Commitment · Lesson 2.4
Triage — Work vs. Noise
The thirty-second decision that protects the rest of the team
~10 min
What you'll learn
- Apply four triage questions to any candidate task
- Use Negation explicitly to dismiss out-of-scope candidates without guilt
- Recognize the three most common noise patterns
- Build a daily triage cadence that takes ten minutes or less
Auto-capture replaces an old problem (work falls through the cracks) with a new one (too many candidates). The new problem is actually easier — you can solve it with a daily ten-minute pass — but only if the team has a triage discipline. Without one, the candidate queue grows, and within two weeks people stop opening it. This lesson is the triage discipline.
Four questions, in this order
Open a candidate. Ask, in this order:
1. Does this advance a stated Vision? If yes, continue. If no, dismiss it. Not 'maybe someday.' Dismiss. The conversation log will preserve it; you do not need to keep it on the board to retrieve it later.
2. Does this cross a Negation? If yes, dismiss with a one-line explanation in the dismiss reason. The Negation that caught it just earned its keep — those reuse events show up in the negation-hits metric. If you find yourself dismissing many candidates against the same Negation, the workspace or project charter is doing its job.
3. Is this the smallest unit of work that produces a checkable outcome? If yes, accept. If no — if it is so big that you cannot tell whether it is done — break it down. The agent's create_task tool accepts sub-task lists; ask it to break the candidate into checkable pieces. If you cannot articulate the checkable pieces, the candidate is not yet ready to be work.
4. Does it have a clear owner who is the right owner? The agent's recommendation is usually right; check it. The most common override is when the right owner is on leave or in a focus block — Module 3 covers this with the availability system. If there is no right owner, the candidate is interesting but premature; park it as a 'needs an owner' entry, not a task.
Four questions, thirty seconds each. The throughput on triage should be roughly two candidates per minute once you have done a few hundred.
The three most common noise patterns
Recurring operations as candidate tasks. The agent will sometimes propose recurring operations work as one-off tasks ('reply to the Q3 customer email' becomes a candidate every week). Dismiss these and either set up an automation in /automations or accept the work as an ongoing responsibility of a role, not a task. The dismissal teaches the agent.
Genericized commitments. 'We should improve performance' is a topic, not a task. The agent will sometimes float topics from conversations where multiple participants noodled on a problem. Dismiss with the reason 'no checkable outcome'; revisit when someone makes a specific commitment.
Double-counted commitments. If the same intent showed up in chat, then in email, then in a meeting, the agent's dedupe usually catches it — but not always, especially when phrasing varies. Triage merges the duplicates into the canonical candidate before accepting.
These three patterns account for the majority of dismissals in well-instrumented workspaces. The dismissal reasons themselves become a useful signal: a sudden spike in 'no checkable outcome' dismissals usually correlates with a phase of the project where the team is thinking out loud, which is a signal the project Vision needs sharpening.
The Negation axis is a noise filter, not a wall
The temptation is to treat Negation as restrictive — the wall the team cannot cross. The more useful framing is: Negation is the noise filter that prevents the team from spending time on candidate work that the leadership has already decided is not for them.
When a candidate gets dismissed because it crosses a Negation, three good things happen at once. The candidate goes away. The Negation gets a hit, which surfaces in the negation-hits-per-week metric as evidence the boundary is doing its job. And the conversation participants get a soft signal — through the agent's reply or the dismiss reason — that this is not the kind of work the team is taking on right now.
A Negation that never produces dismissals is either not relevant or not being read by the system. Either way, the metrics surface it. Module 5.4 walks the negation-enforcement loop in detail.
Cadence: ten minutes a day, batched
Triage works as a batched daily ritual, not a real-time response. Set a recurring slot — twenty minutes in the morning is plenty for most workspaces — and process the candidate queue once. The agent batches candidates while you sleep; you accept, edit, or dismiss in a single pass.
More than once a day is unnecessary, because the cost of capture is no longer urgency. The agent does not forget. The conversation log is retained. You will not lose a candidate by waiting twelve hours to look at it.
Less than once a day is dangerous, because the queue grows past the point where you trust your own throughput. The triage budget is best invested at a small, sustainable daily rate rather than as a heroic weekly catch-up.
Make triage a daily ritual
- 1
Add a recurring Triage slot to your calendar
Twenty minutes, every morning. The agent will have queued candidates while you slept.
- 2
Open the AI agent's proposal queue
Process the candidates four-questions-each. Accept, edit, dismiss with reason.
- 3
Note the dismissal reasons that recur
If the same dismissal reason recurs, the workspace or project charter is missing a Negation. Add it.
- 4
Confirm the queue is empty before ending the slot
Triage is done when the queue is empty, not when the timer runs out.
Triage discipline
- Triage queue size
- Number of unreviewed candidate tasks in the queue at any moment.
- Healthy signal: Under 25 outside of triage slot. Trending up means cadence is too slow.
- Triage time per candidate
- Median seconds spent on each candidate during the triage pass.
- Healthy signal: 30–60 seconds. Above 120 means candidates are arriving under-enriched.
- Dismissal-reason distribution
- Breakdown of why candidates get dismissed: out of scope (Negation), no checkable outcome, duplicate, premature.
- Healthy signal: Useful as a diagnostic — large 'no checkable outcome' share signals an underspecified project Vision.
- Time in queue
- Median wait time between a candidate being created and being triaged.
- Healthy signal: Under 24 hours.
Key takeaways
- ·Auto-capture turns a capture problem into a triage problem. Triage is the smaller problem.
- ·Four questions: advances a Vision? crosses a Negation? smallest checkable unit? right owner?
- ·Three common noise patterns: recurring operations, generic topics, double-counted commitments.
- ·Negation is a noise filter, not a wall.
- ·Cadence is once a day, twenty minutes, batched.
With capture and triage solved, every task on the board is real, has a checkable outcome, and has an owner. The next question is whether the owner is the right one — and that depends on capability, availability, and the choice between human and AI. Module 3.