Choose among three start patterns
Most recurring knowledge work begins through one of these patterns:
- An event arrives, such as approved source notes entering a defined location.
- A schedule arrives, such as a weekly review time.
- A person starts the job after confirming that the material is ready.
The pattern should match the source. A calendar is weak when the required notes arrive irregularly. A source event is weak when a person must first decide whether the work should happen.
Add readiness and duplicate checks
A start condition should answer:
- Is the required input present?
- Does the input belong to this recurring job?
- Has this same item already started or finished?
- Is a person required to authorize the start?
W3C PROV-O offers concepts for relating an activity to the entities and agents involved. The NIST AI Risk Management Framework supports explicit governance around risk-bearing AI work. CommonMark can preserve the start rule in a readable operating note.
Avoid triggers that encode guesses
This method fits recurring work with observable arrivals, stable schedules, or accountable starts. It fits poorly when the initiating event depends on an unrecorded feeling such as “when things look quiet.”
Do not let a trigger manufacture a missing fact. A schedule can start a research check, but it cannot turn an unknown session schedule into a confirmed Saturday offer.
Link the start to nearby choices
See Map an AI Workflow for One Recurring Job for the complete sequence. Use Human-in-the-Loop AI Workflow when a person must decide whether the event is sufficient. Continue with AI Workflow Approvals when later transitions also need explicit decisions.
Tie the start to approved notes
The synthetic job starts when the approved planning notes are selected, not when someone guesses that new copy might be useful.