A boundary prevents accidental scope growth

Without a firm start, old notes and unrelated instructions can enter the job. Without a firm end, polishing continues after the decision has been made. Both problems add work without improving the artifact.

The boundary needs three parts:

  • an entry condition that identifies acceptable source material;
  • an exit condition that identifies the current artifact;
  • exclusions that keep adjacent jobs outside this process.

Stages belong between those lines. They should not redefine them during execution.

Test both edges on one artifact

The synthetic job begins when the selected notes enter the task and ends when the corrected brief becomes the current artifact.

Choose stages that change the work state

A stage deserves a name when it transforms the artifact or changes who can decide what happens next. For this example, source selection, first drafting, evidence comparison, correction, and final recording each create a distinguishable state.

Avoid stages named after software screens or temporary tools. “Open editor” explains an interface action. “Correct unsupported claims” explains a durable job.

W3C PROV-O distinguishes entities, activities, and agents, which helps describe what changed and who was responsible. The NIST AI Risk Management Framework frames risk work as an ongoing practice rather than a one-time label. CommonMark supplies a plain-text syntax for keeping stage definitions readable.

Fit the boundary to the decision

This method fits recurring knowledge work with a stable source type and a recognizable output. It fits poorly when exploration itself is the goal or when every engagement ends in a different kind of artifact.

A narrow workflow can still contain human judgment. The boundary says which job is under examination; it does not dictate how every decision inside that job must be made.

Continue the workflow map

Return to Map an AI Workflow for One Recurring Job for the full cluster. Next, define the material that moves between steps in AI Workflow Handoffs, or decide which stages need assistance in AI Workflow or Automation.