Skip to content
Start an assessment
Return to the framework
Layer 03 of 06

Workflows

How should the AI do recurring work?

Structured, repeatable procedures for the AI's day-to-day tasks, with clear inputs, outputs, and checkpoints.

Why this layer matters

Ad-hoc prompts produce ad-hoc results. A business that runs on AI needs the AI to run the same process the same way every time, the way a trained employee would.

Workflows turn "ask the AI to do it" into "the AI runs the intake-to-delivery procedure with three human checkpoints." That is what a process owner can actually manage.

What good looks like

  • Named workflows for high-value recurring tasks
  • Explicit inputs, outputs, and human checkpoints
  • Defined handoffs between the AI and the people it works with
  • Versioning, so a workflow change is reviewable and reversible
  • Performance data so the workflow improves over time

What a gap looks like

How it connects

Workflows are the connective tissue between domain knowledge (what the AI needs to know) and identity (the role the AI plays at each step). Guardrails set the limits of what a workflow step is allowed to do.