Layer 2 / Decision Architecture
Define what AI can influence, recommend, or decide.
Use this checklist to bound the decisions AI may inform, recommend, or execute.
Checklist scope
One layer. One workflow. One visible gap.
This page does not replace the full Decision Architecture framework page or certify that an AI system is safe. It helps a team inspect one operating condition before scaling.
Concise Readiness Checklist
Assess the current condition, then make one improvement.
Answer from the live workflow and its evidence. If the team cannot point to an owner, record, source, control, or review, treat that as the gap.
Ask now
- What decisions will AI support?
- What decisions can AI recommend?
- What decisions require human approval?
- What decisions require escalation?
Strengthen next
- Create decision rights for each AI use case.
- Place each AI decision on the five-point autonomy spectrum: Human Only, Human Approved, Agent Assisted (the default entry point for every new agent or decision type), Agent Supervised, Agent Autonomous.
- Earn more autonomy only by clearing a measurable confidence threshold set by the business owner, not engineering.
- Define approval paths and escalation thresholds.
Run the Checklist
Apply it to a live workflow, not an abstract AI strategy.
Use one bounded workflow with a named outcome and a decision AI may influence.
- 1
Name the workflow
Describe the trigger, outcome, participants, and the decision AI may influence.
- 2
Inspect Decision Architecture
Write the decision, decider, required evidence, autonomy level, and escalation rule.
- 3
Record the gap
Name one owner, one corrective action, and the evidence that must exist before scale.
Create the Operating Proof
Use the Decision Rights Matrix.
A completed working artifact is stronger evidence than a slide describing future intent.
Start with the checklist gap, complete the tool for the selected workflow, and review whether the artifact is actually used when the next live case arrives.