Architecting an AI-native workflow — a 5-step playbook in code — step 6 of 9

A canonical implicit-policy failure: "the sales process is 'talk to Sarah, she knows how we do enterprise.'" That's tribal knowledge. An agent can't ask Sarah. The fix is to make Sarah's knowledge explicit as a data rule.

The spec below has refund_policy = "ask the support lead". An agent reading this has nothing to ground in. Replace it with an explicit rule dict containing auto_approve_under (dollar threshold below which the agent can refund without human review) and requires_human_above (the same threshold). Use $100.

Expected output:

{'auto_approve_under': 100, 'requires_human_above': 100}