A program manager at a mid-size logistics firm designs an agent that summarizes vendor proposals, scores them against budget rules, sends acceptance emails to winners, and archives contracts in GitHub. Finance requires accountability for spend and legal risk. Which critical decision point requires human judgment and validation?

The correct decision requires prior human approval because sending vendor acceptance emails creates a binding financial commitment and legal obligation, directly triggering the stated finance need for accountability for spend and legal risk, so autonomy must stop before irreversible external action with budget impact. Validating proposal summaries before scoring is a sound quality control for retrieval and summarization accuracy, but it fails by concept because it governs internal information preparation rather than authorization of spend, leaving the high-risk commitment uncontrolled. Reviewing version-controlled contracts after archiving in GitHub is a commonly used auditability and traceability control, but it fails by concept because detective review after archiving cannot prevent or reverse an unauthorized budget commitment already communicated to a vendor. Approving status updates to unsuccessful vendors is a reasonable stakeholder communication and relationship safeguard, but it fails by concept because rejections carry no spend authorization or contract formation risk comparable to acceptances, so it misplaces human oversight on low consequence action. As an exam caveat, distinguish preventive controls before irreversible spend from detective controls like logging, scoring validation, or archiving that improve accuracy but do not satisfy accountability for authorization. As a practical check, ask whether the agent action commits money, creates legal liability, or acts externally without easy reversal, and if yes require explicit human approval with recorded approver, budget line, and retained evidence.