Decision Stubs: Partitioning Human Judgment from Agentic Execution

Decision Stubs: a compact executive briefing that teaches leaders how to partition decisions between humans and agentic systems so you preserve high-value judgment, maintain optionality, and keep execution fast. In ten minutes Adrian explains the core insight: not every decision needs full human ownership, but every automated decision should expose a 'stub' — a minimal, auditable handoff point that preserves context, escalation rules, and undo paths. The episode walks through three supporting points: a simple architecture for stubs (inputs, confidence thresholds, and rollback tokens), a risk-control pattern that ties escalation to incentives and economic SLOs, and metrics that turn agent actions into knowledge capital rather than opaque output. You’ll get a 3-step action checklist to add decision stubs to one workflow this week, plus a concrete one-paragraph prompt and escalation template you can copy. Targeted to executives building agentic systems, the briefing favors practical frameworks over theory so you can take action before the next deployment.