Decision-Scoped Delegation
Hear it: Handing AI the Car Keys — the audio overview
Running time 23:28. The plain-English walkthrough of the Agentic Frontier: what machines may be given, what people answer for. AI-generated audio overview created with NotebookLM, based on Ed Fassio's body of work.
Open the audio file directly if the player does not work in your browser.
Read the paper · DOI 10.5281/zenodo.23126767 Cite the companion · DOI 10.5281/zenodo.23126782
The industry treats delegation like a maturity score: as models improve, you hand them more. Decision-Scoped Delegation rejects that framing. A delegation is scoped to a decision: a named decision-maker, defined limits, a review trail, and the power to halt.
The technical paper "Decision-Scoped Delegation" (v1.1) is published with a Zenodo DOI and defines the admission predicates and four failure modes of delegation: cumulative exposure, approval drift, ambient child authority, and orphaned sponsorship.
The framework is the analytical core of the Constitution of the Frontier and the engine behind the AI Governance and Enablement Foundation offering.
Common questions
What is decision-scoped delegation?
A framework by Ed Fassio in which delegation to an AI is treated as a contract for a consequential decision rather than a maturity badge: each delegation names a decision-maker, defined limits, a review trail, and the power to stop the work.
Who is accountable when AI makes a mistake?
The named human who accepted the delegation contract. Decision-scoped delegation requires a Sponsor of Record for every consequential use — the answer is never "the AI decided" or "the committee".
How much autonomy should you give an AI agent?
Only as much as a named human is prepared to answer for, scoped to specific decisions, with limits and a review trail set in advance — and never the authority to decide what may be delegated.