[arXiv]score: 0.13

Explicit Delegation Contracts Improve AI Coding Agent Reviewability in Controlled Study

June 17, 2026

A 64-run pilot study across two model tiers tested three prompting conditions for coding agents — issue-style prompts, explicit delegation contracts, and contracts with evidence bundles — scored via hidden acceptance tests and mutation checks. The delegation contract framing is proposed as a unit of analysis for measuring agent authority, scope, and work quality.

HOW THIS AFFECTS YOU

●

builderYou can use the delegation contract structure — task, authority, returned work package, acceptance context — as a practical prompt template to improve reviewability of AI coding agent outputs.

●

researcherProvides a small but structured benchmark environment for evaluating coding agent behavior under different authority and prompting conditions.

read original ↗arxiv.org

← back to feed