AI Teammate Mentorship Engineering (ATME) is one of the structured engineering activities proposed in Agentic Software Engineering: Foundational Pillars and a Research Roadmap as part of Structured Agentic Software Engineering (SASE) (SASE). Its stated purpose is to ensure agent-generated code is not just functional but maintainable and aligned with team culture, by treating agent guidance as first-class code — what the paper calls “mentorship-as-code.”
Usage
The paper assigns guidance under ATME to the human coach, durably consumed by agents; rules are authored in the Agent Command Environment (ACE) (ACE) and directly influence agent behavior in the Agent Execution Environment (AEE) (AEE). Its artifact is the MentorScript. The paper’s research roadmap for this activity calls for abstractions expressive enough for nuanced guidance but simple enough for teams to review as ordinary engineering artifacts; quality assurance for mentorship rules themselves — linting, testing, conflict detection, and regression checks, studying how to verify new rules improve agent behavior without unintended side effects elsewhere; and mechanisms for agents to learn and explain candidate rules from repeated human feedback while keeping those rules reviewable and auditable, connecting an agent’s decision to the specific rules it considered.
Related Terms
MentorScript, AGENTS.md, Structured Agentic Software Engineering (SASE)