The Agent Execution Environment (AEE) is one of two purpose-built workbenches proposed in Agentic Software Engineering: Foundational Pillars and a Research Roadmap as part of Structured Agentic Software Engineering (SASE) (SASE), alongside the Agent Command Environment (ACE) (ACE) for human coaches. The paper argues that tools optimized to reduce human cognitive load are often suboptimal for agents, which are unburdened by human cognitive limits and instead thrive on raw, low-overhead tools optimized for computational efficiency and structured, machine-readable feedback — pointing to the fact that many of today’s autonomous coding agents still rely on basic utilities like grep as evidence of this mismatch.
Usage
The paper describes the AEE as needing agent-native tools such as hyper-debuggers capable of analyzing vast state spaces, powerful semantic search utilities, and structural editors that manipulate code as abstract symbolic structures rather than simple text. It must also include monitoring infrastructure to manage agents’ operational health — autonomously spotting security vulnerabilities, flagging agents incurring unexpectedly high computational costs, and repairing or replacing broken virtual environments — so that only problems requiring strategic human intervention are surfaced to the ACE. A LoopScript defined in the ACE is executed by agents within the AEE, and building this agent-centric foundation is described as falling within the expertise of the Platform Engineering community.
Related Terms
Agent Command Environment (ACE), Structured Agentic Software Engineering (SASE)