The Agent Command Environment (ACE) 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 Execution Environment (AEE) (AEE) for agents. The ACE is the command center for the human “Agent Coach,” a workbench optimized for human cognition rather than code-centric editing, offering full observability into agent activities and their associated costs.
Usage
The paper describes the ACE as supporting both 1-to-N collaboration, where one developer works with many agents, and N-to-N collaboration, where a human team coordinates a shared fleet of AI teammates. It is where a coach authors and iterates on a BriefingScript, defines a LoopScript, and reviews structured evidence bundles such as a Merge-Readiness Pack (MRP); the environment also routes, presents, and records a Consultation Request Pack (CRP) when an agent needs to escalate a decision to a human specialist. The paper describes the ACE as needing capabilities missing from standard development tools today: disciplined N-version programming support (letting a developer visualize, compare, and mix components from multiple agent-generated solutions), program-comprehension views beyond simple textual diffs, strategic agent-team management (composing, evaluating, retraining, demoting, or retiring agents by capability and cost), and the ability for the coach to “jump in” to a traditional IDE view for a surgical code change before returning to the coaching workflow. It also proposes voice as a complementary interaction modality for high-level orchestration and mentorship tasks, citing evidence that speech can be faster than typing and that voice-assisted debugging can reduce context switching, and naming Talon Voice and Cursorless for VSCode as existing tools demonstrating the feasibility of speech-based agent interaction.
Related Terms
Agent Execution Environment (AEE), Structured Agentic Software Engineering (SASE)