Agentic Loop Engineering (ALE) 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). The paper describes ALE as deeply rooted in the principles pioneered by the DevOps community, transforming an agent’s work from an opaque, black-box process into a disciplined, auditable, reproducible workflow that moves beyond simple iterative cycles like the Plan-Do-Assess-Review (PDAR) (PDAR) loop.
Usage
The paper assigns the workflow’s definition to the human coach within the Agent Command Environment (ACE) (ACE), while agents execute it within the Agent Execution Environment (AEE) (AEE); its artifact is the LoopScript. Its stated purpose is defining how agents work together (or alone), the patterns of their collaboration, and how they engage their toolset — since agents cannot infer the “stakes” of a task on their own. The paper’s research roadmap for this activity calls for a declarative LoopScript language capturing task decomposition, parallel execution, review checkpoints, escalation rules, and evidence requirements, building on business-process and DevOps automation but with stronger links to code, tests, risk, and human review; interaction mechanisms letting a coach pause a workflow, redirect a branch, add context, or stop low-value work without restarting the whole loop; and standards for sufficient evidence in a Merge-Readiness Pack (MRP), plus more informative tool feedback (structured compiler diagnostics, test failures, or static-analysis findings) to guide agents’ search.
Related Terms
LoopScript, Plan-Do-Assess-Review (PDAR), Structured Agentic Software Engineering (SASE)