The Plan-Do-Assess-Review (PDAR) loop is an iterative workflow for a single agentic task’s lifecycle, discussed in Agentic Software Engineering: Foundational Pillars and a Research Roadmap: a human and AI plan the work, a dev-agent implements it, an agent self-assesses the result, and a human reviews it. The paper describes it as commonly used together with a Product Requirement Prompt (PRP) (PRP), a “minimum viable packet” capturing goals, justification, acceptance criteria, and curated context, and cites Amazon’s Kiro as an industry tool already demonstrating this spec-driven pattern.
Usage
The paper credits PDAR as an early, crucial attempt to impose order on ad-hoc agentic prompting, and states its structured, testable intent aligns with Structured Agentic Software Engineering (SASE) (SASE)‘s own insistence on structure. It positions PDAR as scoped to one-off task execution: by itself, it does not establish durable mentorship, agent lifecycle learning, or cross-task traceability, which SASE treats as first-class concerns addressed instead by artifacts such as MentorScript and LoopScript.
When It Applies
The paper presents PDAR as an existing industry pattern — attributing the PRP structuring specifically to tools like Amazon’s Kiro — rather than a practice it originates, and frames its own contribution as extending PDAR’s single-task loop to team-level, N-to-N human-agent collaboration.
Related Terms
Product Requirement Prompt (PRP), Kiro, Agentic Loop Engineering (ALE)