Enhancing Agentic AI with PDCA Framework

I spent last week at the Retrospective Facilitators Gathering (RFG). The conversations sharpened how I think about agentic AI.

Retrospective facilitators and trainers run learning loops for a living. A Plan-Do-Check-Act cycle applied to teams, product thinking, and organizational change.

Thanks mostly to Andreas Schliep, I left thinking about how to apply what I’ve learned about PDCA in agentic coding to aid other complex human tasks. So I added a scaffold to my Claude skill for building any domain-specific PDCA skill.

Here’s how it works:

A human expert describes a complex repeatable task. The scaffold goes through five layers of Socratic questions: what is the task, where is human judgment essential, what does quality look like, when should the agent stop for feedback, what could improve each cycle. Then it generates an installable Claude skill tailored to the task.

Content pipelines. Data analysis. Client reporting. Operations. Legal review. Anything complex enough to need structure, routine enough to create a repeatable process. Anything that benefits from continuous improvement.

Then, each time it is used, the skill ends with a micro retrospective that identifies specific changes to itself. The user evaluates and adopts them. The skill gets sharper with use.

Retrospection is essential. The RFG reinforced that for me. Applied learning is what elevates tool use to mastery.

#RFG2026