A Plan-Do-Check-Act Framework for AI Code Generation – InfoQ (www.infoq.com)

InfoQ outlines a practical Plan-Do-Check-Act (PDCA) framework for human–AI code generation designed to curb the quality, integration, and delivery problems that have accompanied rapid AI adoption. The author prescribes working agreements and structured prompts that force repo-wide analysis, break features into atomic, testable increments, and keep iterations to 1–3 hour micro-batches. Key evidence motivating the approach: a DORA finding tying heavier AI use to reduced delivery stability, and industry analyses showing large rises in duplicated code and clone-propagated bugs. The framework emphasizes accountability by placing artifacts (analysis, plan, and completion checks) into standard tracking (e.g., Jira) and by requiring the agent to make reasoning visible so developers can intervene. Technically, the PDCA cycle enforces test-driven development and a red-green unit-test loop: agents must write failing tests first, implement code to pass them, then run completion analysis and short retrospectives (5–10 minutes) to refine prompts and practices. The planning phase mandates codebase searches for existing patterns and explicit stop/go criteria; the do phase limits attempts and batches related changes to reduce scope creep; the check phase verifies docs, tests, and acceptance criteria. The result is a repeatable, auditable human–AI workflow that leverages structured prompting (shown to outperform ad-hoc prompts) and empirical PDCA benefits (studies report defect reductions) to improve maintainability and reduce regressions.