Nnenna Ndukwe presents quality as a concern across the software lifecycle, not just a final code review. Agentic coding needs shared context about expected behavior, organizational standards and failure conditions. Repository instructions, skills and planning templates can make those constraints available in the editor, command line and continuous integration workflow.
Nnenna Ndukwe uses a small FastAPI payment service to demonstrate a plan-first approach. Before implementation, the plan specifies the task, applicable rules, acceptance criteria, tests and recovery constraints. Deterministic formatting, typing, security and test tools remain part of the workflow rather than being replaced by an agent's assertion that its work is correct.
Nnenna Ndukwe combines implementation with Qodo's contextual pull-request review. The demonstration surfaces a refund idempotency problem: process-local state is not a durable safeguard against repeated financial mutations. Findings are categorized by correctness, security, maintainability and organizational expectations, then used to guide a focused correction and another verification pass.
Nnenna Ndukwe distinguishes providing an agent with rules or skills from proving that it uses them effectively. Suggested organizational rules require human approval, relevant context retrieval is not guaranteed, and evaluations remain a separate responsibility. The example illustrates a layered review process, not a guarantee that AI-generated software is free of defects.
Watch on YouTube
