SonarQube + OpenAI: Agentic Development — Killian Carlsen-Phelan, Sonar
AI coding agents must provide verifiable evidence of task completion, not just self-reported 'fixed'
“Self-criticism is useful, but it is not an independent oracle, and we often find that the results of the agent and the verifier are often orthogonal to each other.”
A joint OpenAI Codex and Sonar talk at AI Engineer outlined three principles for reliable agentic development: 'done' requires evidence (not self-assertion), isolation/restriction does not equal correctness, and feedback must be inside the agent loop. The talk highlights a systemic failure mode where generated tests propagate the same bugs they're meant to catch, and argues human review cannot be fully replaced in agentic workflows.