Codex keeps Claude in check

In this blog post I detail the rapid development of an agentic engine designed to transform a standard AI chat tool into a proactive business intelligence platform. By integrating Anthropic’s Claude Agent SDK, I
successfully automated complex financial reporting tasks that traditionally required manual human prompts. The project highlights a collaborative "two-AI" workflow where Claude acted as the primary developer and Codex served as a rigorous technical reviewer, catching critical architectural flaws. To ensure data integrity, the system utilizes frozen Zod schemas and deterministic tools, preventing the AI from hallucinating sensitive financial figures. The narrative concludes with a successful live deployment, demonstrating how autonomous browser testing and strict contract boundaries can reliably deliver enterprise-grade automated insights.