Qodo has introduced a plugin for Codex that brings organizational knowledge and codebase standards into developers’ coding sessions. The company says the integration is designed to help coding agents understand the broader systems affected by a change and identify P0 issues before a pull request is created.
Qodo develops AI tools for code review and software quality. Its plugin uses the company’s Agentic Toolbox to give Codex access to engineering rules, codebase knowledge and independent review. Qodo says the system is intended for environments with thousands of repositories, where a change that appears correct locally could still affect a downstream service.
Before Codex begins writing code, the plugin can retrieve relevant repository rules, pull request history and established patterns. Once Codex completes its changes, Qodo reviews the work and sends its findings back to Codex within the same session, allowing developers to address problems before submitting the change for review.
The integration is designed to account for dependencies spanning multiple repositories, including code hosted by different Git providers or connected through services such as Kafka. Qodo co-founder and CEO Itamar Friedman describes the arrangement as a partnership between coding and review agents: one produces changes while the other challenges them.
Qodo’s AI Developer Relations Lead, Nnenna Ndukwe, says developers are using this workflow to build features and address technical debt with GPT-6 models in Codex. The company is also automating the transition between Codex’s coding work and Qodo’s independent review process.
Qodo’s internal workflows include scheduling Codex to summarize weekly accomplishments, identify areas for improvement and research discussions about agentic engineering. The system can also analyze Ndukwe’s code changes and suggest agent skills to create or refine.
Friedman also points to OpenAI’s Decisions API as a potential way to determine which rules and agent skills should apply at each development stage. He gives the example of selecting the relevant checks from a set of 200 rules before a language model reasons through them, which could make the process faster and less expensive while improving review quality.
Qodo says its approach combines AI-generated code, traditional software engineering practices and human judgment. In the company’s vision, coding and review agents would exchange findings and present people with only the issues they cannot resolve, leaving developers to focus on planning, reviewing and making decisions about unresolved problems.
The company describes this model as a foundation for a software factory that customers could assemble and operate themselves, combining coding and review agents with knowledge of how software changes affect production systems.
Sources: OpenAI News.