Follow Us on Google
A Texas-registered software governance company, Foixar, has disclosed that providing artificial intelligence (AI) models with more context improves their knowledge but does not make them accountable for the quality or compliance of the code they generate.
The position was contained in a statement released by the company following the launch of its governance platform this week, in which it explained why it believes enterprise software development requires an independent enforcement layer beyond retrieval-based AI systems.
According to the statement, many engineers and technical buyers initially assume the platform is simply another retrieval-based solution that feeds AI models with documents and coding standards before they generate code.
Foixar, however, maintained that retrieval and contextual information operate only on the input side of AI-assisted software development. While better context may produce better suggestions, the company said those outputs remain suggestions rather than guarantees of compliance.
The statement noted that AI models can still misinterpret retrieved information, ignore architectural rules or generate code that departs from established engineering practices, with nothing inherently preventing such code from being merged into production.
To address this, the company said governance should occur after code generation rather than before it.
According to the statement, the platform analyses the actual code changes contained in pull requests and compares them against the software architecture detected within a repository, flagging code that falls outside established structural patterns or appears in inappropriate locations.
Foixar said this approach differs fundamentally from retrieval because it independently examines the finished code rather than relying on an AI model to follow instructions contained in documents.
The company described the distinction as comparable to the difference between an author writing with reference materials and an independent reviewer assessing whether the completed work complies with required standards.
The statement further argued that governance should apply regardless of how code was produced.
It noted that software entering a pull request may have been written manually, copied from external sources or generated by different AI tools, meaning such code may never have passed through a retrieval process.
According to Foixar, its governance checks evaluate all submitted code equally, irrespective of its origin.
The company also argued that accountability requires an audit trail.
Unlike retrieval systems, which it said simply generate improved responses, Foixar stated that its platform records whether code passes or fails governance checks and identifies the architectural rules applied during the review process.
The statement added that even the company’s repository of architecture decisions serves purposes beyond supplying prompts to AI models.
According to Foixar, those decisions can be versioned, updated and audited while also forming the basis for governance checks.
Foixar’s founder, quoted in the statement, described the difference as one between an AI system knowing a rule and being held accountable to it.
He said additional context may make AI better informed, but it does not make its output answerable to organisational standards.
The company acknowledged that enterprises would ultimately decide whether a dedicated governance layer offers sufficient value over increasingly capable AI coding assistants.
However, it expressed confidence that as AI-generated code becomes more prevalent in enterprise software development, organisations will increasingly prioritise certainty about the checks applied to generated code rather than focusing solely on improving the information supplied to AI models.
Foixar said its platform runs on customers’ own AI infrastructure and is now commercially available.
