AI-generated code is in production at 44.7% of firms

Flux has published a report on the use of AI-generated code in software development. The study found that 44.7% of organisations already run AI-generated code in production.

The findings are based on research by Dimensional Research among 309 engineering leaders and practitioners across five continents. Another 35% of organisations use AI to write code but do not release that code into production because they lack enough visibility into what it changes.

The results suggest AI-assisted coding has moved beyond trial use in many engineering teams, but oversight processes have not kept pace. Flux argues that the main constraint has shifted from generating code to reviewing, understanding and trusting it before release.

Teams are most likely to use AI for lower-risk, repetitive tasks. Documentation was the most common use case at 68.7%, followed by unit testing at 65.9%. Simple functions and code review each stood at 57.7%.

That pattern suggests organisations are applying the tools cautiously. Engineering groups appear more willing to rely on AI output when tasks are predictable and easier to verify.

Review pressure

The report found that 80.5% of organisations have changed development and release processes to account for AI-generated code. Even so, respondents said the hardest issues to catch from week to week were security problems at 49.2%, dependency changes at 47.7% and performance impacts at 44.1%.

Only 3.6% said issues introduced by AI never reach production. That suggests most teams still expect some errors or unintended consequences to pass through existing checks.

The findings also point to concern beyond engineering teams. Respondents reported unease among security teams in 62.5% of organisations, compliance teams in 51.5%, Chief Technology Officers and Chief Information Officers in 46.9%, and legal teams in 40.8%.

These figures show that AI-generated code is being treated as a governance and operational issue as much as a developer workflow matter. In large organisations, concerns about software changes can extend into audit, security review and legal risk.

Tool spending

Some companies are responding by adding more automated checks. Among respondents, 45.6% said their organisations had bought code quality analysis tools, while 39% had added automated code review.

Interest in further controls also appears high. According to the study, 76.4% said a tool that reduces the risks of AI-generated code would be very or extremely valuable.

That demand reflects a broader change in software development since generative AI coding tools spread across corporate engineering teams. While such tools can increase output, they also create more work for human reviewers and for systems that test code, trace dependencies and identify security weaknesses.

Ted Julian, Chief Executive Officer and Founder of Flux, said the core issue is that existing oversight methods were built before AI began producing code at scale.

"Engineering leaders are being asked to embrace AI while simultaneously justifying the expense and mitigating the risk, typically with the same tools they used before AI wrote any code. You can't bolt AI-speed development onto a human-speed view of the codebase and stay in control. Teams celebrate the productivity gains while flying blind on what's changing in their code, but you can't manage what you can't see," Julian said.

Flux describes itself as a code-focused engineering intelligence company, and the report aligns with a growing market focus on observability, governance and quality controls around AI-assisted software development. As more companies deploy code written partly by machines, the commercial opportunity has widened for vendors offering code analysis, automated review and risk monitoring.

The survey results suggest many teams are still balancing speed against caution. The fact that a sizeable share of organisations use AI to write code but stop short of shipping it points to a transition period in which adoption is broad, but trust remains uneven.

Aaron Beals, Chief Technology Officer of Flux, said release decisions should reflect that shift.

"Many teams still measure success by how much code they ship. Instead, they must treat shipping AI-generated code as a risk decision, scaling review to match AI outputs, investing in safeguards, using code-first visibility to surface risky changes and hotspots, and keeping humans in the loop on key decisions," Beals said.