A survey of 213 IT leaders, conducted by CloudBees, finds that while 93% report they are seeing productivity gains that are driven by increased adoption of artificial intelligence (AI) tools, a full 81% also report they have seen an increase in production issues attributable to AI-generated code.
Shared this week at an Agentic DevOps World 2026 event hosted by CloudBees, the survey finds nearly two-thirds (64%) of survey respondents reporting that AI is either widely adopted or fully integrated into engineering workflows.
At the same time, CloudBees is making available a Code Abundance Readiness Evaluation (CARE) Index, a proprietary composite score designed to assess how effectively enterprises can track, attribute, and forecast AI-driven costs against productivity outcomes. Based on six dimensions, the CARE Index establishes a baseline for AI governance maturity that can then be used to benchmark progress.
The CloudBees survey, not surprisingly, finds that 67% of respondents have seen a significant increase in code volume over the past 12 months, but only a little more than half (52%) are seeing higher development output in terms of features and pull requests.
Additionally, only 31% said they can correlate AI spending with specific business outcomes. Well over a third (36%) track AI spend without measuring return on investment (ROI) or don’t measure ROI at all, the survey finds.
At the same time, costs are rising. More than half of respondents (54%) report a significant increase in CI/CD infrastructure spend over the past 12 months, while 53% say testing, security scanning, and deployment costs have risen alongside growing code volume. Only 27% have set hard limits or quotas on token usage, and just 18% have implemented automated controls.
Finally, the survey finds 70% of IT leaders now view test suite maintenance as a bigger burden than writing code itself.
Shawn Ahmed, chief product officer for CloudBees, said the survey makes it clear that organizations have yet to address governance, validation and accountability issues that inevitably arise when AI is embedded within a DevOps workflow. Too many organizations are pursuing a “token maxing” approach that encourages developers to use AI tools as much as possible without considering the cost of the total number of tokens consumed, he added.
In fact, the CARE Index finds that less than half (45%) of organizations are finding AI spend to be very predictable quarter-to-quarter.
Ultimately, DevOps teams will need to revisit how workflows are being managed as the volume of code continues to exponentially increase in the AI era, said Ahmed. It’s not exactly clear how much AI-generated code is actually making it into production environments, but there is little doubt that the pipelines through which that code flows will need to be optimized, he added.
In some cases, that may require organizations to adopt new tools and platforms, but regardless of approach, fundamental governance issues will be addressed, noted Ahmed.
The only thing that remains to be seen now is how many issues will arise in the absence of that governance that could have been avoided with a little extra forethought.