New research from feature management platform LaunchDarkly reveals a growing "AI Control Gap" in software development: while AI tools are significantly speeding up code production, they are simultaneously driving up operational risk, developer burnout, and customer disruption.
According to the survey, 94% of teams say AI has accelerated code development. At the same time, 91% say they have become more cautious about pushing AI-accelerated changes into production environments, highlighting a widening gap between speed and confidence in live releases.
Risk under pressure
The study paints a picture of development teams working under significant deadline pressure. It reports that 81% of teams admitted to knowingly shipping risky code in the last six months due to deadline pressure. That behaviour has knock-on effects in operations and support.
More than a third of teams, 38%, now spend more than a quarter of their time resolving incidents. That level of time spent on firefighting suggests a persistent cycle in which fast delivery creates more breakages, which in turn erodes capacity for new work.
The internal strain is reflected in staff wellbeing. As 42% of professionals report lower morale and increased burnout due to the current pace and risk profile of delivery.
The report states that nearly all organisations have some form of safeguards in place. A total of 99% say they have runtime guardrails around production systems. Despite that, about 70% still roll back changes at least weekly, indicating frequent problems serious enough to reverse live deployments.
It also finds that 69% of teams apply hotfixes or perform rollbacks weekly or more. Those activities tend to add complexity to release pipelines and increase the risk of further defects.
Customer impact
The LaunchDarkly data suggests that the effects of unstable AI-driven releases extend beyond internal teams. More than 60% of organisations report customer-visible issues such as performance degradation, missed service-level agreements or customer churn. These issues point to direct commercial consequences from flawed or rushed releases.
Furthermore, 45% of respondents report that slow or risky release cycles have resulted in missed market opportunities. That sentiment reflects a concern that organisations either move too fast and damage customer trust, or move too slowly and miss competitive windows.
Use of AI in software development remains widespread despite these concerns. The report says 83% of respondents believe that releases containing AI-generated code are somewhat or very likely to cause production issues. That perception underscores a lack of trust in AI-authored changes once they leave controlled development environments.
The study indicates that engineering and operations teams attempt to mitigate these issues through guardrails, monitoring, and rollback mechanisms. The frequency of reversions and hotfixes suggests those safety measures are under strain as the volume and complexity of AI-assisted changes rises.
Call for control
LaunchDarkly commissioned the research from independent firm UserEvidence. The survey covered 767 senior engineering and DevOps professionals. Respondents worked in mid-market and large organisations across sectors including software, financial services, telecommunications and retail.
The report introduces the phrase "AI Control Gap" to describe the mismatch between how quickly AI enables code to be written and how safely organisations can deploy it into production.
"Our research reveals the AI Control Gap: the gap between how fast AI enables us to ship code and how safely we can deploy it in production. The tension between speed and control is becoming one of the biggest challenges in modern software delivery.
"Success shouldn't just be measured by how fast teams ship, but how safely they do it. As 77% of teams still struggle with the pace and risk of AI-accelerated development, the business case for resilience is growing, and enterprises must lead by making controlled release, progressive delivery, and runtime stability core business metrics, alongside speed and cost," said Joe Byrne, Global Field CTO, LaunchDarkly.
The company frames the findings as evidence that organisations will place greater emphasis on controlled releases and stability metrics as AI-driven development continues to expand.
