CloudBolt finds trust gap in Kubernetes automation
CloudBolt has published research highlighting what it describes as a trust gap in Kubernetes automation. The survey found that only 17% of practitioners allow continuous autonomous right-sizing in production.
The findings are based on a survey of 321 Kubernetes practitioners at organisations with more than 1,000 employees. They suggest that while automation is widely used in software delivery, many teams still want human review when CPU and memory settings are adjusted in live environments.
According to the research, 89% of respondents said automation is mission-critical or very important. At the same time, 59% said they deploy to production automatically without manual approval, suggesting automated software delivery is more widely accepted than automated infrastructure tuning.
The distinction was clearest in right-sizing, where 71% said they require human review before resource optimisation is applied. A smaller group, 27%, said they allow guardrailed automatic application of CPU and memory changes.
Scale pressure
The report also points to the operational burden of manual review in larger Kubernetes estates. More than half of respondents, 54%, said their organisations run more than 100 clusters, while 69% said manual optimisation starts to break down before about 250 changes a day.
The research suggests the issue is not a lack of data about system usage. Instead, teams appear more cautious when automation directly affects performance, reliability, and cloud spending in production systems.
Respondents said they were looking for clearer conditions under which automation could be trusted. In the survey, 48% said visibility and transparency would most increase their trust, 25% pointed to proven guardrails, and 23% cited instant rollback.
Mark Zembal, Chief Marketing Officer at CloudBolt, said the reluctance begins when automated systems are given authority over production resources.
"Everyone says they trust automation right until it requires the authority to act," Zembal said. "Teams will auto-deploy code via CI/CD 50 times a day without blinking an eye. But the moment automation touches cost, performance, or reliability in production, hesitation creeps in. That hesitation is where delegation dies. Teams won't hand over the keys unless the system is explainable, bounded by guardrails, and reversible on demand."
Maturity divide
The data suggests many companies are still at an intermediate stage of Kubernetes operations, where they can observe usage patterns and receive recommendations but stop short of giving systems authority to make live optimisation decisions on their own.
That gap matters because overprovisioned environments can persist when teams recognise inefficiencies but decide manual oversight is safer than autonomous action. The report presents that choice as a practical response to production risk rather than a failure to adopt automation tools.
Methodology for the study was provided by Gather, which worked with CloudBolt on the research. The survey focused on enterprise Kubernetes practitioners and combined structured responses with qualitative input.
Yasmin Rajabi, Chief Operating Officer at CloudBolt, said the survey results reflect a broader progression from visibility to operational trust.
"If you map all of this to a maturity model, it creates a clear continuum from Observe to Advise to Automate to Trust. Most companies are stuck in the early middle," Rajabi said. "They can see the problem. Some can even accept recommended fixes some of the time. But they stop short of letting the right-sizing system act autonomously. The final stage isn't more insight, it's trust. And until teams trust automation to optimise right-sizing in production, they will always be constrained by manual limitations that can never effectively scale."