Kubernetes: Trust Automation to Deploy Code, but Not to Touch the CPU. AI Raises the Stakes

The Automation Dilemma in Kubernetes

Kubernetes teams have embraced automation for code deployment without a second thought. CI/CD pipelines run dozens of times a day, autoscaling adjusts replicas in seconds, and no one hesitates to trust an automatic rollback. However, when it comes to optimizing CPU usage or adjusting resources at the node level, distrust appears. This paradox, highlighted by The New Stack, intensifies with the emergence of artificial intelligence, which promises to automate these critical decisions as well.

Why Do We Trust CI/CD but Not CPU Tuning?

The answer lies in the perception of risk. A deployment failure is easily reverted; a mistake in CPU allocation can degrade the entire cluster. SysAdmins and DevOps know that overloading a node affects multiple services, and the impact is less predictable. Moreover, automatic resource tuning tools (like the Vertical Pod Autoscaler) are not yet as mature as CI/CD pipelines. Generative AI and machine learning models promise to change this, but trust is not earned overnight.

Business Impact: Efficiency vs. Control

For the business, full automation means lower operational costs and greater speed. But resistance to delegating decisions on critical resources hampers efficiency. Infrastructure teams spend hours manually adjusting CPU and memory limits, while AI could do it in real time. The key lies in implementing monitoring and governance mechanisms that allow auditing and rollback of automated decisions, similar to what is already done with deployments. At ForgeNEX, we have analyzed cases like firewall configuration in cloud environments, where automation with oversight has proven effective.

AI as a Catalyst for Change

Artificial intelligence is raising the stakes. Tools like Kuberhealthy or Goldilocks already suggest adjustments based on historical metrics, but new AI systems promise to learn load patterns and anticipate spikes. However, trust requires transparency: teams need to understand why AI recommends certain changes. The solution involves combining automation with human-in-the-loop in initial phases, then moving toward full automation as trust grows. This approach is similar to that applied in security in AI development environments, where human oversight remains critical.

Source: The New Stack. ForgeNEX analysis.