In the world of IT operations, there’s a long history of “cowboys”—system administrators who would often log into servers directly, making quick fixes with little planning or repeatability. This cowboy approach led to chaos and outages, prompting enterprises to adopt better tools like configuration management, immutable infrastructure, and strict access controls. Now, a new kind of cowboy is riding into town—agentic AI—that promises to make sysadmins’ lives easier by handling routine tasks automatically.

The Return of the Cowboy in a New Form

Agentic AI, especially large language models, can perform tasks like logging into servers, fixing broken applications, or updating old systems—all without human intervention. This sounds great in theory. It could mean fewer manual tickets, less on-call stress, and around-the-clock problem solving. But there’s a catch: these AI models are inherently unpredictable. Given the same prompt, they might produce different results each time, which can be dangerous for critical systems.

For example, an AI might decide to apply different configurations or deploy varying patches on different days, even if the initial approach was successful. This non-determinism could send organizations back to the chaos of the cowboy era, risking inconsistent systems, security gaps, or outages. Even seasoned sysadmins know that relying blindly on AI without safeguards can lead to problems—especially when systems are in production and need to be stable and predictable.

Balancing AI Use in IT Tasks

IT teams need to think carefully about when and how they use AI tools. The key is to separate tasks into two categories: deterministic and non-deterministic. Deterministic tasks are those that must behave exactly the same way every time, like running scripts, deploying containers, or executing pipelines. These should be tightly controlled and predictable to maintain system integrity.

Non-deterministic work, on the other hand, involves exploration and experimentation—searching documentation, testing different configurations, and troubleshooting issues. AI is well-suited for this kind of task, helping engineers find solutions or generate ideas. Technologies like retrieval-augmented generation (RAG) and tool-calling models can bridge the gap by grounding AI outputs in real data and systems, making the results more reliable.

Ultimately, the goal is to leverage AI to assist with exploratory work while keeping critical, repeatable operations under strict control. This way, organizations can enjoy the benefits of automation and innovation without sacrificing stability or security. It’s about finding the right balance so that AI becomes a helpful partner, not a source of chaos.

Inspired by

• https://www.infoworld.com/article/4119285/ai-agents-and-it-ops-cowboy-chaos-rides-again.html