Podcast – Is AI Reintroducing the Chaos DevOps Was Built to Eliminate?

AI adoption is accelerating across engineering organizations, but according to Lev Andelman, TeraSky’s CTO, many teams are unintentionally reintroducing the exact operational chaos DevOps was designed to eliminate.

That was one of the most interesting themes discussed in a recent podcast episode focused on AI, platform engineering, and operational governance.
The conversation wasn’t anti-AI. Far from it.

The real argument was that most organizations still haven’t developed a clear framework for where AI belongs inside operational systems – and where it doesn’t.

For the last decade, platform engineering has pushed companies toward deterministic infrastructure: reproducible deployments, Infrastructure-as-Code, CI/CD pipelines, GitOps, Kubernetes, and policy enforcement. The goal was consistency, predictability, and the removal of manual drift from production systems.

Now GenAI is changing the equation again.

During the discussion, Lev described a growing tension between deterministic operational workflows and probabilistic AI systems. In areas such as organizational knowledge discovery, incident correlation, or navigating fragmented internal systems, AI can create significant leverage. These are environments with massive search spaces and constantly changing context – exactly where large language models are useful.

But problems start when organizations begin inserting probabilistic decision-making directly into deterministic operational paths.

As Lev put it:
“ClickOps with extra steps.”

It’s a sharp observation because it captures what many engineering leaders are starting to realize. The industry spent years removing variability from production operations, and now some teams are quietly putting it back under the banner of AI automation.

The episode also explored the rise of what Lev described as “Harness Engineering” – the growing need for governance layers around AI-generated workflows. The future advantage may not come from writing better prompts, but from building stronger operational controls around AI systems themselves.

That includes validation pipelines, policy enforcement, orchestration logic, security boundaries, and evaluation frameworks that allow organizations to adopt AI without sacrificing reliability or governance.

The core challenge is no longer whether AI can generate code or automate tasks.

The question is whether organizations can maintain operational clarity and system reliability as software becomes increasingly abstracted.

If you’re currently evaluating how AI fits into platform engineering and operational workflows, this episode is worth listening to.

Listen here: https://www.pageittothelimit.com/harnessing-ai-with-lev-andelman/