Today, I want to explore the striking similarities between AI agents and Infrastructure as Code—and why this comparison reveals the future of our industry.

Having written hundreds, if not thousands, of IaC definitions over the past decade, I’ve watched this concept mature dramatically. From standard CloudFormation templates and CDK to Pulumi and Kubernetes definitions, the underlying principle remains constant: define your desired infrastructure state in code, then let the system work to achieve it.

Kubernetes exemplifies this perfectly. It continuously reconciles desired states, pursuing your specified configuration infinitely until achieved. Its loosely coupled architecture means each component operates independently, without rigid deployment dependencies.

Here’s where it gets interesting: Generative AI models, particularly agentic execution systems, operate on remarkably similar principles. When you provide input requesting a specific output state, the model reasons through its expertise and available tools to reach that desired outcome.

The IaC ecosystem was built to reduce cognitive load for humans and keep large-scale infrastructure maintainable. Yet even with these tools, maintaining consistency across multi-tenant enterprise systems remains challenging.

You might argue that AI’s non-deterministic nature makes it unsuitable for critical infrastructure management. But consider this: we humans already handle these responsibilities with all our imperfections. The industry has invested billions creating DevOps ecosystems to manage infrastructure while fighting the relentless pace of business changes, user demands, and production growth. I’ve personally witnessed—and experienced—the tears that come with CloudFormation rollback failures and Terraform state corruption.

Within the next few years, we’ll see AI agents performing at the level of well-trained engineers with a decade of experience. They’ll operate at a fraction of the cost with greater accuracy, working around the clock.

This will inevitably displace some jobs.

Managing complex infrastructure will become trivial, requiring less specialized knowledge as expertise becomes embedded in the models themselves. The true experts will shift toward building and maintaining these agents, applying their hard-earned knowledge to guide AI behavior.

Throughout my career, people have asked if it’s possible to “clone” me because 12 hours of work per day isn’t enough. That reality is approaching. In 2015, managing complex auto-scaling infrastructure across multiple regions was a full-time job. Today, you can generate a complete Terraform module during your coffee break that handles it perfectly.

If you’re early in your career and feeling overwhelmed by this change, I understand. I’ve been there multiple times, and you probably have too. Success lies in continuous learning. Change fatigue is real—nobody wants to abandon hard-won expertise or alter ingrained habits. But adaptation is intrinsic to our industry, where Moore’s law drives relentless innovation.

Focus on scaling the right skills for both your interests and your role.

Stay open-minded, but don’t jump off the bridge too quickly.​​​​​​​​​​​​​​​​

— Pierre