Why Infrastructure Architecture Is Entering Its Agentic Era

For a long time I assumed that cloud architecture would remain one of the last areas of engineering that resisted automation. Writing code could be delegated. Testing could be delegated. Even parts of design could be delegated. But architecture felt different. It required context, judgment, and the ability to understand how many moving parts fit together. It felt like a space where humans would always have the upper hand.

That belief has been fading quickly.

The shift did not come from bigger models or more parameters. It came from the rise of agentic workflows. Once you start working with agents that can plan, reflect, revise, and orchestrate tools, you begin to see that the barrier was never intelligence. The barrier was structure. Our infrastructure patterns were simply too implicit for machines to reason about.

This became obvious to me when I started running agents locally. I upgraded my home lab from a Tesla K80 to a Tesla V100, not because I wanted to train models, but because I wanted full control over my agent workflows. I wanted privacy, determinism, and the freedom to experiment. Running agents on my own hardware changed how I saw the problem, mainly due to the constraints. The agents were not struggling because they were incapable. They were struggling because the systems we build are vague.

Infrastructure as code is still code. It has dependencies, interfaces, and flows of information. Yet we often treat it as a loose collection of modules and files. We rely on tribal knowledge, naming conventions, and unwritten assumptions. Humans can survive in that environment. Machines cannot. And once you see that, you cannot unsee it.

This is where ABC, the abstract base cloud pattern comes back into the picture.

Abstract Base Cloud

I created the ABC pattern years ago when I was juggling several projects for a customer. Each project used different languages and different stacks. Without a unifying pattern, the IaC would have turned into a mess and fried my brain. My colleagues thought I was overthinking it. They believed that strict patterns were unnecessary. Their codebases eventually collapsed under their own weight. Mine did not. ABC kept me sane.

At the time, ABC was a personal survival strategy. I never imagined it would become relevant again. But with the rise of agentic AI, it suddenly feels like the missing piece.

ABC gives infrastructure a shape that machines can understand. It defines a clear hierarchy of Application Stacks, Logical Units, and Resource Groups. It forces every construct to declare what it needs and what it produces. It restricts data flow to two simple directions: down from parent to child, and up from child to parent. It eliminates lateral references and hidden dependencies. It turns architecture into something explicit.

Once you do that, agents can reason about it. They can generate it. They can validate it. They can refactor it. They can port it across clouds. They can explain it. They can evolve it. They can do all the things we once believed only humans could do.

The breakthrough is not that AI suddenly became smarter. The breakthrough is that I finally have a pattern that exposes the structure AI needs.

This is why I believe architecture is entering its agentic era. Not because humans are being replaced, but because the role is changing. Architects will spend less time wiring modules and more time defining the models that agents operate on. They will focus on constraints, boundaries, and intent. They will supervise rather than assemble. They will curate patterns rather than hand‑craft every detail.

And honestly, that is a future worth embracing.

ABC gives me a way to build infrastructure that is explicit, deterministic, and safe to automate. It gives us a foundation for tools that can finally understand what I mean, not just what I type. For years, I have been waiting for AI to catch up to architecture. It turns out architecture needed to catch up to AI.

Important

The Abstract Base Cloud specification is freely available under a permissive Creative Commons license.

The specification also comes with implementation-specific agent guides (AGENTS.md).

Todo

Provide a from-scratch guide for creating a deployment-ready cloud infrastructure through generic natural language.

---

Comments

Feel free to leave a public comment on my Why Infrastructure Architecture Is Entering Its Agentic Era blog post.

Before you comment...

In order to comment, you need to authenticate yourself with a valid e-mail address. The e-mail address will not be publicly displayed, or shared with anyone, as I (Tiara) also operate the commenting service on my own server, on which your e-mail address is securely stored.
Choose your username in accordance with your privacy expectations.