A few weeks ago, Nikita Bier, Product Lead at X, posted something that stopped me cold. He'd set out to build a video editor into the platform, expecting three months of engineering time. Instead, with a single prompt he had a working prototype in fifteen minutes with AI. His follow-up question: "Will videos even be edited manually in three months?"
I'm ridiculously excited about what AI can do. But that question reveals something everyone in tech is feeling and few are willing to say plainly: when AI compresses the cost of creation to near zero, it doesn't just change how fast things get built. It changes what gets built, how much of it gets built, and whether any of it is worth anything.
We're already watching this play out everywhere. Since ChatGPT launched, credential phishing attacks have increased over 700%. More than half of all spam is now AI-generated. AI-powered content farms are publishing across thousands of websites in sixteen languages. Cold email open rates dropped nearly ten percentage points in a single year because inboxes are flooded with AI-generated outreach nobody asked for.
The pattern is consistent: AI reduces the cost of production, volume explodes, and value collapses. Spam emails, AI-generated social media posts, automated cold calls—they all follow the same curve. More noise. Less signal. Diminishing returns for everyone.
Here's what nobody in infrastructure is talking about: this exact pattern is now playing out in DevOps. And I say that as someone who helped create the problem.
The New Wave of "AI for Infrastructure"
Over the past eighteen months, a new category of infrastructure tooling has emerged. The pitch sounds compelling: AI that writes your Terraform, generates your Kubernetes manifests, and produces your CloudFormation templates so your engineers don't have to.
The number two IaC platform just pulled a third of its 130-person workforce to build an AI agent they describe as "your newest platform engineer"—one that can do in minutes what used to take weeks. Another well-funded startup uses AI to scan your existing cloud environment and auto-generate infrastructure-as-code to represent it, literally creating a code representation of infrastructure that already exists and is already running. Yet another raised over $70 million to build a management platform for the IaC that AI is now generating at scale.
And the hyperscalers? Every single one of them announced AI-powered infrastructure code generation at their flagship conferences last year. One generates CloudFormation and Terraform from natural language. Another builds visual design tools where architects drag components and Terraform materializes automatically. A third transforms legacy templates into its preferred configuration language using AI agents. Let's be clear about what this actually is: it's a preservation play. These are legacy platforms built on decades of architectural decisions they can't unwind. They're not going to rebuild their foundations—the business model won't allow it, the installed base won't tolerate it, and frankly the organizational will doesn't exist to do it. So instead of doing the right thing, they're making it easier to do more of the wrong thing, faster. Dress it up with AI, call it innovation, and hope nobody notices that the underlying complexity just got worse.
Venture capital is pouring fuel on this fire. AI accounted for more than half of all invested VC dollars globally in the first half of 2025. The top infrastructure-focused funds are publishing theses about "AI-native infrastructure" and projecting that AI will write 95% of all code by 2030.
The investment logic is clean: DevOps is a $30 billion market, developers hate writing YAML, AI can auto-generate code, therefore AI-for-IaC is a massive opportunity.
I understand why investors find this compelling. I also understand, more than almost anyone alive, why it's exactly wrong.
Automating the Symptom, Ignoring the Disease
I created Salt. I helped define how the industry thinks about infrastructure-as-code. I watched SaltStack get adopted by millions of teams, get acquired by VMware, and become part of the standard infrastructure stack everywhere. I am genuinely proud of what we built.
I'm also telling you plainly: we created a monster.
Not because the technology was bad. Salt solved real problems. But the model we established—describe your infrastructure as code, manage that code, version that code, debug that code—created an industry of complexity that was never inevitable. It was a choice. And the industry ran with that choice for twenty years in the wrong direction, stacking YAML on top of YAML, abstraction on top of abstraction, until teams needed entire platform engineering organizations just to manage the configuration of their configuration tools.
And now the solution being offered is to automate the production of that YAML at machine speed.
I want to be direct: this is not innovation. This is acceleration toward a cliff.
Let me walk through what these tools actually do. They don't make infrastructure simpler. They don't reduce the number of moving parts. They don't eliminate the configuration drift, the dependency conflicts, the brittle pipelines that break at 2 AM. They produce more of the same infrastructure code, faster. The AI generates a Terraform module in seconds instead of hours. Great. You still have a Terraform module. It still needs to be tested, maintained, version-controlled, and debugged when it inevitably drifts from reality. You've automated the creation of your future technical debt.
With IaC infrastructures every infrastructure is a snowflake, rather than finding patterns and commonality it drives complexity. The complexity isn't incidental, and just like snow, it accumulates. Every generated module is a new surface area for failure. Every abstraction layer you can't see into is a future outage you can't diagnose. When AI is generating your infrastructure code, you're not just accumulating technical debt—you're accumulating technical debt you don't fully understand, at a rate no human team can audit.
This is the infrastructure equivalent of AI-generated spam. You haven't solved the communication problem—you've just made it cheaper to flood the channel.
The IaC market is growing at 24% annually. DevOps adoption has climbed from 33% to 80% in seven years. And now we're adding AI that can generate configuration at machine speed. If you thought managing infrastructure code was complex before, wait until every team in the organization is producing it ten times faster. Google's DORA research consistently shows that teams forced to use complex internal platforms see decreased throughput and decreased change stability. More tools, more configuration, more layers of abstraction—they don't compound into better outcomes. They compound into more failure modes.
I helped build the first generation of this problem. I'm not going to applaud the AI-powered version of it.
The Right Problem to Point AI At
None of this means AI doesn't belong in infrastructure. It absolutely does. But it matters enormously where you point it.
Using AI to generate more Terraform faster is like using AI to write more cold emails faster. You're accelerating an activity that was already producing diminishing returns. The right question isn't "how do we write infrastructure code faster?" It's "why are we writing this much infrastructure code at all?"
We should use AI to eliminate the boring things, yes—but first we have to be honest that many of those boring things should never have existed in the first place. The fact that we've normalized thousand-line Terraform configurations as "just how infrastructure works" doesn't mean that's how infrastructure has to work. It means we've been building on a flawed foundation for so long it started to look like bedrock.
An infrastructure platform built on a sound foundation—one that doesn't require tens of thousands of lines of YAML to describe what should be simple—lets us point AI at problems that actually matter: optimizing resource allocation, predicting scaling needs, identifying security anomalies before they become breaches. Real problems. Hard problems. Not "please generate more of the configuration we've convinced ourselves is necessary."
The distinction matters enormously. There are infrastructure solutions that use AI to automate the production of complexity. And there are infrastructure solutions that eliminate the need for that complexity in the first place. The first category is getting all the funding. The second is where the actual value lies.
And here's the part that should make every infrastructure engineer furious: the first category puts you in less control the more you use it. Every layer of AI-generated configuration between you and your actual infrastructure is a layer of opacity. You lose the ability to reason about what's running, why it's running that way, and what will break when something changes. You become dependent on the tool that generated your infrastructure because you no longer truly understand your infrastructure. That's not a platform strategy. That's a trap.
After building datacenter infrastructures for over 20 years, I've watched this movie before. The industry finds a pain point, builds tools to manage the pain, then builds tools to manage the tools. Every layer adds cost, adds complexity, adds headcount. Adding AI to the top of that stack doesn't break the cycle—it accelerates it. I know, because I was one of the people who built the stack.
The infrastructure industry doesn't need AI that writes more configuration code. It needs architecture that eliminates the need for that configuration. AI should reduce complexity, not automate its production.
That's what we're building at ContextOS. Not AI that generates your YAML—a platform where we've removed the layers of complexity that should never have existed. When the foundation is sound, AI becomes genuinely useful instead of just prolific. When you eliminate the unnecessary abstraction layers, you get infrastructure that's actually understandable, actually debuggable, actually yours.
The companies racing to build AI-powered IaC generation are solving a real problem. I don't question their engineering talent. But they're pointing one of the most powerful technologies in human history at the wrong layer of the stack. I know that layer well. I helped design it. And I'm telling you: automating it harder is not the answer.
The teams who fix the problem at the architecture layer—not the configuration layer—will be the ones who win the next decade of infrastructure. With less complexity, they'll be the load-bearing wall supporting the boom of AI applications, enabling innovation faster, and empowering their organizations to actually compete. Everyone else will be exactly where they are today, just buried deeper.
Writing more YAML. Just faster.