What Changes When AI Becomes Infrastructure
A LinkedIn post crossed my feed recently discussing the backlash surrounding Anthropic’s suspension of access to Fable 5. Most of the post was fairly standard commentary on AI adoption, but one sentence caught my attention:
“Once AI becomes part of someone’s daily workflow, restricting a capability isn’t just a product decision. It’s a trust decision.”
The more I thought about that statement, the less interested I became in the specific details of the Anthropic situation and the more interested I became in the idea underneath it. What struck me wasn’t the notion of trust itself. We trust software vendors every day. We trust cloud providers to remain operational. We trust operating system vendors not to break critical functionality. We trust service providers to continue supporting products that our organizations rely upon. Trust is already a fundamental part of modern technology. The more interesting question is why people reacted so strongly to the possibility of losing access to a particular AI capability.
I suspect the answer has less to do with AI and more to do with dependency.
For most of the software era, products have largely been viewed as tools. Some tools are more important than others, but organizations generally assume that alternatives exist. A company might decide to move from one CRM platform to another. It might replace an analytics solution, migrate email providers, or adopt a new development platform. Those decisions can be expensive and disruptive, but they are still decisions. The organization retains agency because the technology is ultimately viewed as something it uses rather than something it depends upon.
That distinction sounds subtle, but I think it matters.
Consider how we think about electricity. Nobody inside a business spends time debating whether electricity remains valuable enough to justify its continued use. The same is true for internet connectivity. These technologies are not evaluated continuously because they have become embedded within the assumptions that make modern work possible. Entire processes, business models, and organizational structures have been built around the expectation that they will be available tomorrow. Their importance is not measured by how frequently people think about them. It is measured by how disruptive their absence would be.
What makes infrastructure different from a tool is not the underlying technology. It is the degree to which people have reorganized their behavior around its existence.
That is the thought I keep coming back to when I look at AI adoption.
A year or two ago, most people were experimenting. They were testing prompts, generating images, and trying to understand what these systems could do. Today, many professionals have incorporated AI into the way they work. Developers use it while writing code. Analysts use it while researching unfamiliar topics. Marketers use it to develop content. Executives use it to challenge assumptions, summarize information, and explore strategic questions. In many organizations, AI is no longer a novelty. It has become part of the workflow.
That doesn’t automatically make it infrastructure. Plenty of useful technologies never reach that threshold. What it does suggest, however, is that the relationship between people and these systems may be changing. When a technology becomes integrated deeply enough into daily operations, decisions about its availability begin carrying different consequences. A temporary outage is no longer merely an inconvenience. A capability restriction is no longer simply a product update. The effects ripple outward into processes, expectations, and ways of working that have gradually formed around the technology.
This is one reason I find the broader AI governance discussion so interesting. Much of the conversation focuses on model capabilities, safety, competition, and regulation. Those issues are important, but they are also relatively visible. Less attention is paid to the possibility that organizations may be building dependencies faster than the surrounding norms, expectations, and governance structures are developing. Historically, whenever a technology becomes sufficiently important, questions eventually emerge around reliability, continuity, accountability, and control. Not because policymakers decide those questions are interesting, but because enough people have begun relying on the technology that the answers start to matter.
The internet did not become infrastructure because someone declared it infrastructure. It became infrastructure because businesses, governments, and individuals gradually reorganized themselves around the assumption that it would be available. Looking back, that transition seems obvious. Living through it was probably much less clear.
I find myself wondering whether we are beginning to see a similar transition with AI. If we are, then some of the most important questions facing the industry may have very little to do with benchmarks, model rankings, or even regulation. They may instead revolve around what obligations emerge when a technology stops being something people use and becomes something they depend on.