AI is Not a Technology, It's a Subscription Company

Back in the 2010s, we had our own hype cycle. It was silly because it hijacked a term we were all already familiar with, then rebranded it as if it were new technology. Those two terms were "Server" and "Cloud." Yes, children, there was a time when we didn't say "cloud computers". We just called them servers.

I remember watching that Larry Ellison video where he rants about cloud computing. Eventually, he conceded, and Oracle started selling "cloud computing."

We grew tired of the term "cloud", it was everywhere. It was the web3 of our time, the NFT of our time, the AI of our time. So a hero decided to take matters into his own hands and created a Chrome extension that replaced all instances of "the cloud" with "my butt."

For a long time, I'd forget I had this extension installed and would be shocked when an article I was reading suddenly referred to my butt.

But today, I thought of something new. We're drowning in AI hype, but what happens when you replace the term "AI" with "subscription company"? Suddenly, everything makes more sense and reveals what we're actually dealing with.

The Great Substitution Test

Here are some recent news headlines and what they would look like:

Original Title	New Title
Some early-career workers are losing their jobs to AI	Some early-career workers are losing their jobs to a subscription company
AI is leading to thousands of job losses, report finds	A subscription company is leading to thousands of job losses, report finds
Jobs that are most at risk from AI, according to Microsoft	Jobs that are most at risk from a subscription company, according to Microsoft
AI is linked to a fourfold increase in productivity growth	A subscription company is linked to a fourfold increase in productivity growth
Senior Developers Ship nearly 2.5x more AI Code than Junior Counterparts	Senior Developers Ship nearly 2.5x more code from another subscription company than Junior Counterparts

Technology vs. Service

When we think of AI, we imagine a technology, a way computers think and solve problems for us. True technology becomes part of your toolkit; you own it, control it, and can use it indefinitely. But what we currently call "AI", generative AI and large language models, operates fundamentally differently.

For most people, AI doesn't run on their hardware, and they have no control over it. Instead, it's an external company like OpenAI, Google, or Anthropic presenting a product that they control. You are merely a user, paying monthly fees for access to their servers.

This isn't like buying software and owning it forever. Stop paying, and you immediately lose access to everything. Your workflows break, your productivity gains vanish, and you're back to square one.

The Subscription Trap

When we say people are losing their jobs to AI, it's not as if their employer now has a model they trained that can handle common tasks. Instead, it's an enterprise subscription to a service that promises to replace workers. Companies like Klarna quickly realized that whatever service they were using to replace their workers wasn't up to par. But by then, they were already dependent on the subscription model.

When we say developers using AI are more productive, it's not that they now have a local model that can better understand and produce code. Instead, they have another company scanning their code and providing generative suggestions. This productivity boost lasts only as long as they keep paying that monthly fee.

Once your workflows depend on these services, the switching costs become enormous. You've built processes, trained teams, and structured work around a particular service. This creates powerful lock-in effects that traditional software tools never had.

Your Data is Their Product

While companies market these services as "assistants," AI is not your assistant. It's a company extracting value from your interactions. As these services provide assistance, they simultaneously harvest your data to improve their models and sell you additional services.

Anthropic, the maker of Claude, recently updated their privacy policy with this statement: all user conversations are now used for training by default, and users are automatically opted in. You must manually opt out to prevent your chats from being used to train their models. This isn't the behavior of a neutral technology tool. It's the business model of a data-hungry subscription company.

Your prompts, your creative work, your business processes, your problem-solving approaches, all of this becomes training data that strengthens their competitive moat while creating your dependency on their service.

The New Outsourcing

We're still at the forefront of this shift. For now, the only way we can be replaced by AI is if we make space for another company to take over our functions.

While people often expect robots to take over their jobs, the reality is more akin to how jobs get exported overseas. Previously, companies would outsource manufacturing or customer service to other countries. Now, they're outsourcing cognitive work to subscription companies that promise AI solutions. In both cases, you have a contracting company managing those positions, and the original company loses direct control over that function.

Even though this article makes me sound like a skeptic, I've worked at an AI company where we built an automated agent that was quite good at customer service. From the perspective of our client's employees, it was an actual agent working alongside them. From our perspective, our entire company essentially had a seat as an agent within our client's organization, and they were paying us monthly for that privilege.

---

By definition, current generative AI and large language models cannot run on everyday hardware. They require massive computational resources that only well-funded companies can provide. This isn't a temporary limitation. It's the fundamental architecture of the business model.

We're not witnessing the democratization of AI technology. We're witnessing the emergence of a new category of subscription dependency, where cognitive functions that could theoretically be owned are instead perpetually rented from companies that maintain control over both the service and the data you provide to make it work.

---