AI at the tail end of 2026

I’ve been meaning to weigh in on AI for a while. I mean, do I even exist if I don’t write something on the internet about AI? Will anyone ever be able to take me seriously if I don’t contribute to the media churn about the most recent computer fad? I missed “the cloud” and “cryptocurrency”; I won’t make the same mistake with AI!

There are three things I want to comment on. The first is to add to the page ranking of a blog entry by Rich Hickey, which might be the most novel and insightful observation I’ve come across, amongst all of the regurgitation of opinions by people who not even don’t understand the most recent iteration of AI, but who don’t know they don’t understand it. Rich argues that “‘AI’ is Not a Tool”. I don’t want to spoil it, because it’s short and worth the time to read, but Rich proposes that AI is not a tool, it’s automation, and he makes a compelling case in less than a page of text. That’s it: I have nothing to add, I just want to funnel more people to the article.

The second thing is that my wife (who’s not in IT) recently mentioned that she had stopped using it at work because it wasn’t adding any value. By itself, that could just be one person’s experience, but as I was closing the door of her office on a WFH day I heard the presenter on her conference call say: “we ask that nobody use AI note takers on this call”. It wasn’t a special call; it was just a bog-standard meeting. Businesses are slowly starting to realize that they risk leaking company secrets if they use LLMs, and that the purveyors of AI tooling are essentially unable to prevent it even if they wanted to. This leads back to the truism that’s well-known within IT circles, even if less heard outside of it: the cloud is just other people’s computers. And AI is all in the cloud. On other people’s computers. And while non-IT people my find the distinction fuzzy, IT executives within companies should well understand that there’s a fundamental difference between running your infrastructure in the cloud, and using LLMs hosted by a small handful of companies.

It may be that companies will start to run local models to avoid leaking IP, whether by standing up bespoke hardware for true security or time-sharing with their competitors like they currently do. However, there’s a clear consequence, and this is my third observation: if this self-hosting becomes common, it threatens the revenue of projects like Gemini and outright tolls the death knell for companies like Anthropic. AI companies are already underwater, by some estimates spending $10 for every $1 they realize. Anthropic is leading an AI investment collapse, with recently leaked documents stating that the company’s income is $20 billion lower than expected. When the financial sector starts to realize companies are avoiding using ChatGPT and its brethren because of privacy concerns, it’ll cause another hit to the tech stocks like the massive 1.5% loss the NASDAQ took on the Anthropic news.

The recognition that there is an AI bubble has escaped the IT space and is now being discussed openly in mainstream media. It’s less a question of if it’ll burst, as a question of when. How many hits like Anthropic’s revenue prediction miss will it take before the bubble bursts? I don’t know; if financial markets were predictable there’d be a lot more Warren Buffets. The only thing I’m sure of is that, because so much of the money involved in AI markets today is phantom money, it’s going to be painful for everyone when it hits.