Let's Talk About AI

Some of you might not know, but my day job is technology. I have a MSCS degree and have held many different roles in my life, from coding to management. At the moment, I’m a part-time consultant.

Let me just say that I’ve seen a lot of tech hype in my life. From my perspective, the AI hype train is going down the same tracks as a bunch of other hype trains. The tracks are:

1. Early adopter individuals: These are people who are going to try out AI, tell their friends about it, and generally be big fans who the AI companies want to recruit.

2. Social Media Users: These are the 2025 equivalent of AOL users. They’re generally not that interested in tech, and they’re influenced by the early adopters in their social circles.

3. Companies that use software: This is where the money is. Vendors like IBM, Amazon, Salesforce and others can charge them big bucks for their AI tools.

I’m sure there are other tracks, but this is simplistic enough for a blog post.

Let’s talk about the early adopters and others who pay little or nothing for AI. There are plenty of them who are getting value from AI because they spend the time to give it good prompts. I know content creators (what we used to call “writers”) and programmers who get some value from AI. AI is genuinely useful in some limited domains, and it will continue to be so.

Moving on to the Social Media users — companies are just jamming AI down their throats. Google’s “Gemini” AI search is an example. Unless you work to turn it off, it occupies a good part of the screen and is sometimes right, and sometimes wrong. It depends on the search. Here’s your tip for the day: go to tenbluelinks.org and use it to turn off Google AI search. The only reason that AI is pooped into search results is to juice engagement stats. You should be able to choose whether to ask AI a question or just search Google. It’s enshittification to have it just thrust upon you, and Google is doing it for the AI hype.

Let’s move on to companies, also known as the “enterprise” market. As anyone who’s worked in a large company knows, IT projects there often fail and often suck. So, there’s always interest in the enterprise market for the new thing. AI is that. Guess what — like every other big hype new tech, AI is a failure at the enterprise level.

That’s a 95% failure rate, my brothers and sisters in Christ. The dream of “here’s a problem, throw AI at it, voila!” is just that, a dream. There’s an old book on software engineering called The Mythical Man Month. It articulates the basic problem with writing enterprise software that existed in 1975 and exists 50 years later: users don’t know what they want, writing code is hard, the code writers and the people who want software have a hard time communicating, and management is always impatient. AI doesn’t solve this problem. Like all other fads, there are a few success stories driven by extremely talented software engineers coupled with smart, articulate users — those projects would be successful using almost any other technology.

So, to be clear, what sucks about AI isn’t AI itself — that technology works for certain applications. It’s, as usual, the hype, the coffee is for closers sales culture, and the gullible managers who buy into the hype.

Unfortunately, AI is also incredibly resource intensive. Krugman has two recent posts (here and here) about the incredible amount of energy that AI consumes. Reader Scott sent a link to a Prospect piece about the data center boom that is plaguing cities — as usual, tech companies are looking for a handout. Even the power companies don’t want to power data centers. They know that this is a boom, and they don’t want to create excess capacity that will sit idle once the boom busts.

This has gotten long, but here are the key points about AI and power:

• Any datacenter project in your backyard should be resisted. They have a huge environmental impact for questionable return.

• At the moment, like most early tech, AI is extremely wasteful. The optimizations available, both hardware and software, can make it much less wasteful. Software is not like the physical world — some small tweaks can make code run 2, 3, 4 or 100 times faster. Similarly, expect special-purpose hardware to make AI tasks run a lot faster and take less energy.

• The AI bubble, like all tech bubbles, is being powered by free stuff given to people who might or might not pay for it. When they turn off the free AI, demand will plummet, as will demand for power.

Or, tl;dr: that echo chorus lied to me: