The AI Economy Is About to Change (Here's Why)

本站添加: 2026-05-23

[00:00:06]Now, you may have noticed a bit of a shift in the culture around building with AI recently in the memes because something has changed. We started off with the memes about you can build anything with one prompt to now people complaining about clawed rate limits in life whilst I'm waiting for my clawed rate limit to reset where they're rock climbing and they're, you know, scuba diving, whatever. So, I'm going to talk about exactly what is going on here, and it's probably not what you think because every video I've watched on this topic is either too simplistic or just basically wrong. And I'm going to split the video into two sections. The first perspective is going to be from the consumer. So, probably you, someone who uses these AI tools. Then I'll give you the perspective from the AI company's side. And then I'll give you my opinion on how I see this impacting the future of coding and what you can do to prepare. But let's start with the perspective from the average consumer who uses AI. Now, if you're familiar with venture capitalists and also Silicon Valley startups, then this might seem familiar because a playbook is being used which is actually the same playbook used by dealers. So, let's talk about it. Now, there was a time where you pay $200 a month and you could basically get unlimited AI usage, which is kind of insane looking back. But what happened was you have the moderate user and the power user. And I saw a video recently that which really just like encapsulates this idea of the power user and how because of the power user, the economics around AI token users is just not sustainable for these AI companies.

[00:01:37]There is this 18-year-old on the internet somewhere with this remarkable haircut and this very bold shirt. He has 12 $200 a month codec subscriptions and he can't code and he doesn't know what git is. But he set up this thing where every time he hits a rate limit on one of these 12 accounts, he deletes the account and starts a new one. This guy has spent 60 billion tokens. He is generating hundreds of thousands of lines of code. He is merging everything into one repo because he doesn't understand like work trees and stuff.

[00:02:09]So, this guy is I don't know what the hell he's building, but I hope the Amazon rainforest is still around by the time he's done with this project, but he is a definition of a power user. This guy is like Open AI and Anthropic's worst nightmare. So, you have these power users who figured out you could just run agents 24/7 and just generate loads and loads of code. The thing is though, these AI companies factored this in. They were always aware that this could happen because they're following a very specific playbook. And for context, I want to explain what is going on with this business strategy which is well known in Silicon Valley called the growth and squeeze phenomenon and how this is playing out and how it's very similar also to what dealers do. So there's three stages and this is currently playing out right now. Stage one is subsidized growth. So venture capitalists will throw billions at AI companies. For example, Open AAI has received a incredible amount of VC money. And what they're doing is they're investing that in these AI tools to offer them at prices way below the operational costs which make it sustainable. So you want to flood the market and get a dependent user base who love this product. So in that sense, the losses are in the strategy. Stage two, you want to integrate that into their workflow. So train these users using these very powerful tools to the point that they become so essential in their daily work. For example, with vibe coders in terms of software engineering, you now have companies who are judging their employees based on their AI usage.

[00:03:43]You have engineers, engineering managers who've spent years learning to code, but when you press a button and you see it just generate all this code to go back to manually writing a lot of code, people now want to take care of the architecture stuff. So we've reached this point now where we have a dependent user base on these products and what they want to do is to create these switching costs which are very high. So like user lockin where for them to switch to even to a different model provider or to to switch away from coding with AI it's just going to be so painful they'll just swallow the costs.

[00:04:16]Stage three is the squeeze which is what is going on right now. So they raise prices, cut features, so it's a worse product and they shift to a usagebased pricing and this is how you recoup all that VC money which has been flooded in achieve profitability at the expense of the user experience. An example of this is GitHub Copilot. So they offered loads of free features, access to the top models, and now they've just stripped away a lot of that. Now this doesn't mean that we're going to go back to coding all the boiler plate and everything manually. The culture I think is clearly shifted to AI assisted coding and what we're actually seeing is a reduction in the price of tokens due to the reduction of inference costs and that is being driven by hardware. So there's big progressions happening there with Nvidia driving a lot of that model efficiency. So there's different techniques they're using to make these models more efficient and also the rise of small language models which they can host locally. Now, with this video, I wanted to give you a balanced perspective so you can see both sides of what's going on with the AI economy. So, we've talked about it from the consumer level. Let's talk about it from the AI labs point of view because they can just say that this is just normal SAS style pricing maturity which you see even with mobile apps where usually you get a free trial, right? And SAS products usually reward early adopters. They get it cheap. There's free tiers and now the product is maturing and shifting to a pay-per-use model. Now, also a massive elephant in the room here is the lack of GPUs and compute because AI isn't actually getting more expensive per token. Prices have fallen dramatically over the last few years. The real issue is that people are using way more tokens than anticipated, like the guy I showed earlier in the video. And I think tokens also are a misleading metric. They're easy to build and measure because they mostly track how much text you send and receive, plus API calls, but not how smart or useful the model actually is to you. Okay, but now let's talk about the future and how I see things going and how I think this impacts coding with AI, you know, indie hacking, vibe coders, but also big developer teams is that there's going to be a much more of a culture change shifting towards token efficiency, token management, actively being a part of that process of managing how you're using your tokens. Things like caching aggressively, and just maintaining a clean code base, so you're just more efficient with the way you use tokens. So all these things are making tokens cheaper. But what's actually happening is that we're seeing the real cost, the real price of this. This free era we had of, you know, unlimited AI usage. That was the trailer to the main event. This is the main event and this is what these products actually cost.

[00:06:52]But what's happened now is that you have this huge user base and a lot of people are just addicted to using AI and a lot of people are addicted to coding with AI. So, you've built a whole ecosystem of vibe coders who literally they can't code without AI. And there's one message which I'm consistent on on this page and all of my content. And it's this idea when you're thinking about AI use, there is productivity and there is learning. A lot of companies now are forcing their developers to be more and more productive. So, you you have to be productive, but it's a balancing act.

[00:07:23]Your job and even if you're a vibe coder, your job is to learn as much as you can. So, you got to wrestle with them and make sure you're having that balance of productivity and learning because if there's one thing they can't put a rate limit on, that is your skills. Anyway, hope you enjoyed the video. See you in the next one. Happy coding. Ciao.