“The cost of a thing is the amount of what I will call life which is required to be exchanged for it, immediately or in the long run.” — Henry David Thoreau

A few years ago, or perhaps even up until late 2025 or early 2026, a free AI plan was more than enough to slam your entire codebase as context and just get shit done without worrying much - playing a casino-like game in your terminal, pulling the lever until you hit a jackpot mindlessly. But that’s not the case anymore. That window has closed, and even a $20 subscription won’t last a day if you’re doing some serious work with your AI coding agents.

The AI economy was never real to begin with. All the large corporations, investors, and leaders were subsidizing your usage, your game of casino, betting on market share. That bet is getting called in. More and more people are frustrated with not being able to vibe-code all day mindlessly like before, just because AI frontier companies have slammed a huge price tag on every token you spend. A simple task to modify a function or implement a new one can use up a million or more tokens, eventually costing you more than a dollar depending on the model and the task’s complexity with respect to the size of the project’s context.

On How People Actually Use This Stuff

I don’t think most people even realize what’s physically happening on a scale when they send a simple prompt to their favorite therapist chatbot.

These are people running hours-long conversations with LLMs about their personal problems, traumas, and whatnot. People are generating thousands of images that they barely care about or might use once. Entire social media feeds, news outlets, article platforms, book reviews, and so on are generated by AI and then fed back into AI for more training, in the end barely read or consumed by anyone. Each of those events eats up a hell of a ton of electricity and earthly resources, which we know in fact are limited. GPUs in data centers, cooling systems, silicon chips, and whatnot - these are actually way more expensive than a normal user thinks, and that’s why we are starting to already pay the real, unsubsidized price of these cloud-based AI models.

I am not judging how people spend their time and what they do; I simply don’t care about all that. My point is that the current highly addictive and mindless usage cycle cannot scale further the way people assume it will. The sign is crystal clear, and the system is trying to outrun its own costs.

The Shape of What Comes Next

Perhaps if somehow, there is a magical, fairy-tale breakthrough in the future that allows chips to be cheaper, accessible, and less earth-destructive, then maybe there is light at the end of the tunnel. Each person may have their own local models running on these chips, solving this unsolvable problem. Only the most insanely complex work would be handed over to cloud-based AI models.

What This Means if You’re Writing Code for a Living

A lot of developers, over the last few years, have been doing something that superficially resembles programming, but is somewhat different. Paste some file or context, describe the issue loosely, accept the output, and move on. It works, prematurely, until you’re forced into a situation where you actually have to understand what the code does, and you hit the hard realization that you’ve been moving blazingly fast without building an underlying map of what you are actually doing.

As costs rise and providers start tiering access properly, this habit also becomes expensive in a direct sense. Using a large context window to fix something trivial will feel different once you’re paying for it by the token at its real price.

The skill that survives this is the old one: understanding the system underneath well enough to know when the AI answer is wrong. That’s real-world engineering, and most people in the computer field have long lost this art of engineering things. The last few years were unusual, and it led developers into thinking they could skip the comprehension part and get shit done. Well, you’re wrong, and that gap is closing day by day.

I am more than happy to see this happening in real life; this is going to be a hard pill to swallow for everyone who has been vibe-coding day and night without any fundamental knowledge about computers and calling themselves successful just by the sheer luck of winning the lottery at the slot machine.

---

“Technology is a useful servant but a dangerous master.” — Christian Lous Lange