Just one more lane bro (source)
My personal experience with LLMs has been unbelievable. As it has with many other of my peers (according to my social media feeds and conversations, at least). I can learn things faster, ship working code faster, make decisions faster, process information faster. It sure feels like AI is saving me a whole lot of time.
In many ways it’s unprecedented. In other ways, it’s quite precedented.
Let’s take a step back for a second and look at our civilization from a broader view. Sober analysts of our energy situation will reference a paradoxical phenomenon regarding energy efficiency, called the Jevons paradox. It is the simple observation that relative gains in energy efficiency do not translate to absolute reductions in total energy use. For example, the Jevons paradox would be in full effect if someone replaced all of their incandescent light bulbs with more energy efficient LED ones, only to find that they end up with a higher electricity bill! At a collective level, this is exactly what we are doing as a global economy.
There are countless examples of these kinds of seemingly paradoxical situations, also deemed the “rebound effect”. From faster cars and more highways leading to more time spent commuting or household appliances leading to more time spent doing household chores.
If we are to treat AI as Normal Technology, then we should expect the rebound effect to follow course. Of course, not everyone thinks of AI as normal technology. Emile Torres has done a brilliant job of understanding the ideological drivers behind seeing AI as something more God-like, with their TESCREAL bundle framing — I highly recommend their work.
The rebound effect of AI's productivity gains might well be a net increase in the time spent working. Although the technology makes it easier and faster to build something, we’re likely to fill that extra time with more work rather than using it to live our lives more fully. It’s also unclear whether the extra time spent working will even be enjoyable:
Let’s also not forget about the process of enshittification of which we are only in the early days of with tools like Claude Code or Cursor. Eventually, these tools will move towards extracting immense value and won’t be as emancipatory as they currently seem.
If we want AI to truly save us time for our data labour activities, vibe coding won’t do the trick. Instead, it’s going to take collective bargaining efforts, as argues.
Spending less time working is as much a political and social effort as it is a technological one — if not more.