Vibe-coding your own desktop environment: https://isene.org/2026/05/Audience-of-One.html As this toot points out, however, this personalization comes at the price of making oneself dependent on a coding agent.

the alternative being not being able to?

You can eject, but then you are stuck with the current state of your tools forever. Which in practice means until it is broken by an earthquake in its infrastructure, typically after a year or so. A settings dialog is very different from a coding agent. Most programmers would be able to write their own, it's just not worth the effort if you can use a ready-made one. The degree of dependency is very different.

then why people don't use emacs, smalltalk more? in fact most people never did, not even programmers

I pretty much stick to default, less bugs that way. I use claude code to fix issues with my mac though, which is pretty cool. I think "coding agent" is selling it short. Apps with internal reasoning, thats a true expressivity game changer over a config or setting panel. It can configure itself to things unanticipated by the developer, and it can creatively combine posix tools towards a goal better than I can. I also do not like being dependent on a huge service. But it makese sense, the computers are really big and expensive and can only justify their cost if running continuously. They are not self-hosting sized, unfortunately, even if the software was available. I think I can take the LLM dependency if 1. you can get the LLM from a few different providers, that helps with a downward pressure on the market to be competitive. 2. the LLM is for the control plane i.e. the software works without an LLM (and offline in general), but I accept reconfiguring it might be difficult by hand (your losing the reasoning part) My productivity has sky rocketed with LLMs so I can't really see them going away or not being integral to development and/or operations now.

As someone who only ever bothers with Nvim distros, defaults to fish shell and generally rarely checks the man doc let alone the config's mandoc, as someone who doesn't really do programming recreationally, I had a rather few eye opening moments this year where the 5 minutes I spent looking at config manuals made incredible impact on the comfort of my computing environment. These range from undervolting a laptop to learning a simple tmux cmd. The reason I'm sharing is that these moments made me realize that as precious those 5 minutes are, they can actually be very well spent at customization. Reading manuals, I tend to associate that with nights wasted distro hopping and desktop ricing of my youth but those are but extremes and swinging back to the opposite extreme only had me settling for a sharp and hard defaults. We're all so different after all. Anecdote aside, one thing that really helped me was having a better understanding of what made me uncomfortable. I maintained a todo list for personal computing maintaince tasks (which I usually tackle when procrastinating something else). I used atuin to analyze what commands I used most. I used LLMs too in some cases, they're a big help when it comes to sanding edges. Categorically, I (and everyone here) have always been and will always be a statical outlier when it comes to customizing their computer. That being said, the virtue there is really real me thinks. Defaults are overrated.

@Mariano Guerra

then why people don't use emacs, smalltalk more? in fact most people never did, not even programmers

Some do, some don't. As you say, there was probably never a time when most people did, and yet, these systems have survived for decades because they remain important for enough people to keep them alive. People have different priorities, and that's fine with me. What I think everyone should do, however, is explore the pros and cons of their tech choices, short term and long term with the latter involving some form of scenario planning. Just to avoid bad surprises. In particular if your tech choices impact others as well. In short, do what @Tom Larkworthy just did above: explore the trade-offs.

energy usage vs usefulness and its application and funding by the military could be used against computing itself in the early days, they got cheaper, more efficient, personal.

1. Anything that is in the world when you’re born is normal and ordinary and is just a natural part of the way the world works.

2. Anything that's invented between when you’re fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it.

3. Anything invented after you're thirty-five is against the natural order of things.

Indeed, and I expect LLMs will follow this trend as well, assuming they don't crash the economy or the biosphere before (that was a problem that early computing did not have). And I'll be happy to revise my LLM attitude if and when I see serious work towards reducing their negative impact. So far I see mostly denial. There is obviously a lot of a margin in the energy consumption of computation, as proven by biological brains. Running an LLM on a digital substrate is enormous waste. Digital technology is at one extreme end of a trade-off between energy efficiency and precision. Our hardware has an incredible level of long-term determinism which is mostly wasted on LLMs. But I don't see anybody investing serious efforts on doing better so far. Your three stages of normality are a caricature which is overall true, but it's besides the point. There are many technologies which are normal for pretty much everyone living today, and yet harmful for the long-term survival of our societies. Air travel, for example. LLMs may well become normal in the same sense in a few years. That will only accelerate the demise of our societies.

re: But I don't see anybody investing serious efforts on doing better so far https://taalas.com/products/

I agree it's not likely to happen, but it would be a great path to take! Taalas looks interesting, that's where LLMs should. But also for training. And it's only the first step, I am sure we could do a lot better than that, though not with the goal of running existing LLMs on new hardware.

Thanks for sharing @Konrad Hinsen.

Could we say that coding agents potentially adjust some of the stickiness of existing tools? You feel less barrier to entry with trying something new?

Yep, @wtaysom, I feel like they amp stickiness for good or for ill.

Interesting. Prompting for shims is one pleasant way to use the an agent because just getting a thing working tends to be such a tedious trial-and-error task. The reasonable consequence is that with everyone personally sanding rough edges, the edges get rougher over time. However, I've tended to see a different dynamic: when adoption costs for some framework are high (requires a different language runtime for example) an agent can be quite effective at mining it for the desired parts. The reasonable consequence is that frameworks potentially being less well developed because its easier to take bits without adopting it and entering into the culture around it.

Great examples. Which can be seen as special cases of the general phenomenon that I expect to happen: less collaboration and consensus-seeking between humans, replaced by more dependence on LLM-mediated services for working around tech incompatibilities (and human ones - think of machine translation). This might well be an acceptable trade-off in a world in which LLMs are economically accessible to everyone and a diverse LLM market ensures that no single party can inject biases into LLM-based mediation. Which is unfortunately not our current world.