Table of Contents
The Missing Tsunami
This year, LLMs finally entered usable territory. As I wrote in the code review post, DeepSeek V4.1 Flash crossed the Rubicon of actual, indisputable usefulness on real tasks.
And yet, look around. Where is the software getting better? Where are the people successfully monetizing their expensive subscriptions? Where are grand new ideas being turned into cool products and stable revenue streams?
What I mostly see is a mountain of vibe-coded software that is, at best, personal. The trillion-dollar question is why. Many were bracing for a Cambrian explosion and got a pile of half-finished side projects instead.
Even professional, mass-market software has gotten worse. Uptime is down, and the embarrassing bugs are more frequent and more public than they used to be. We can probably attribute that to growing pains, but the current state of software is a shit show.
Skill Was Never the Blocker
The usual narrative is that skill was the scary blocker preventing good people with bright ideas from being successful, and that removing the skill barrier would democratize the industry.
Needless to say, it’s hard to imagine a more open and meritocratic industry than software, but apparently this “peasants’ rebellion” fairy tale was a popular marketing trope for LLM hucksters.
Skill was never the actual blocker. Skill is a derivative of giving a fuck. It is what you acquire on the way to doing something you genuinely care about and want to bring to this world. Spend long enough thinking about something and you cannot help becoming an expert at it. Expertise never had gatekeepers, the gates were wide open from day one.
Nobody needed a computer science degree to build something they were dreaming about. They needed the obsession, and the skill came along for the ride. The people who couldn’t code were, with rare exceptions, the people who didn’t care enough to learn.
Good Ideas Are Already Taken
Most good ideas were implemented a very long time ago. The obvious mass-market products exist. They exist because someone else already cared enough to build them, and cared enough to keep building them after the first prototype.
Now we have an army of lazy, egocentric people, bamboozed by marketing overpromises, running around in circles, convinced they are sitting on the banger idea of the decade. Billions of tokens wasted on fan fiction about how their app will change everything. Then they ship it and learn the boring truth: other people see a barely working, unintuitive, ugly pile of slop.
That gap between how it feels from the inside and how it lands from the outside is the crucial part. Building a personal tool does not require you to model another human being. The self-centered approach is fine for a tool only you will ever use.
It is fatal for a commercial or even an open-source product. To make something other people like and use, you have to be honest about the fact that they are not you, and that they also have their own goals, workflows and preferences. The zero clue “just make it work, I don’t care how”, self-centered approach can never produce that. It produces slop that the author finds brilliant and everyone else abandons after one look.
Who Doesn’t Want to Be a Plumber
So how do you cross the gap? The prosaic reality is that producing marketable software is no fun.
It is about brutal efficiency and polish rather than feature bloat. It is empathy, the dull discipline of imagining what a stranger needs and would buy instead of what you find interesting. It is product training; it is heavy marketing, shouting into a void that does not care, over and over, until you hear back and adapt to feedback.
The code was always the easiest part, a means to an end, so even your choice of programming language is usually performative. We can safely say the code is significantly cheaper to write and maintain now, and yet the gap is still here. The real gap lies in giving a fuck about the person on the other end.
Conclusion
LLMs are the best thing to happen to my own workflow since autocomplete. But “lines of code per week” was never the bottleneck, and producing more has clearly diminishing returns.
You still have to obsess until you improve your understanding, and until things click and start to feel right. And then you still have to do all the unglamorous, unbillable, unautomatable work that turns a thing you find cool into a product other people actually want.
Skill is a derivative of giving a fuck, and so is everything worth using. I saw someone on Twitter define AGI as “Actually Generating Income.” A good framing, and a good reminder. So far, no amount of tokens can generate that ex nihilo.
One more thing. Everything above is about software, the least interesting part of the story. The aspects of LLMs I actually find revolutionary and exciting have nothing to do with computers and everything to do with politics and actual power. That will be the subject of one of my future posts.