The Death of the Manual Programmer
Categories: Thought
The title is borrowed from a rewritten version of a song by Omnipotent Youth Society that I loved more than a decade ago. The scene in that song is obviously different, but the feeling of watching a familiar world slowly get replaced feels strangely similar.
I had not taken a shuttle bus in years. The last time was on a trip to Prince Bay Park, when traffic restrictions meant I had to leave my car outside and take the free shuttle in.
There was a little girl in the seat ahead of me, scrolling through AI-generated short videos the whole ride. The images were rough, the stories felt fake, and one clip looked almost exactly like the next. What surprised me was how absorbed she was. The like counts were high too. Watching that scene made me unexpectedly sad. I kept wondering whether my own kids might grow up surrounded by this kind of synthetic, low-effort content, and slowly stop recognizing what is actually beautiful.
Since AI arrived, production has become easier across the board. Making content is easier. Building software is easier too. In the past, if you wanted to ship something decent, you usually had to spend a long time refining it, solving real problems, and earning enough confidence to put it in front of other people. Now a lot of that friction is gone. Writing something is easy. Building a product is easy. Buy some tokens, ask a model for help, wire together a workflow, drop in a template UI, and very quickly you have something that runs.
Today I saw someone say they could recreate Claude Code in two days. I believe it, and I do not. Over the past few months, dozens of voice AI products have appeared almost at once, and many of them are honestly not bad. Even Doubao has jumped into the space. I have also seen quite a few wrapper clients around Claude Code lately, and some of them are genuinely well made.
A lot of things that used to require technical depth, a real learning curve, and years of accumulation are turning into commodities. In the future, what we will not lack is software that looks like a product: usable, runnable, polished enough. You can still create value by making it faster, smoother, or packaged a little better, but AI is catching up to that layer very quickly.
I was talking to a colleague over dinner recently about something related. Over the past year I have become strangely obsessed with cassette players. Songs sound unusually durable on them. Why do old cassettes, CDs, TV shows, and even many older books often feel more carefully made? The answer is simple: production and distribution used to be heavy. If you wanted to release an album, you had to make something good enough to justify pressing tens of thousands of tapes. Otherwise it would not sell, and the company might never back you again. Publishing a book worked the same way. You could not just write something at night and push it in front of everyone the next morning. The act of making and distributing the thing was already a filter.
Now you upload a song to a platform. You post an essay to WeChat. Software is moving in the same direction after AI. AI can even push code to GitHub for people who used to find that intimidating, and set up release CI that they never would have written themselves. A lot of the hurdles that once demanded long-term accumulation have been flattened by tools. The result is predictable: the world fills up with things that are similar, passable, and easy to mistake for quality.
The problem is not only that average quality drops. Over time, our sense of quality drops with it. When rough work becomes the majority and gets distributed everywhere, especially when money rewards speed and volume, people’s judgment shifts. What they get used to is fast stimulation, fast feedback, and fast satisfaction.
That is what bothered me about the girl on the shuttle. She was not just watching a few low-quality videos. She may be growing up inside a stream of content that is cheaper, more frequent, and emptier than what came before.
Software development has clearly entered the same phase. Soon, complete beginners will be able to use AI to build products that solve their own problems. Product managers will be able to make things that previously required engineers to join the effort. So what will still belong to real engineers in that world? That question deserves serious thought.
Recently I have heard that quite a few big tech executives have become obsessed with vibe coding too. In a single afternoon they can build a demo they believe is usable, and some of them get deeply hooked. This could have a real effect on people doing the actual engineering work. Once a boss gets one happy-path demo running, it is easy for them to think software is not that hard after all. If a project used to take six months, should it now take one? If it used to need one hundred people, can ten do it? The implications are hard to ignore.
Engineers will still have room to build better products and more efficient systems. But if we stay only at that layer, competition is going to get much tighter. More people will enter. More people will be able to make something that looks convincing. That part of the market will get crowded fast.
What seems more interesting to me is the way musicians and actors escaped pure reproduction pressure. They still release albums, but they also create concerts, theater, and live performance, things you cannot replace just by wrapping the same core in a slightly different shell. Those forms contain organization, density of detail, and a sense of completion that only appears after long refinement. They face the world directly.
I think software will increasingly work like that too. Everyone will be able to vibe code a product. Everyone will be able to build something that mostly works. What will create real distance between teams is still systems thinking, engineering depth, domain understanding, and all the things that are hard to see at a glance but ultimately decide whether a product has real weight.
The faster the outside world moves, the less willing we should be to lower our own standards. Cheap supply will keep increasing. That does not mean we should become rough with it. The products that usually last are still the ones that feel effortless, calm, restrained, and reliable, the kind where you can tell the builder treated the work seriously.
Maybe the next thing worth doing for me lives in another dimension: products that combine hardware and software, platform-level products that previously only a giant company with thousands of people could build, or something that breaks out of the current frame entirely. I do not know yet. That part still needs more thought.
When everything around you starts to look alike and compete on the same narrow axis, moving outward may be the only way forward. Face a larger market, different users, and harder requirements. In that environment, you cannot survive by packaging faster or grabbing timing alone. It forces you to make things more solid, and it forces you to figure out what you actually want to build.
AI has made many things easier. Precisely because it is easier, it has become harder to know what is actually worth doing, and what is worth spending years of your life on. Deciding what to build may matter much more than figuring out how to build something quickly.