Agreed 100%, especially with the sentiment that a lot of people have been telling themselves stories about the importance of quality in the industry (it was quite obvious, long before AI, that quality was not a concern unless marginaly in the rare cases where it mattered for competing. But quality for the sake of customers? No way, not now, not in the past, never.

But that's only valid in the industry though. Most of the software I'm using are free software, created by passionate people who used to care about quality because that's their name on the label, their reputation as an individual, their badge of honor, and also that's what makes us all feel good at the end of the day.

So, if I have few expectations from the quality of commercial software, I'm genuinely worried that open source authors would adopt AI generators too. Of course they will, I know I did. It's hard to resist. The speed gives you the fun you've lost writing good code your way.

Is code generation going to ruin the quality of open source software? I actually don't think so. I believe the benefits in terms of finding bugs (next postgres release will be the best when it lands!), conversions into better techs (finally, that old buggy python tool that's so convenient will be rewritten to rust or something), and we will have more new ideas, more people trying more things, being more ambitious etc.

Actually, it reminds me of when I transited from assembly to C, and even more from C to OCaml, for my personal projects: for a while, it was not fun! I was unhappy with the memory layout, with the GC, with the 31 bits integers, I could hear the bytes screaming at me: "come back and take care of us" and felt guilty. But, on the other hand, I was no longer typing kilometers of repetitive boilerplate, and had more time to think. Coding was no longuer typing at the keyboard, but more about thinking about designs and abstractions, and it felt both wrong to leave so much in the dark but also appealing to be able to think bigger.

And that's a bit of the same when driving an LLM: you have to spent much more time than usual designing, specifying, spelling out everything that matters; in one hand you feel that it would be faster to type the code yourself, but it's actually not true because you have an LLM to discuss the design with; whereas left to your own devices you would type and type for nights a design that you will revise several times over, rewriting most of everything several times. Now "coding" starts with a lengthy process of arguing with Claude until a good design is written down, recursively down to some good amount of details, until you can unleash the agents and voilà, you have it all done and working the next morning.

And there is some pleasure in that, too.

I'm worried for the users, for the workers, for the citizens, for myself, but not for the programmers part of us: machines will still be fun.