The moving tiny rectangles framing is interesting, it gets to the heart of why I find all the anti-AI takes so difficult to comprehend. If you never made any effort to connect what you do with what value is added in real life, then it's no wonder better tooling is leaving you lost. Programming (other than code golf) has always been an implementation detail for solving problems IRL.
The programming itself is the reward for people who love doing it. It attracts the sort of detail-oriented thinkers who enjoy the doing and don't frame everything in terms of "value added."
AI is attractive to the sorts of people who have their secretary write their Christmas cards.
I think there's a middle ground. I like coming up with solutions to problems (mostly technical problems, may even be very low level ones). But I always found writing the code generally tedious. Basically, once I had a good detailed idea of what the implementation would look like, actually executing the plan would bore me.
AI is still not competent enough to come up with good solutions in many things I work on. So, at least so far, AI has made me happier.
AI helps you to focus on the aspects that you are interested in. Perhaps you care about database nifty stuff, but you may need the front end to make an end to end solution. You can delegate that part to AI
Not disagreeing, but apart from coding the most boring part of the entire job is the one that's the hardest to correctly delegate to AI: testing and ensuring the whole thing is sound.
Which is why it's changing the calculus on junior devs: if you're not mature enough to do self-review and self-QA, you're just dead weight for the team/company.
You don’t, there’s always a choice about what to include and what to exclude in your system.
Unless you are hand selecting every atom that goes into a thing (maybe you make nuclear weapons?), you always make choices about what you focus on and what is irrelevant to your project.
Carpentry has always been an implementation detail for making furniture. They have been able to purchase flat-pack chairs for all of their lives, but for some reason there are people who learn this skill and have fun slowly making things that factories already make at scale. A subset of those people have made lucrative businesses out of the very human craft that is carpentry, and are able to create custom pieces on-demand that you could never justify retooling a factory to create.
It is okay to view code as a means to an end. I disagree, preferring to treat code as craft, and striving for better systems that are easy to understand, maintain and extend. And I think that's the source of our disconnect; deeper than one's opinion about AI is one's value of human skill and the effect that has on the output. Maybe I overvalue it, and maybe creating code "manually" is going to look more like carpentry in the future; but you cannot expect to convince a skilled carpenter that an IKEA chair is just as good and accomplishes the same task.
b) There's a clear problem being solved (you need furniture).
Stretching your analogy to fit my point: pretend that programming is manually sanding wood, while AI-assisted programming is using a belt sander. If you're focused on the chair being built, getting a belt sander to help is great! If you're sanding for the craft (?) of it, focused on the wrist mechanics of rubbing sandpaper up and down, you'd be disappointed.
Maybe the disconnect, then, is that you don't consider programming to take place in the real world. I don't share that experience, any more than someone coding batch jobs on punch cards or an author committing chapters via typewriter. Maybe the medium has parts that only have meaning within our minds, but we are real (dammit!) and we punch out those characters with our real hands and we get frustrated and deal with impedance mismatches and want to scream and laugh and cry. There is a piece of us that lives in what we build. It's evidence that we are real.
This ties in with your second point. There are uncountably many ways to accomplish the goal of making a chair or writing a program. And if you are a carpenter working on a one-off matched dining set for a fickle client, the problem might not be as clear as even many software tasks are. Your skill and experience is highly likely to play into the eventual form and structure of the finished work. The customer might not know where you hid the dovetail joints or dominoes, but they can absolutely notice the grain continuity and lack of obvious engineered joinery evident in a factory piece.
If you don't care, then fine! You can focus on the other things that bring you joy. But I hope you can appreciate that some of us want to experience solving these problems with a bicycle for the mind instead of a Waymo.
There’s a third group of people: people who did it, got good at it, and got bored.
I do both carpentry and programming and both activities have long since become repetitive. There are only so many dovetails or distributed systems you can make.
That’s why I don’t care if AI can replace those parts. I’m in it to do the designing, not the crafting.
That would assume that the AI is as good at the repetitive parts as you were, which isn’t my experience. You still have to review and correct it, which is more boring and repetitive than implementing it yourself.
But I also disagree about the getting bored on the “crafting”. It may depend on what you do, but there are always new design decisions and trade-offs to make all the way down. This isn’t a solved problem, and AI doesn’t change that.
> pretend that programming is manually sanding wood, while AI-assisted programming is using a belt sander.
That analogy falls flat, because there is little creative difference between these two modes of sanding. In particular, there is approximately zero variation in what the belt sander does as a function of how you control it. It is a reliable, deterministic, very predictable tool. That’s as different from generative AI as a compiler is.
It easy to convince a college student with $20 that an ikea chair is good. Artisanal is overkill for plenty of scenarios, and definitely those where time or money are constrained.
If you see programming purely as a means to an end, then yeah, I get this perspective. But to many there is enjoyment in the _doing_ and the craft of it beyond the end result. It’s why people get into woodworking or knitting despite the fact that it’s much cheaper, faster, and easier to buy a table or a sweater than to make one yourself. Value is subjective, and for some the value of code is not primarily in what you can sell to others.
I think all it means when we say 'solve problems in real life' is just the stuff you have to do that tooling can't abstract you away from any more.
The sharp end of the debate now is around what exactly that means in the LLM world. It's extremely unclear what exactly the new level of abstraction unlocked is, or at least how general/leaky it is.
There's obviously just the stance of enjoying the craft, and that's one thing off to the side, but I think the major source of conflict for those who are more oriented towards living in the top level of abstraction (i.e. what you can do in real life) is between some of the claims being pushed about said level of abstraction and what many still experience in actual reality using these tools.
If you've never made any effort to connect what you do to the underlying mathematics, then no wonder you think it's all an "automatible" implementation detail, despite three decades of the industry trying and failing.
