From Today, Software Engineering Is Dead

I don't get it. Writing code yourself is the best way to control how the code functions and is designed. If LLMs were to disappear tomorrow, how would one lose control over the degree and specificity of the code produced that one could not simply compensate for with a bit more time and skill investment up front?

The author only hints at the notion that developers who do not use LLMs are like painters upon the advent of photography, but does not go on to suggest who these developers analogous to painters might be today, nor to whom their works might be valued.

If we're going to torture visual media into being a good analogy for LLM versus hand coding, I think writing code oneself is more like having a inkjet printer that one controls manually. You get a great deal of explicit control and theoretically unlimited expressiveness but at the cost of a greater time investment. Comparatively, painting might not give you that same degree of precision. Photography surrenders that control but puts a great significance on capturing fleeting moments in time.

I wonder why people think that is a good argument. Spending weeks on a single design decision makes the case that you’re either not a good developer, your time was stolen by other concerns, or that design decision was a major process change. In the last case, this is one of the goals of software engineering to make those less painful.

Anyone using this argunent should add anecdotes so that others can appreciate the previous scope of work.

No it didn't. You can go find wildly talented painters today who work in incredibly realist styles…in fact just a few years ago as visual social media really took off, there were many challenges to "fool" people into thinking they were looking at a photograph when it was actually a painting.

The whole "painters didn't have to paint a realist style anymore because of cameras" is an ad-hoc rationalization. That was entirely due to taste and other cultural factors, nothing really technological about it whatsoever. The only claim you could convincingly make is that painters _who made a living painting portraits for random people who hired them to paint their portrait_ had to pivot because now people could hang a cheaper photograph of themselves and their family on the wall instead of the more expensive painted portrait. That's about it really.

Handcoding great software for people who want it is alive and well. It's just not at all the industry "software engineers" who can't see past what Google or Meta like to shove out of their corporate offices have been working in from the jump.

I'm not a senior engineer and I trust those that tell me LLMs are currently incapable of high level coherent engineering. But at the same time, this is a rapidly developing tech whose progress only seems to be accelerating, and as long as it's gaining on every area of coding and planning, anyone arguing AI will never be equal or superior to human SWEs is wish thinking.

We've lived through all this before with the Go community, who used to be adamant a machine could never do better than a human - because of some ethereal semi-magical quality only existing in the skulls of bipeds and totally inaccessible to machines. SWE's Alphago Zero moment could be in the post.

AI has created more situations where an algorithmic twist or a really clever implementation makes a significant difference. But those situations are still rare compared to everyday work in software development.

If you can slop it out with a coding agent it's probably not the kind of thing that needed an engineer in the first place, just someone who can write the code. Which is not the same thing.

You certainly can do software engineering with LLMs (and in some cases it supercharges it). But then you're doing actual software engineering anyway, with or without the LLM.

The only thing that's changed is being able to call yourself a software engineer when you weren't any such thing in the first place. I agree it's harder to do that now.