Comment by bdcravens
1 day ago
> A lot of code is being written and a lot of it is going to be a real future burden.
This assumes that the models of the future won't find it easier to just throw the code away and rebuild it
This also assumes that the same application build by humans wouldn't become a "spaghetti mess".
Reminds me of Niven/Pournelle's "The Mote in God’s Eye" where the aliens have a very pragmatic/Jury-rigged approach to technology and everything is more or less improvised.
I have noticed this with co-workers also, when you have the ability to read/write/understand things very quickly, you tend to spend much less time on making things tidy, clear and maintainable.
But then you end up in a state, 2 years down the line, where every single person that works in that code base haaaates it and adding any new functionality takes weeks or months.
Compare that to clean, organized code where patterns and extensibility are established. I’ve seen product manager surprised at how fast a new feature was added when everything is laid out well.
Two years down the line I suspect "persons" won't working directly on code bases they hate. Two years from now, I'd wager that even today's most ardent AI critics will be using them for dirty jobs -- like refactoring ugly, but useful, balls of mud.
3 replies →
This is currently my workflow lol. I will build features and functions rapidly then extract and rebuild into smaller more focused pieces of software. I think of it like oil painting. I’m able to paint over, scrape off and repaint relevant sections without being burdened by the complexity of what has already been painted.
You use an AI to make the spaghetti.
You use another AI to document the spaghetti.
You use the third AI to make a spicy meatball from the spec.
Interestingly, I have personally felt this way about projects in which I was involved in the past(written both by me and by others)…
sometimes starting from scratch just faster and/or easier.
That reasoning is why I worry about the AI endgame ;-)
Man … I never think about or am concerned about the the terminator apocalypse. But what a parallel
1 reply →
>just throw the code away and rebuild it
What about all the undocumented "adjustments" ("bug fixes" in a professional context) that were made to make it actually useful?
Well first it could write a spec, THEN throw it away :) Like you are supposed to do with a prototype.
Well it could miss relevant things to include in the spec.
6 replies →
> This also assumes that the same application build by humans wouldn't become a "spaghetti mess".
nah, it reflects on how applications built by humans usually do become spaghetti messes with all the resulting brittleness and unintended negative side effects of changes that result
but it probably doesn't matter for a little toy piano app
Have better agents rewrite it in the future.
And the models of the future will surely be better than they are today. This stuff is still a very long way from maturity.
Sometimes it seems like they're moving very slowly. That makes sense: It's easy to get used to how they work today and it is also easy to forget how much worse they were last year.
When we look back and realize that just 4 years ago these tools didn't really exist at all, it becomes clear that the rate of progress is rather amazing.
In 4 years, we've gone from "hah, good luck with that crap!" to "little kids writing music-learning games on their own in a few minutes"
That's pretty friggin' awesome, and it's not finished yet. :)
> And the models of the future will surely be better than they are today. This stuff is still a very long way from maturity.
Since this is Claude, the models of the future might be more expensive, or more locked down, or might decide your 8yo is actually trying to build a cleverly disguised bomb so her request gets silently downgraded to a dumber model, etc.
People tend to not realize how far these models have become.
I remember when they would always hallucinate APIs that wasn't there or make up fields that didn't exist.. those problems are virtually solved now.
So with that in mind, why wouldn't AI be able to write better code?
The code would have to be maintainable by AI itself (operating based on the assumption that the future will be Agentic Engineering)
9 replies →
Models in general have gotten better, not specifically Claude, is what they mean; 4 years ago was when ChatGPT without any Anthropic was released.
There needs to be studies of people like this.
Same when one points out that yes, quality of life is in general much better now than 20 years ago even.
Someone like you will just pull out some stat "well THIS is worse!! I cant buy a house at 22!! The west is doomed and I LOVE China!"
2 replies →
I think that sort of careless waste is a very human behavioral pattern, but then we are the ones training these bots.
ed sp
errors compound
Just like in language learning?
...Whether you are an LLM or human. And, on average, LLMs write better code than humans.
Not you, of course. You write exceptional code with zero errors that never needs rework. I'm talking about the rest of us.
they write code faster too, which means things compound faster.
i write pretty good code when i need or want to, yeah. sadly LLM code gives me a headache so the current psychosis standard has made work a bit painful.
the codebases are becoming LLM-first human-second so even if i wanted to be natty it's not really viable. might as well spend my employer's money.
Eh, not really. It's the same issue with porting. You always rely on features you didn't properly articulate.
Good luck to future models figuring out which weird section of code are bugs and which are features.
AI writes faster, so the rate it incurs tech debt is proportionally higher. However it's ability to have large context kept on memory compared to humans is also a key component fighting against it. These are occasionally forgotten when code quality of AI for large code bases are discussed. So Yes I think humans also build spaghetti, but as they write slower they get to the same place a lot later. However humans can't correct it, or can't correct it fast enough. AI can.
It does somewhat depend on the application size. Seems to me that for regular software projects (that aren't enterprise SaaS) a good programmer will create better software than Claude. Maybe the dehumanizing way to say it is that humans have more efficient/improved retrieval. The amount of time I see repeat code for no reason, or code/context that has been obviously missed is absurd.
I agree with this. Clean code is important today because it helps humans to understand it. Cleanliness is not otherwise intrinsically valuable.
Today I don't always need to understand the code. When I do, I can ask the LLM to explain the code or to specify the algorithm in pseudocode or even re-write the code until I can understand it.
I think, as an industry, we haven't yet fully grasped the impact of agentic coding. We don't yet know what software engineering is going to be like in 5 to 10 years.