Can LLMs improve faster than the tech debt catches up to you. A lot of people/companies are betting on “yes”.
I honestly don’t know where I stand. I can’t, in good conscience, yolo code into my company’s code base. I use LLMs extensively but I’m not ready to abdicate my understanding and review of new code.
I have 100% vibed little prototypes or side projects for myself where I’ve never looked at the code. I’m just not ready to do that for my day job or my side business. But what was once a “I’d never do that” has shifted into “Is this possibly what the future will look like?”.
I’ve been playing with LLMs from the start and used tools like Aider before coding agent harnesses we really “a thing” (pre-Claude code/codex/opencode/pi). Initially I dismissed them as “really cool, but holy god it’s so much slop” but that’s starting to change (really started for me in Dec 2025) and I’m finding fewer and fewer mistakes in the LLM-generated code and becoming more and more impressed with it’s output.
I’m started to wonder if my need to feel in control of the code and understand it is actually still worth anything. I don’t understand assembly, I don’t like working in C, I already work I higher-level languages, is this the next step?
Yes, yes, there is a world of nuance between ASM->C vs C->LLM but what used to be a hard red line for me looks a lot blurrier today.
I like your way of thinking. Java meaningfully abstracts away memory management concepts. But would I be better off if did know memory management? Marginally. There were a few times when these concepts were useful.
I would call Java effectively “sealed” because 99.9% of the times you don’t need to go deeper into abstraction layer.
LLMs don’t have this property. You do need to understand some syntax and some low level design. So I wouldn’t call it “sealed”. That doesn’t mean it is not useful. You just need 10% or so of the understanding of Java syntax and other concepts.
The unresolved question is:
Can LLMs improve faster than the tech debt catches up to you. A lot of people/companies are betting on “yes”.
I honestly don’t know where I stand. I can’t, in good conscience, yolo code into my company’s code base. I use LLMs extensively but I’m not ready to abdicate my understanding and review of new code.
I have 100% vibed little prototypes or side projects for myself where I’ve never looked at the code. I’m just not ready to do that for my day job or my side business. But what was once a “I’d never do that” has shifted into “Is this possibly what the future will look like?”.
I’ve been playing with LLMs from the start and used tools like Aider before coding agent harnesses we really “a thing” (pre-Claude code/codex/opencode/pi). Initially I dismissed them as “really cool, but holy god it’s so much slop” but that’s starting to change (really started for me in Dec 2025) and I’m finding fewer and fewer mistakes in the LLM-generated code and becoming more and more impressed with it’s output.
I’m started to wonder if my need to feel in control of the code and understand it is actually still worth anything. I don’t understand assembly, I don’t like working in C, I already work I higher-level languages, is this the next step?
Yes, yes, there is a world of nuance between ASM->C vs C->LLM but what used to be a hard red line for me looks a lot blurrier today.
I like your way of thinking. Java meaningfully abstracts away memory management concepts. But would I be better off if did know memory management? Marginally. There were a few times when these concepts were useful.
I would call Java effectively “sealed” because 99.9% of the times you don’t need to go deeper into abstraction layer.
LLMs don’t have this property. You do need to understand some syntax and some low level design. So I wouldn’t call it “sealed”. That doesn’t mean it is not useful. You just need 10% or so of the understanding of Java syntax and other concepts.