Comment by supertroop
1 day ago
This is an example of a perspective that is solely lacking: I was a professional programmer for 40 years. I loved writing code in my youth. But then I started to love something more: making bigger things. I’ve grown out of the infatuation with my own self satisfaction with writing code. Now I’m addicted to architectures. There’s a point when you realize coding gets in the way and needs to be delegated. That’s why AI is appealing to me. I can put aside the wrench turning and focus on the bigger construct. It’s a game changer and it accelerates my ideas. I feel like I just got a second wind even though I’m close to retiring thanks to AI.
>Now I’m addicted to architectures
But architecture creation can also be delegated to ai if you give agent enough context about functional/non-functional requirements, certain business side restrictions etc. Your work is to be feedback loop mechanism between people and llm, no? Please, give me encouraging examples of technical tasks that cant be delegated to ai with accurate prompting
I have a feeling that a prompt accurate enough to cover all the important stuff (functional/non-functional requirements, certain business side restrictions etc) would be close in its volume and complexity to the actual software code and configuration (given the proper tools, frameworks and infrastructure in place)...
You might be thinking too near-term. If a sufficiently advanced coding agent has read all the language specs, all the code, all the software engineering/architecture books, and all of human knowledge along with it, why would it need any longer prompt than "create complete video streaming infrastructure to compete with Netflix"?
2 replies →
It won't be close. And result will be better. Anyway, this is different type of work
I guess I wasn’t very clear. At this point I haven’t found of level of abstraction where AI cannot be helpful. I have set it upon architecture problems but the main obstacle is the domain knowledge that it lacks. My company is working on uploading all of our internal specs as skills, but AI gets confused with poor English and outdated specs. Some groups have succeeded and the high level arch suggestions are being reviewed.
The challenge is guiding the model through decades of human slop: poorly written or missing specifications!
Ah, this sounds so unfun to me. As I thought, actual "thinking" is providing context due to bad, not self sufficient, documentation
I think I am just not result driven. I like to do stuff myself, not providing info to some super intelligence to do cognitive work
This role is also require way less qualifications and cognitive ability. You need local knowledge, not fundamental knowledge: why in this exact company things the way it is. Mostly not transferable
2 replies →