Crazy how many engineers in here just say they are using another prompt on top. From my experience that makes things worse. It does abstractions, but the wrong ones. It overcomments, confusing future calls of the LLM.
To me building on multiple scalable systems this has been the most dangerous part of LLMs. On a good codebase it will work good, but it will maek it worse, so you keep using it, till it doesnt work and then you have to pay the bill and fix for what you didn’t build before.
If you put an agent on a fresh codebase 2 things are often given:
-> You have a mental model of the code
-> The code is somewhet concise
After multiple iterations both is lost and LLM performance degrades. To solve this you can regular refactor, but it’s not a nice experienc. So my best solution is:
I use LLMs for exploration and for review, but I write the code myself. I find it hard to believe why so many engineers try to avoid it. It’s not consuming much of my time. And it’s actually the most enjoyable part.
Sometimes I race AI i give it a prompt /bug to fix and at the sametime im greping/symboling through the codebase and tryto fix it myself. AI isn’t always faster.
The issue is that whilst the loops will initially lead to good results they will be less and less as context gets bigger and bigger and tougher to understand for human and AI.
Whilst I understand cases in which duplication is preffered. I generally think abstractions are underused. Sometimes I would abstract something away that is only done once, not because i want to have less code, but because it allows me to solve bigger problems and when i look at a function I don’t have to worry about it. It allows to create systems. Obviously your abstractions should be good
IMHO most founder will fail because leaning to heavy into ai and creating a system where they never experience the friction necessary to build the domain knowledge which ultimately could be the deciding factor.
Just think about website design, I don’t think it’s far-fetched to say that a non ai design website will outperform an ai designed one. These percentages add up in multiple disciplines.
I would argue betting against ai is your best chance of succeeding frankly (not in all cases but certainly as displayed here)
I don’t think HN is the right place to post this and also I don’t think the title/slogin as fitting. This might be helpful to beginners, but nobody who’s scared of skill rot, will use this on a mobile app.
Honestly I think there is no benefit in any mobile coding app at all. It makes coding seem way more tedious than it is. I would rather recommend a book, laptop or youtube video than to do coding tasks on a phone keyboard.
This is very interesting and the biggest glimpse of hope I’ve seen the last couple months here. I haven’t really paid attention to Fusion even though I got the email. I didn’t even assume it would be comparable.
I wonder if this makes the world a fairer place actually. Because it allows people that are attentive and have time to dig through this to save money. The lowest price would be higher without these patterns.
It’s a polarizing idea, but frankly it’s what the world moves too and seem to work on the market. Some people are lazy or don’t have time and pay more money and some people have less time and dig through dark patterns, collect coupons or utilize ramp up subsidizing.
All those people that actually "work" through it will have less of a prime as if these patterns wouldn’t exist.
The question still remains would the world be a better place without these patterns,as it wastes time and acts against user intent.
It’s just a fascinating question to me, because a lot of things are not as simple as they seem of the first glance.
To me building on multiple scalable systems this has been the most dangerous part of LLMs. On a good codebase it will work good, but it will maek it worse, so you keep using it, till it doesnt work and then you have to pay the bill and fix for what you didn’t build before.
If you put an agent on a fresh codebase 2 things are often given:
-> You have a mental model of the code -> The code is somewhet concise
After multiple iterations both is lost and LLM performance degrades. To solve this you can regular refactor, but it’s not a nice experienc. So my best solution is:
I use LLMs for exploration and for review, but I write the code myself. I find it hard to believe why so many engineers try to avoid it. It’s not consuming much of my time. And it’s actually the most enjoyable part.
Sometimes I race AI i give it a prompt /bug to fix and at the sametime im greping/symboling through the codebase and tryto fix it myself. AI isn’t always faster.