I'm really struggling to see how to make architectural decisions with an agent. It's great when you're at a total loss for ideas, but when you already have some of the pieces it ultimately wants to drive all of the thinking and takes over. Then it just feels like you're deferring your experienced judgement. I've seen colleagues lose the ability to reason any more without asking the agent to do it for them, because they inherently don't see the point if the agent is going to end up doing the whole piece (and probably auditing/overruling anything they came up with on their own).
I tend to find this part of the work enjoyable. It is just a faster and more productive version of what I would do with any engineer working for me who owns a large feature. I barely have to hint at my concern or drop the right keyword and the agent (Fable/Astra) will immediately understand.
"Couldn't this be stateless?" "Do you have a plan to be able to shard this?" - We will almost always pivot from the agents initial design but the agent is able to easily understand the reasons and benefits and align quickly.
If I was writing the code myself, I'd often have to make compromises between the ideal architecture and the level of effort required to implement it - now I can just always have the ideal architecture.
> I can just always have the ideal architecture.
Now you have what you think is the ideal architecture. Since you didn't implement it, you didn't discover it was not the ideal one mid way into the implementation.
May be it is too complicated, but you wouldn't know, because LLM is doing the implementation. If you did implement it yourself, you might have spotted a critical point that might simplify the whole thing...
There is a difference between LLM-generated code that somebody merged essentially unseen and LLM-generated code which a human then looked at carefully and massaged until they were happy with it.
The first version can hide a lot of crap. All you see is the end-to-end functionality and there can be corner cases or scenarios which you have not tried and in which it is terrible or plain incorrect.
The second version hides just as many issues as human written code.
Unfortunately, the debate that our field has about LLM-generated code mixes both quite freely even though they are quite distinct.
trwhite · · focus · HN ↗
stefangordon · · focus · HN ↗
"Couldn't this be stateless?" "Do you have a plan to be able to shard this?" - We will almost always pivot from the agents initial design but the agent is able to easily understand the reasons and benefits and align quickly.
If I was writing the code myself, I'd often have to make compromises between the ideal architecture and the level of effort required to implement it - now I can just always have the ideal architecture.
qsera · · focus · HN ↗
Now you have what you think is the ideal architecture. Since you didn't implement it, you didn't discover it was not the ideal one mid way into the implementation.
May be it is too complicated, but you wouldn't know, because LLM is doing the implementation. If you did implement it yourself, you might have spotted a critical point that might simplify the whole thing...
themgt · · focus · HN ↗
This is the "Jodie Foster in Contact listening for the SETI signal with headphones" theory of how production software systems work.
dolni · · focus · HN ↗
But then, one small detail changed everything. And it wasn't something I'd have thought of until I was writing the code.
I shudder thinking about the number of these issues hiding in LLM-generated code.
teiferer · · focus · HN ↗
There is a difference between LLM-generated code that somebody merged essentially unseen and LLM-generated code which a human then looked at carefully and massaged until they were happy with it.
The first version can hide a lot of crap. All you see is the end-to-end functionality and there can be corner cases or scenarios which you have not tried and in which it is terrible or plain incorrect.
The second version hides just as many issues as human written code.
Unfortunately, the debate that our field has about LLM-generated code mixes both quite freely even though they are quite distinct.