After months of coding with LLMs, I'm going back to using my brain

This rings true to me.

I still use LLMs heavily. However, I now follow two rules:

* Do not delegate any deep thought to them. For example, when thinking through a difficult design problem, I do it myself.

* Deeply review and modify any code they generate. I go through it line-by-line and edit it thoroughly. I have to do this because I find that much of what they generate is verbose, overly defensive, etc. I don't care if you can fix this through prompting; I take ownership over future maintainability.

"Vibe coding" (not caring about the generated code) gives me a bad feeling. The above approach leaves me with a good feeling. And, to repeat, I am still using them a lot and coding a lot faster because of it.

I also tried to go as hard as I could with agents and I ran into a ton of problems. The core issue I found is that inferring "intent" from existing code is hard and it makes errors that accumulate and build up. This ends up wrecking the software project by adding tons of complexity and duplicate ways of doing things and changing assumptions depending on which part of the code base you are in.

I think that the solution is to somehow move towards an "intent" layer that sits above the code.

I still use Cursor, but I use AI judiciously so that I don't wreck the project and it is only acts as an aid.

I don't get the whole "all-in" mentality around LLMs. I'm an iOS dev by trade, I continue to do that as I always have. The difference now is I'll use an LLM to quickly generate a one-off view based on a design. This isn't a core view of an app, the core functionality, or really anything of importance. It's a view that promotes a new feature, or how to install widgets, or random things. This would normally take me 30-60 min to implement depending on complexity, now it takes 5.

I also use it for building things like app landing pages. I hate web development, and LLMs are pretty good at it because I'd guess that is 90% of their training data related to software development. For that I make larger changes, review them manually, and commit them to git, like any other project. It's crazy to me that people will just go completely off the rails for multiple hours and run into a major issue, then just start over when instead you can use a measured approach and always continue forward momentum.

I tend to agree with this. These days I usually use LLMs to learn about something new or to help me generate client code for common APIs (especially boto3 these days). I tried Windsurf to help me make basic changes to my docker compose files, but when it couldn't even do that correctly, I lost a little enthusiasm. I'm sure it can build a working prototype of a small web app but that's not enough for me.

For me LLMs are a game changer for devops (API knowledge is way less important now that it's even been) but I'm still doing copy pasting from ChatGPT, however primitive it may seem.

Fundamentally I don't think it's a good idea to outsource your thinking to a bot unless it's truly better than you at long term decision making. If you're still the decision maker, then you probably want to make the final call as to what the interfaces should look like. I've definitely had good experiences carefully defining object oriented interfaces (eg for interfacing with AWS) and having LLMs fill in the implementation details but I'm not sure that's "vibe coding" per se.