Don't have it do things for you. Have it do things with you.

It can't do things with me like a human, it's not human, it's not intelligent, it's not thinking, it's not aware. It's an aide I use, not a tool I rely on.

I have had things like your React instead of Vue problem. I solved it by always having Claude write a full implementation spec/plan in markdown which I give to a fresh context Claude to implement. Typically, I have comments and make it revise until I am happy.

It has basically eliminated surprises like that.

Back then, also around May, I had Claude 3.old destroy a working app. Those were sad old days.

Hasn't happened in a long time. Opus 4.6 is a miracle improvement.

An advantage I have enjoyed is that I am insanely careful about my fundamental architecture and I have a project scaffold that works correctly.

It has examples of all the parts of a web app written, over many years, to be my own ideal structure. When the LLM era arrived, I added a ton of comments explaining what, why and how.

It turns out to serves as a sort of seed crystal for decent code. Though, if I do not remind it to mimic that architecture, it sometimes doesn't and that's very weird.

Still, that's a tip I suggest. Give it examples of good code that are commented to explain why its good.

I would love it if someone explained what their ten agents Ralphing away were actually told to do.

I suppose if you are doing something that truly can be decided based on a test but, I just don't see it, at least for anything I do.

even if it can be decided on a test, there is the point of "it works, but the code is shit". For instance, I'm doing a website with some e-commerce. Upon executing a sale, I connect to the accounting software to create an invoice, and then send a confirmation email to the customer attaching the invoice. Claude Code decided to put all the functionality to connect to the accounting software inside the `sendConfirmationEmail` function. The code was placed in the right order so it "worked", but hiding invoice creation inside a function named after sending confirmation email is, IMO, bad code.

Some people might say that, with AI, having well structured code no longer matters because no human is expected to read it any more, but I don't think we're there yet.

I think ralphing is for purely vibe coded stuff, where you're literally never looking at the code and only asking for changes to the final output.

If I'm reviewing all the code, so far I'm still the bottleneck even with a single agent and I don't see an easy way to change that.

YES YES YES!! I so wish that we could go back in time and never, ever have even suggested anything other that what you say here. AI doesn't do it for you. It does it with you.

You have to figure out what you want before the AI codes. The thinking BEFORE is the entire game.

Though I will also say that I use Claude for working out designs a lot. Literally hours sometimes with long periods of me thinking it through.

And I still get a ton more done and often use tech that I would never have approached before these glory days.

The hours of design thinking with Claude is exactly it. That's the part nobody talks about because it isn't 'sexy' and doesn't make for a good demo or tweet. But it's the secret sauce IMO.

This. Code generation is cheap, so you can rapidly explore the space and figure out the architecture that best suits the problem. From there, I start fresh and pseudocode the basic pattern I want and have Claude fill in the gaps.

Every time I try to use Claude Desktop, I quickly feel like it's like trying to type wearing mittens. No bueno, at least for me.

My wife calls Claude my girlfriend.

I do a ton of programming but I also use it to learn all kinds of stuff. I'm into physics, history and philosophy and have done wonderful explorations.

Now I tell it what I had for breakfast just to see what it says. Half the time it says something interesting and I end up exploring another new thing.

"My people" for sure and everyone is mad at me because I think that.

Also, I don't care what they think. I am all about the fun.

The brainstorming, investigation and planning are so much fun, aren't they?

Having an infinitely patient, super smart colleague available all the time is amazing.

I have been making web apps for years. A few year ago I converted my base stack into a scaffold that lets me spin up a full working project with API, CLI and UI.

I use NodeJS with a highly structured ExpressJS app for the API. It uses an npm module, tools-library-dot-d to implement a carefully scooped plugin structure for endpoints, data model and data mapping. It has built-in authentication and database (sqlite).

Nuxt/Vue/Vuetify/Pinia for the UI. It has a few components that implement things (like navigation) the way I like. It supports login and user editing.

The stack includes a utility that looks at a directory for executable CLI tools (usually NodeJS or BASH) and adds them to the session PATH. The API stack has boilerplate to treat CLI apps as data-model services.

Does that help?

Definitely! Thanks for the reply.