[VertanaNinjai]

Claude’s web UI has some strange limitations. You may be better off using another provider that just uses Claude 3.5 Sonnet API. One example of many is Kagi’s interface. You can set a system prompt and Claude will actually follow it instead of trying to be too clever. If your prompt says “give full and complete code files at all times unless specified otherwise by the user” it tends to follow that way more accurately than Claude.ai. May be worth a try for you to check out some other providers/interfaces.

[UncleEntity]

Perhaps...

Seems more like throwing good money after bad as they appear to make it so you have to get an API key to do anything useful and then there's the possibility of spending significantly more than the $20 month 'pro' plan. Not something I'm really all that interested in considering this has nothing more than entertainment value and, from a video on the youtubes that broke the costs down, they have the highest per token rate.

I have enough yaks to shave to justify $20/month but anything over that...

[techbrigades]

I’ve had really good luck building micro-SaaS up to a certain point of complexity, and beyond what a coding assistant will do.

A lot of the problems you describe are due to the need to provide adequate context, project details, and rules in your prompt.

Claude will generate files it doesn’t think already exist, and in a SaaS project, you have many files.

Claude will struggle with syntax and proper usage of library/module/package versions beyond its training data. You have to provide your project with knowledge to work around this.

Lastly you will hit usage limits while working on projects because it’s a fixed cost offering. You can track and generate a project status to pass on to the next agent. This works “ok” and when I hit my sonnet limit I use haiku to bug and type fixes.

Bottom line, an out of the box chatbot is a great playground to flush out your techniques, but most software projects have complexities which must be managed in a separate system designed to manage all the details and break down the project into hundreds or thousands of individual tasks.

I want this magic wand too, but you have to build that yourself (or buy it when it becomes available). It’s been a fascinating learning process.

[UncleEntity]

The problem is I'm not doing anything as complicated as what you're describing.

The task was/is to take a grammar for APL from some long forgotten paper and turn it into a lemon parser. Easy, peasy, well within its wheelhouse and it had spectacular initial results with the help of DeepSeek-R1 analyzing its work.

"Oh, good job, robot," me types, "let's work on a lexer. Hmm... you seem to have clipped out some important rules at some point, we need to add those back." Then, boom, Claude is completely worthless.

I want Claude to succeed. It was doing so well then it hit a self-reinforcing wall of failure that it just can't get over even though it can analyze its behavior and say exactly why it keeps failing.

I mean, exactly zero people think the world needs an APL interpreter written by the robots but the point of the project is to see how far they can get without having a human write a single line of code. I know they have limitations and have no problem helping them work around them.

But, alas, this project is shelved until the next big hype cycle.