Claude AI is perfectly supporting Hollywood 11 and all plugins. I just created a TEX viewer using HTex loading a text file as parameter and showing it. If more than one page you can use arrow keys. All in perhaps 30 minutes

It is supporting lots of amiga specific languages, even ace basic or amos

you can also let it rewrite arexx in regina

not tested yet of course

just tested developing a picture viewer using data types in Amiga-E (evo) in just 5 minutes. Unbelievable

From what I have noticed about AI, unless you are using an advanced version or a commercial version the standard free AI tools are good for doing research. The code provided should be regarded as non-working because when you try to compile it there are many errors! I believe it's done purposely to insert errors so we are persuaded to use the more advanced or commercial version. The errors in the code are consistent.

It's up to the developer to judge whether it is functional, working code and to fix any errors in the code. As long you understand that what AI provides may not work as is. It needs a little help from a human to edit the code and fix it. However, using AI does help speedup production!

I tested Hollywood code in free edition and it worked, then I took the Pro version. In about 30 minutes I created a simple TEX viewer. I have also tested evo and created a small utility that uses datatype system to show pictures in E. Including 2 changes and testing I needed 5-10 minutes to have a working version.

@miker1264
It’s true that commercial versions are more sophisticated, but they come at a cost—and even they generate errors.

The problem is that AI systems get all their information from the web, including forums; in addition to reading correct information, they also read incorrect information.

The errors are caused partly by incorrect information but mainly by the confusion among the various operating systems. They forget very easily—just ask something unusual, and they’ll change the subject and go haywire, forgetting everything.

AROS is often confused with AMiGA or Linux; you always have to keep your eyes wide open and check the output strings. Eventually, by repeatedly attaching the error logs, they’ll manage to solve your problem.

Keep in mind that all the programs I’ve created were developed using free A.I.

A few things you can do to improve the experience, depending on how deep into the rabbit hole you want to go.

1. Documentation is crucial. LLMs are probabilistic; they need grounding. Write down what you are trying to achieve. Show examples, draw wireframes, take screenshots
2. Install superpowers: https://github.co...uperpowers
3. Try different approaches, e.g., maker-checker: one model creates a solution, the other tries to break it; another is reviewer: one agent writes, the other reviews and advises
4. Have the source code - for problems with AROS vs. Amiga, it's very helpful to simply have the source code of AROS and say: verify with the source
5. Watch out your context window: /clear and /compact are your friends. It's better to break implementation into small steps (that's why point 1. is crucial: document describes the big picture, you split it in steps) - if LLM makes an error somewhere, it will store it in the context window and each next step may feel that it's getting dumber.

And a private one: building apps is great, especially when it works one-shot, but let's help each other - build/port things others can build on top: new Zune cl**** new library, datatype etc

Have fun!