I specifically have used Claude, Opus 4.1 model. Love how it creates and shows you the code. I tell Claude to develop a HTML5, Javascript single file game and it just does it. The way it organizes the code, in Artifacts, which when I first started on the Claude AI coding journey didn’t know existed. In Artifacts you can toggle on and off if you want to see the code or run the code.

An example of creating a game. Told Claude:

And it created:

I can create a game with from a small paragraph. The issue is, which is a big one. You start playing the game that was just built and it’s not very good at all. Gives you a starting point to build a much better game, or web app. Which can be grueling task. Meaning I can start the project, build a base, add more features and update the design. Run into issues, go through copying and pasting error’s to help Claude on what the issue is. Then you get a “Your message will exceed the length limit for this chat.”. Then you open a new chat, and start all over again. Need to train the chat on what you want and the system specs.

Recently Claude launched with the Max plan that you can enable Claude to search previous chats but it just searches, it doesn’t remember. It helps reference a chat but you have to train Claude again. And in my experience you can create a perfect prompt explaining everything in detail, upload the necessary files required, but it still doesn’t remember everything even during the same chat. If you are building something, don’t have Claude blindly build a script that changes production, WITHOUT reviewing every detail of what it’s doing. I broke production for a while because I trusted Claude to create a migration script from DEV to PROD, of one of my web apps.