I thought I would share some techniques that I have found useful and others can feel free to do the same.
I have been mainly using ChatGPT but have tinkered with Claude a bit as well.
I like to generate an outline in pseudo-code before we start so that the AI can keep a picture of what we are building towards. I also have a general rule of thumb that the smaller of a task that you ask it to do at a time, the more accurate the result will be. So I try to ask for small things at a time where possible.
Something I have been playing with lately... Its been documented that AI will respond differently depending on the level of expertise they detect from the user. I think AI tends to produce better results when you show that you have some understanding of whatever subject you are discussing, prior to generating any code/assets/whatever.
I have not played with agents for game development yet. But I have used the "workforce agents" feature from ChatGPT at my workplace... There are some very cool possibilities there for sure, as you can basically create self-updating instructions and scheduled tasks. Would be interested to know if anyone has played with this.