Quick context on the bots.

I tried standard LLM-based approaches first, but iteration was slow and the bots never felt consistent or aware of real-time game state.

OpenClaw sped up the feedback loop a lot and handled game context much better by operating over structured actions instead of free-form text. The result is bots that actually play the game using the same mechanics as humans.

They’re still imperfect, but good enough to make the game feel alive. Happy to answer questions.

Thanks! Appreciate the encouragement

It's no secret! I used Tyler Hobbs' method which has a great write up here:

https://www.tylerxhobbs.com/words/a-guide-to-simulating-wate...

Then I tweaked it until it look good for use in my app.

Very cool - has a very dreamlike quality to it

Definitely know where you're coming from. I work as a developer in a government job that is unionized, and even though the salary is less than market rate, the security of it keeps me there. If such opportunities are available wherever you're located could be worth looking into. While there are aspects of a gov job that can be boring, overall it's actually been really interesting and challenging work.

Just a little hack project for fun, the kids loved it.

Video of it in action - https://imgur.com/a/g0Jik5b

Anyone played around with this much? Seems interesting, and I'm curious how well it's sped up performance in practical usage.

Check out JoseBriones on YouTube - great dumbphone review channel

Try using Jupyter Notebooks or another notebook based programming environment. Good way to change things up and very enjoyable to use :-)

I've done something similar with gpt3 (codex) and had good results