Post

finally embracing the ralph loop. this is actually pretty nice. most of my LLM usage recently has been letting it run and just babysitting it and approving commands as it follows a detailed spec. now i'm using docker sandboxes and letting agents run with bypass permissions enabled without worry.