You gotta give them guardrails. It's like AI in chess. Engines work because the game has a limited rule set and can explore possibilities ina closed space. When you are talking to an LLM you are *defining those rules of that "game"* for each session. You are literally making the game as you go. If the "game" is small enough, LLMs work. If it is too big or too undefined, there won't be enough tokens to do well. View quoted note →