Opposites attract. Large language models like GPT4o are by their very nature unpredictable. If you want to scale out the complexity of apps that leverage them then you need to partner 'em with something very predictable. That's where state machines like XState come in. State machines provide a predictable container to run your AI apps within, allowing you to focus on the logic and letting your users play within the sandbox your state machine places them in. You can then integrate with new features like AI tooling to let the large language model inform you when a state transition should occur. This increases the observability of your apps while reducing the bugs, gives visibility of complex workflows and allows you to aim for even bigger problems to solve.