How vibe coding a game made me design an AI agent protocol (blog.carlid.dev)

In a recent article, a developer shares their transformative journey from skepticism to embracing AI, sparked by the release of Anthropic's Opus 4.5. This pivotal moment led them to explore the application of large language models (LLMs) in creating a new MMO titled "SAO: Slop Art Online," where NPCs are controlled by LLMs, mirroring player interactions. The developer's innovative approach involves modeling NPC behavior with a hybrid technique that utilizes a default behavior tree alongside dynamically regenerated actions based on the game’s state, addressing challenges like latency and scalability. This initiative is significant for the AI/ML community as it introduces a protocol for standardizing how AI agents interact with applications, enhancing real-time responsiveness through a push-based architecture. By contextualizing actions within the application's state and decoupling consumers from providers, the protocol offers a structured way for agents to receive relevant information and execute actions efficiently. Although the game project is on hold, the developer plans to refine and publish the protocol, emphasizing the need for creative and adaptable approaches in utilizing AI tools for future developments.