LLM-powered agents require structured, discoverable, self-documenting APIs, and GraphQL fits perfectly with its introspectable schema and precise data fetching. This talk explores the emerging intersection of GraphQL and AI agents. We'll cover why GraphQL's type system gives agents a major advantage over REST for tool use, how to expose a GraphQL schema as an LLM tool with automatic argument construction from the schema, practical patterns for letting agents compose queries dynamically while keeping cost and complexity bounded, and real-world examples using function calling with OpenAI, Anthropic, and open-source models. We'll build a working agent live that discovers, queries, and mutates a GraphQL API, and discuss guardrails like query allow-lists, mutation scoping, and human-in-the-loop approval for destructive operations.