A GOTO 2025 conference talk by an AWS Java specialist making the case for Java and Spring AI as a viable stack for building agentic applications in enterprise settings. The talk covers the core building blocks of agents (LLM, prompt, memory, RAG, tools), demonstrates how Spring AI abstracts these components, and shows a live travel booking agent demo. Key practical insights include: the real value of LLMs is converting natural language to structured API calls, not pure reasoning; fully autonomous agents are still unreliable at 70-80% accuracy; the pragmatic approach is sprinkling LLMs into existing validated workflows rather than replacing them; and small focused agents (3-10 steps) beat monolithic approaches. Spring AI's annotations and advisers make it easy to expose existing Java beans as MCP tool servers without rewriting enterprise services in Python.
Sort: