From Assistants to Agents: Self-Improving Agentic Systems with Spring AI
summary
Spring AI turns a few concrete primitives into agentic systems: the ChatClient plus structured output advisor and interceptor-style Advisors for memory, guardrails, retries and tool calling. Demos use Tool Calling APIs, a tool search layer that halves token use by progressively registering tools, and agent skills that run scripts in sandboxed tools; retries and judge-LLM loops fix 1–3% malformed outputs but can add latency. Subagents isolate context per subtask and the patterns consistently reduce boilerplate to a few lines of code.
AI-written summary · about
chapters (14)
- 0:00Agentic systems with Spring AI
- 3:17Advisors and interceptors
- 7:27Chat memory with advisors
- 10:05Guardrails and output validation
- 13:37Self-improving loops with judge LLMs
- 17:24Tool calling as advisor
- 21:58Progressive tool disclosure
- 28:18Agent skills and context
- 36:20Agent skills with shell scripts
- 40:54Task order with to-do lists
- 44:57Clarification tools
- 48:43Sub agents for tasks
- 50:49Advisor patterns in Spring AI
- 51:10Q&A
related talks
54mFrom Single-Shot LLMs to Intelligent Agents: Building Scalable AI Systems with Spring AI and MCP
44mReal-World AI Patterns with Spring AI and Vaadin
24mBuilding Agentic Applications with Spring AI
27mBuilding AI Agents with Spring & MCP
59mFrom Pair to Peer Programmer
01h 52mDiving into Spring AI: Building Production-Ready Agentic Applications
34mAI Agents: Transforming Autonomous Decision-Making
40mAgentic AI: The Next Evolution of Automation