Keynote: Fireside Chat
summary
Defines agentic AI as iterative, multi-step workflows and lays out four design patterns: tool calling, planning, reflection, multi-agent collaboration, argues frameworks (LangChain, LangGraph, Crew AI, Haystack, LlamaIndex) are useful but optional, and reports diverse memory approaches (snippet stores, graph memories; startups like Zep and Cogni) with no single architecture dominating. Shows knowledge graphs improving RAG precision—moving examples from ~90% toward ~98% for high‑stakes tasks—and lists practical tools used today: Gemini, Cloud Code, OpenAI Codex, Open‑source research prototypes.
AI-written summary · about
chapters (17)
- 0:00Welcome.
- 0:26Agentic AI - What is it?
- 2:094 Design Patterns of Agentic AI
- 4:12Framework or no framework?
- 6:05Where is it going (with frameworks)?
- 6:42Shape of memory?
- 9:04Interplay of memory with "RAG corpus"?
- 10:29Future of agentic AI?
- 12:41Multi-agent and SLMs?
- 14:13AI adoption in the Enterprise?
- 17:25Transformation playbook?
- 19:15Getting data AI-ready?
- 23:38What is it about a KG that improves accuracy?
- 25:49How do you learn?
- 27:27On a normal day, how much time learning?
- 28:23Personal AI tool stack?
- 29:13Thank you!
related talks
01h 01mKeynote Panel: Future of AI agents
05mKeynote: How Open Protocols Unleash AI’s Next Wave of Creativity and Impact
32mPanel Discussion: Trends in Generative AI
27mEmerging Patterns Agentic AI
15mKeynote: The Common Pathway of AI and Software
10mKeynote: Next-Gen AI Applications via Multi-Agent Conversation
19mBeyond agents: AI as a creative partner- 10mKeynote: Next-Gen AI Applications via Multi-Agent Conversation