Autonomous Al Agents for Cloud Cost Analysis
summary
Describes a Spotify system that uses LangGraph to orchestrate specialized agents: a self‑healing SQL expert tuned for BigQuery (up to three automatic repair attempts), a React-style agent that generates and sandbox-executes Python for precise math, and web grounding via Google Search API with Gemini. Claims evaluation shows the multi-agent stack reaches 58% vs 22% accuracy on real Slack questions using OpenEval LLM-as-judge, and discusses N-way/MCTS search and continuous replanning.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:21Cloud cost challenges
- 1:03Traditional cost limits
- 1:45First workflow agent
- 2:29SQL reliability issues
- 2:50Python agent improvements
- 3:32Self-healing SQL agent
- 3:53Web search for knowledge
- 4:35Task planning architecture
- 5:17Verification and replanning
- 5:58Multi-agent architecture
- 7:03System reasoning example
- 7:25Evaluating real questions
- 8:29Results and gains
- 9:12Scaling in three dimensions
- 9:55Future autonomous goals
- 10:36Q&A
related talks
34mAI Agents: Transforming Autonomous Decision-Making
30mBuilding Robust, Secure LLM and Agentic AI Workflows
25mWho Let the Bots Out? A Guide to Evaluating AI Agents
30mBuilding Reliable AI Agents: An Open Source Approach To Evaluation and Observability
01h 27mBuilding Vertical Agent with AG2 AgentOS
24mAI Agent Swarms Are Amazing
49mAI Agents: Collaboration in Real-Life Applications
01h 01mKeynote Panel: Future of AI agents