When AI Agent Teams Work (And When They Just Create More Work)
summary
Argues that agent teams pay off only when orchestration, tool access and governance are concrete not conceptual: shows Walmart Element agent mesh running specialist agents across HR inventory and digital twins, demos Crew AI and NAD building real email and calendar workflows, and uses cloud code to auto‑generate a Towers of Hanoi UI that runs in milliseconds for about $3. Highlights the cost math too — Towers of Hanoi: 12.4 million LLM calls yielding $71k solutions versus $2k open alternatives — and warns about security, unpredictability and maintenance.
AI-written summary · about
chapters (12)
- 0:00Welcome
- 1:29AI agents future computing
- 7:12Agentic workflows and teams
- 17:13Multi-agent vs orchestrator
- 20:04Walmart’s multi-agent system
- 27:20Building agent teams
- 36:10More agents better?
- 39:02Cost and efficiency
- 44:00Security risks
- 47:29Real-world agent teams
- 51:04Future ad hoc teams
- 52:51Q&A
related talks
29mWhat exactly are AI Agents?
47mHuman Robot Agent: Scouting AI Beyond Hype and Doom
26mBuilding Agentic Workflows: Patterns for Orchestrating Intelligent Systems.
19mLessons from Transforming Teams to AI Native Workflows
22mHOW TO BUILD YOUR OWN AI AGENT: THE COMPLETE ARCHITECTURE FOR ENGINEERING TEAMS
30mAI Agents Unchained
49mAI agents won’t save your productivity– human–agent collaboration will
01h 17mThe year of the AI Agent: Beyond Automation