Ryan, Travis and guest Shayne Boyer from Microsoft discuss “agent orchestration,” or humans coordinating multiple specialized AI agents to complete multi-step tasks. They cite recent developments like Perplexity Computer, Google/Samsung Gemini multi-step mobile agents, Open/Claude tools, and Microsoft Copilot Tasks, and explain that routing work to the best model and giving agents tool access are key trends. The conversation stresses that despite hype, agents are brittle, often produce low-value output, and require heavy human “composer/puppet master” supervision, clear prescriptions, guardrails, evaluation, and delegation skills. They compare multi-agent setups to specialized human teams, planning/execution/eval roles, and even autopilot risks around over-trust, while noting sustainability and cost/token limits. They encourage listeners to start small and gradually delegate tasks without becoming paralyzed.

00:00 Welcome and Guest Intro

01:05 What Agent Orchestration Means

03:18 Why Agents Are Everywhere Now

04:15 Travis on Agentic Computing

06:11 Shane on Jarvis Dreams

08:26 When Agents Fail Hilariously

09:56 Tools and Model Routing

11:38 Delegation and Trust Risks

14:50 How Orchestration Works

18:25 Ant Farm Multi Agent Experiment

21:14 Why Multi Agent Helps

25:49 Baseball Team of Agents

27:56 Sustainable AI Pace

28:55 Empowered PR Culture

30:26 Grumpy Reality Check

34:57 Gardening the Agents

38:46 Supervision Is the Job

42:27 Managing Agent Teams

44:07 Multi Agent Life

45:28 Token Costs and Access

47:43 Demystify the Hype

50:06 Try It Step by Step

51:17 Wrap Up and Thanks

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aiunprompted.substack.com