The AI Agent Iceberg
summary
Sierra builds customer-facing agents for large enterprises handling tens to hundreds of millions of interactions, and ties revenue to outcome-based pricing instead of tokens or seats. Production work surfaces concrete failures: voice models hit ~40% on a single booking task and collapse by eight repeated iterations per the Tao benchmark, simulations show both sides can be agents, and real deployments require PII detection, payments flow, multilingual Taglish handling, and robust human handoffs. Product pieces include Ghostwriter and Explorer for agent generation and live optimization.
AI-written summary · about
chapters (9)
- 0:00Introduction to Sierra
- 1:11AI agents in production
- 5:01The AI agent iceberg
- 8:54Complexity in voice agents
- 10:38Tao benchmarks explained
- 11:42Language and cultural nuances
- 13:05Designing for failure modes
- 14:08Ambidextrous AI teams
- 14:49Ghostwriter and Explorer tools
related talks
20mThe AI Agent Iceberg
19mRise of the AI Architect
56mAI Agents That Don’t Suck: How to Build Ones That Actually Drive Business Value
30mAI Agents Unchained
18mFrom POC to Profit: Deploying Customer-Facing AI Agents That Actually Work
36mAI Agents in Automation | 3 Real Production Cases with ROI
43mTesting Agentic AI Applications: Beyond Traditional QA
49mHow to be an AI First company