Building Frontier CX Agents | Interrupt 26

In this keynote from Interrupt 2026, Cisco Customer Experience Fellow and Chief Architect Carlos Pereira pulls back the curtain on how one of the world's largest enterprise organizations is building and scaling frontier AI agents for its Customer Experience (CX) division. Carlos walks through Cisco's real-world journey from early chatbots and the "Agentic Foundation" to a fully AI-native "Renews Teammate" running in production across a $26B+ recurring revenue business. He shares candid lessons learned: why 95% accuracy wasn't enough for adoption, how bolting AI onto broken workflows can actually accelerate failure, and why the key to scaling agentic systems is keeping supervisors lean, introducing domain-specific planners, and building in self-correction by default. You'll come away with a practical architecture playbook—covering supervisor graphs, nested subgraph planners, deterministic task workflows, routing-first design, and long-term memory—plus an honest look at the organizational challenges of getting people to actually use these systems. Building Frontier CX Agents | Interrupt 26 00:00 Introduction & Cisco CX Overview 01:22 The Land, Adopt, Expand & Renew Model 02:36 From ChatGPT Hype to B2B Agentic AI 03:16 Why 2026 Is the Year of AI-Native Business Workflows 04:18 The Problem with Bolting AI onto Broken Workflows 05:01 The Agentic Foundation (Built in 2025) 05:49 Evolving to the "Renews Teammate" Concept 06:29 Architecture Deep Dive: Supervisor, Planner & Agent Layers 07:51 Introduction of the Planner Node 10:21 Zooming In: How the Supervisor Handles Complex Queries 12:14 Dynamic Replanning & Self-Correction Loops 14:43 Deterministic Task Workflows vs. LLM Reasoning 16:57 The Adoption Plateau: Why Users Ghosted the System 18:11 Forced Curiosity & Evolving from Chatbot to Teammate 18:50 Flipping the Mindset: Humans Help Software, Not Vice Versa 20:29 Key Features: Long-Term Memory, Scheduling & Proactive Intelligence 20:53 Lessons Learned: System Level 2