Google Build with AI Keynote: Agentic Reality Check

Key Themes
Design for Trust: Moving from "black box" automation to transparent, governed systems.

The "Flight Recorder" Concept: The necessity of logging and observability in agentic workflows to ensure accountability and error correction.

Defense in Depth: Implementing "Model Armor" and multi-layered security to protect against common 2026 exploits and unauthorized data access.

Orchestration Over Autonomy: Why "multi-agent" is often just a polite term for a distributed system, and how to manage the resulting complexity.

Evaluation as Leverage: Shifting the perspective of evaluation from a chore ("broccoli") to a strategic advantage ("leverage") for refining system output.

The Pragmatist’s Playbook
The deck concludes with actionable strategies for architects and product leaders to build agent-to-agent (A2A) systems that don't share memory or compromise security, ensuring AI integration remains an asset rather than a liability.

Transcript

2. Noble Ackerson Google Developers Expert for AI/ML Senior Director, AI & Agentic Solutions . Leidos.