Day 1 – April 22, 2026
Solution Talks
SOL300 • Compute, Data Analytics • Technical
Beyond memory: Architecting self-evolving agent fleets
location_on
Solution Talks 3 - Palm H
schedule
4:00 PM - 4:45 PM
While models are evolving, most enterprise agents remain static, until now. In this deep-dive, we apply DeepMind's Nested Learning architecture to build self-evolving AI agent fleets. We will architect a Dual-Velocity system on Google Cloud: a Fast Loop for immediate RAG corrections and a Slow Loop that crystallizes institutional knowledge into agent policy via an Atomic agent building approach. Using Vertex AI Agent Engine and BigQuery, discover how to build Atomic Agents that don't just execute instructions—they rewrite them.
Read more