Overview
- Google announced the change on April 22 at Cloud Next ’26 in Las Vegas, folding Vertex AI into a new platform built to design, run, and audit autonomous agents at scale.
- The stack spans four areas — Build, Scale, Govern, Optimize — with tools like Agent Studio for low‑code design, the Agent Development Kit for code‑first workflows, Memory Bank for long‑term state, and an Agent Registry and Gateway to control which tools agents can use.
- An Agentic Data Cloud aims to ground agent decisions in business context through a Knowledge Catalog that auto‑ingests and enriches data, graph‑aware retrieval in BigQuery Graph, and a Cross‑Cloud Lakehouse that queries data in AWS and Azure without moving it, with key federation features in preview.
- Security and oversight are built in through cryptographic Agent Identity, a policy‑enforcing Gateway with Model Armor to block prompt injection and data leaks, and standards like the Model Context Protocol for safe tool access and agent‑to‑agent communication for coordinated tasks.
- Google is pairing the software push with specialized chips — TPU 8t for training and TPU 8i for inference — planned for general availability later in 2026, while early customers like Comcast, Color Health, L’Oréal, Payhawk, and PayPal pilot the platform as migration and maturity are tested in the coming months.