The short version: Google renamed Vertex AI to the Gemini Enterprise Agent Platform at Cloud Next 26 (April 22, 2026) and folded the old Agentspace and the Gemini Code Assist enterprise tier into it — one console, one bill. Your API endpoints are unchanged, so nothing in your code breaks. What changed is where things live, what they're called, and the mental model behind the whole thing. Here's the map.

The rename you can ignore, and the one you can't#

Most of this is cosmetic and backward-compatible. Per Google's own name-changes doc, the request surface didn't move — if your service was wired up against Vertex AI, it kept running the day the sign on the door changed. There's no migration fire drill. Vertex AI simply stopped appearing as its own product in the Cloud Console around May 21, 2026, and its features — Model Garden, AutoML, the Model Registry, Endpoints, Pipelines — now sit as components inside the Agent Platform.

The rename you can't ignore is quieter: Agent Engine is now "Deployments." That's the managed runtime you use to deploy and scale an agent. Same service, new menu name. If your runbooks, internal docs, or onboarding guides say "Agent Engine," they're now pointing at a label that isn't in the console. Grep your wiki and fix the word; you don't have to touch the workflow.

The real story is the inversion#

Strip the branding and here's what actually happened. Vertex AI was a model platform — training, tuning, serving — that had grown some agent features on the side. The Gemini Enterprise Agent Platform flips the hierarchy. Agents are the headline now; model training, AutoML, the Model Registry, and Endpoints are demoted to sub-features of an agent-first stack. As AIwire put it at launch, Google expanded Vertex "into a full agent stack" rather than adding another SKU.

Two moves make the intent obvious. Google shipped the A2A protocol v1.0 as the platform's default agent-to-agent interoperability layer — the same Agent2Agent standard it has been pushing across the industry — and bundled a no-code builder, Workspace Studio, on top of 200+ models in the Model Garden. When a hyperscaler makes interop the default and hands non-engineers an agent builder, it isn't defending a model business. It's betting on the full agent stack.

The one line item worth a second look#

Buried in the consolidation is Memory Bank: a managed service that stores long-term memory for agents so they carry context across sessions. If you're building on Google Cloud and had "stand up an agent memory layer" on your roadmap, this turns it into a build-vs-buy question you should answer deliberately, not by default.

That's not a rubber stamp for the managed option. Agent memory is a crowded field, and a managed Memory Bank trades control and portability for one less thing to run — the same tradeoff every memory decision comes down to. But it's now sitting in the same console as your deployments, which means the path of least resistance points at it. Know it's there, benchmark it against a dedicated memory system on your data, and choose on the merits instead of on where the button happens to be.

What to actually do#

Three things, in order. One: don't panic — your endpoints work, this is not a migration. Two: update the vocabulary in your docs and dashboards (Vertex AI → Gemini Enterprise Agent Platform; Agent Engine → Deployments), because stale names cost your team search time. Three: if agent memory or agent-to-agent interop is on your roadmap, re-open those decisions now that Memory Bank and A2A v1.0 are first-class in the platform you're already paying for. The name change is free to ignore. The re-org underneath it is worth ten minutes.