What Is the Gemini Enterprise Agent Platform?
If you went looking for Vertex AI recently and found something called the Gemini Enterprise Agent Platform, you did not take a wrong turn. Google announced the change at Cloud Next in April 2026, and it went generally available on April 22. Vertex AI is the old name. Agent Platform is the new one.
The rename is not just marketing. It reflects a real inversion in how Google organizes its AI products, and it comes with a second product that shares most of its name, which is where most of the confusion starts.
Two different things are called “Gemini Enterprise”
This trips up almost everyone evaluating it.
Gemini Enterprise is the app. It is what employees open. Search across company data, chat with an assistant, run agents somebody else built. It is licensed per user, per month, in editions, and it is the thing that absorbed Agentspace.
Gemini Enterprise Agent Platform is the developer platform underneath. It is where engineers build, deploy, govern, and monitor agents. It is the renamed Vertex AI Agent Builder, it bills on consumption rather than per seat, and you reach it through the Google Cloud console.
If a vendor, a consultant, or a search result says “Gemini Enterprise” without specifying which, ask. The buying process, the pricing model, and the person who owns the decision are all different.
What the platform is made of
Google split the platform into four jobs: build, scale, govern, and optimize. The named components map onto those.
Build
- Agent Development Kit (ADK) is the code-first framework, available in Python, Go, Java, and TypeScript. Google says it now processes more than six trillion tokens a month. It is open source, and it is the piece with the most independent adoption.
- Agent Studio is the low-code visual builder for designing reasoning loops, wiring data sources, and testing prompts without writing the agent by hand.
- Agent Garden is the template library you start from instead of a blank file.
- Model Garden carries more than 200 models, including Google’s own Gemini and Gemma lines alongside third-party models such as Anthropic’s Claude.
Scale
- Agent Runtime is the managed execution engine. Google rebuilt it for sub-second cold starts and workflows that run for days rather than seconds.
- Agent Memory Bank holds long-term context so an agent recalls prior constraints and decisions instead of restarting cold on every invocation.
Govern
- Agent Identity issues a cryptographic identity to each agent, so authorization is per agent rather than per human service account.
- Agent Gateway is the single control point for connectivity and policy enforcement.
- Agent Registry is the approved catalog of tools, agents, and skills that developers are allowed to pull from.
- Agent Sandbox is the hardened environment for running generated code and driving a browser.
Optimize
- Agent Simulation tests agents against synthetic interactions before they touch production.
- Agent Evaluation scores behavior continuously on live traffic.
- Agent Observability exposes execution traces and the agent’s reasoning path.
The through-line is that the model is no longer the top-level object. The agent is. Model training, deployment, and the registry all now sit underneath agents, which is the actual architectural claim behind the rename.
What it costs
The platform bills on consumption, not seats. In practice that means several meters running at once: compute for the runtime measured in vCPU-hours and GB-hours, storage for sessions and stored memories, per-query charges when the agent searches your data, and foundation model tokens billed separately by whichever model you called.
Rates move, and Google has repriced pieces of this more than once, so price it against the current Google Cloud pricing calculator rather than any published figure, including one you read here. The structural point is more durable than the numbers: an agent that idles cheaply and an agent that runs long multi-day workflows over a large corpus will produce very different bills from the same architecture.
The per-user Gemini Enterprise app editions are a separate line item. Buying one does not get you the other.
Who it is for
The platform is aimed squarely at engineering teams inside companies that already run on Google Cloud and need governance as a first-class requirement. Named early adopters include Comcast, PayPal, L’Oréal, and Geotab, which tells you the intended shape of the buyer.
That focus is a strength and a filter. Identity per agent, a policy gateway, an approved tool registry, simulation before release, and continuous evaluation are exactly what a regulated enterprise needs before letting an autonomous agent touch a customer record. They are also weeks of setup that a five-person company will never recoup.
If you are a solo operator or a small team, the honest read is that this platform is not built for you, and the answer is not to use a smaller slice of it. The answer is a different category of tool.
Not every agent problem needs a cloud platform. If what you actually want is an agent that watches your inbox and calendar and acts on what shows up, Carly runs that without a Google Cloud project behind it: free Zapier-style workflows, with AI agents from $35/month.
Where Workspace Studio fits
Google shipped a second, unrelated no-code agent builder called Workspace Studio, and it is a genuinely different product with a different audience. Workspace Studio lives inside Google Workspace, is included with Business and Enterprise editions rather than sold separately, and targets an operations person automating their own work in Gmail, Drive, and Calendar. No cloud project, no billing account, no code.
Agent Platform targets a developer shipping an agent to production with an audit trail. The two are not competitors and they are not tiers of each other. If you are evaluating “Google’s agent product” and have not decided which of these you mean, you are evaluating two products at once.
How agents talk to each other
The platform leans on two open protocols, and they solve different halves of the same problem.
MCP connects an agent to tools and data. If you have read about MCP servers, that is the layer being described, and Google now offers managed MCP servers with Apigee acting as the bridge from an existing API to an agent-callable tool.
A2A, or Agent2Agent, connects agents to each other, so an agent from one vendor can delegate to an agent from another without a custom integration between them. Google reports A2A v1.0 running in production at around 150 organizations.
The practical consequence is that the platform is less of a walled garden than the branding suggests. An agent built with ADK can call tools it did not ship with and hand work to agents Google did not write.
FAQ
Is Vertex AI discontinued?
No. It was renamed. Existing Vertex AI Agent Builder services carry forward under the Agent Platform umbrella, and Google’s own product page still carries the “formerly Vertex AI” label. Some individual modules have been marked deprecated on their own schedules, so check the deprecation notice for the specific service you depend on rather than assuming the rename alone changed anything.
What happened to Agentspace?
It was folded into the Gemini Enterprise app, and existing Agentspace agents migrate automatically. The standalone product name is going away.
Do I need Gemini Enterprise to use the Agent Platform?
No. They are sold separately. The platform is consumption-billed through Google Cloud, and the app is licensed per user. Many teams buy one without the other.
Is ADK tied to Gemini models?
No. ADK is model-agnostic and open source, and Model Garden carries more than 200 models including Anthropic’s Claude. You can build on ADK and call a non-Google model.
Can a small team use this?
Technically yes, practically rarely. The governance layer that justifies the platform for a large enterprise is overhead that a small team pays for and does not use. Smaller teams generally land on a hosted agent product instead, and the comparison worth running is against the best AI agents for productivity rather than against another cloud platform.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
