Vertex AI Agent Builder: What It Was, What Replaced It
Vertex AI Agent Builder was Google Cloud’s bundle for building production AI agents. As of April 22, 2026, it is the Gemini Enterprise Agent Platform. Google’s own product page now reads “Gemini Enterprise Agent Platform (formerly Vertex AI).”
If you are here because a tutorial, a job posting, or a half-finished internal doc referenced Agent Builder and the console no longer matches, this is what changed and what did not.
What Agent Builder actually was
It was less a single product than a set of pieces sold together, which is part of why the rename made sense.
Agent Development Kit (ADK) was, and still is, the code-first framework. Python, Go, Java, and TypeScript, model-agnostic, open source, with a graph-based structure for multi-agent setups where a supervisor routes work to specialists. ADK is the piece with real independent adoption, and it survived the rename untouched.
Agent Studio was the low-code console for designing reasoning loops, connecting data sources, and testing prompts visually instead of writing the agent by hand.
Agent Garden was the template library, so common patterns started from something rather than nothing.
Agent Engine was the managed runtime. Deployment, scaling, session handling, and memory, so you were not operating the infrastructure yourself.
Memory Bank gave agents persistent recall across sessions, which is the difference between an agent that remembers a customer’s constraints and one that asks the same question every Tuesday.
Model Garden provided access to more than 200 foundation models, Google’s own alongside third-party options including Anthropic’s Claude.
Around all of it sat the governance layer: per-agent identity, a policy gateway, and content safeguards on model input and output.
What changed on April 22, 2026
Google announced the consolidation at Cloud Next and shipped it the same month. Three things actually moved.
The name. Vertex AI became the Gemini Enterprise Agent Platform. All Vertex AI services and future roadmap now ship under that umbrella rather than as a standalone Vertex AI product.
The hierarchy. This is the substantive change. Under Vertex AI, the model was the top-level object and agents were something you built with it. Under Agent Platform, the agent is the top-level object and model training, deployment, and the registry sit underneath. The console reflects that reordering.
The neighbors. Agentspace, Google’s separate enterprise search and assistant product, was folded into the Gemini Enterprise app at the same time. Two product lines became one story. If you were an Agentspace customer, that migration is covered separately.
What you have to do about it
For most existing Agent Builder users, nothing. The services are the same services under a new name, and code written against ADK keeps running.
The exception worth checking is module-level deprecation. Consolidations of this size usually retire specific older components on their own timelines even when the umbrella product carries forward, and Google has done exactly that with pieces of the old Vertex AI surface. Third-party write-ups have circulated specific cutoff dates that are not consistently reflected in Google’s own deprecation notices, so verify against the deprecation page for the individual service you depend on rather than trusting a blog post, including this one. A wrong date here costs you an outage.
The other practical change is naming. Search results, Stack Overflow answers, and internal runbooks will say “Vertex AI” for years. When you are debugging, treat the two names as synonyms rather than assuming a version difference.
How it is priced
Agent Builder never had a seat price, and Agent Platform does not either. It bills on consumption, with several meters running at once:
- Runtime compute, measured in vCPU-hours and GB-hours for the time your agent is actually executing
- Session and memory storage, charged per stored event or memory
- Search queries, charged per thousand when the agent retrieves from your indexed data
- Model tokens, billed separately by whichever model the agent called
Published rates for each of these have moved more than once, so price a real workload in the Google Cloud pricing calculator rather than working from any figure you find written down. The shape is the useful part: an agent that answers a question in two seconds and an agent that runs a multi-day workflow across a large corpus bill very differently even when the code is identical.
Is it still the right choice?
It depends almost entirely on whether you need the governance layer.
The case for it is straightforward. If you are putting an autonomous agent in front of customer records inside a company with auditors, then per-agent cryptographic identity, a central policy gateway, an approved tool registry, simulation before release, and continuous evaluation on live traffic are not overhead. They are the reason the project gets approved at all. Nothing in the hosted-agent market matches that depth, and Google reports ADK processing more than six trillion tokens a month, so the platform is not a science project.
The case against it is equally straightforward. All of that is weeks of setup and a permanent operational surface. If your agent needs to read your inbox, notice that a client asked to move a meeting, check your calendar, propose three times, and update the CRM, you are paying for an enterprise control plane to do work that does not need one.
That second description covers most small teams and solo operators. If it covers you, Carly does that job without a cloud project, a billing account, or a deployment pipeline: free Zapier-style workflows, with AI agents from $35/month. The best AI agents for productivity covers the wider field if you want to compare before deciding.
FAQ
Is Vertex AI Agent Builder discontinued?
No. It was renamed to the Gemini Enterprise Agent Platform in April 2026. The services carry forward, and Google labels the new product “formerly Vertex AI.” Individual modules have their own deprecation schedules, which is a separate question from the rename.
Do I need to migrate my agents?
Generally no. Existing Agent Builder services moved under the new umbrella rather than being replaced, so agents keep running. Check the deprecation notice for any specific service you depend on.
Is ADK going away?
No. The Agent Development Kit is open source, model-agnostic, and available in Python, Go, Java, and TypeScript. It is the component Google has invested in most visibly, and it carried through the rename unchanged.
What is the difference between Agent Platform and Gemini Enterprise?
Agent Platform is the developer platform, billed on consumption, reached through the Google Cloud console. Gemini Enterprise is the employee-facing app, licensed per user per month. They are sold separately and bought by different people.
Can I use Agent Builder without Google Cloud?
No. It is a Google Cloud product and requires a project and billing account. ADK itself is open source and can run elsewhere, but the managed runtime, memory, and governance pieces are tied to the 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."
