Key takeaways
- Google Cloud describes Vertex AI Agent Builder as a suite of products to build, scale and govern AI agents in production; on 22 April 2026 it became part of Gemini Enterprise Agent Platform, and Vertex AI Agent Engine became Agent Runtime.
- Developers build agents in code with the open-source Agent Development Kit (ADK) or in the low-code Agent Studio, run them on Agent Runtime with Sessions and Memory Bank, and govern them with Agent Identity, Agent Registry and Agent Gateway.
- It is priced by use: model tokens plus Agent Runtime at $0.085 per vCPU-hour and $0.009 per GiB-hour after a monthly free tier, and its Middle East agent location is Tel Aviv, not Doha or Dammam, as of November 2026.
What Vertex AI Agent Builder is now
Google Cloud described Vertex AI Agent Builder as its open platform to build, scale and govern reliable agents, with the Agent Development Kit (ADK) at its core. At Google Cloud Next on 22 April 2026, Google launched Gemini Enterprise Agent Platform, which it calls the evolution of Vertex AI, and said Vertex AI services will from then on be delivered only through the Agent Platform.
| Earlier name | Name now |
|---|---|
| Vertex AI Agent Builder | Part of Gemini Enterprise Agent Platform |
| Vertex AI Agent Engine | Agent Runtime |
| Agent Engine Sessions and Memory Bank | Agent Platform Sessions and Memory Bank |
| Vertex AI Studio | Agent Studio |
| Vertex AI Search | Agent Search |
| Vertex AI Model Garden | Model Garden |
Gemini Enterprise is a different product: the app where employees use agents. Agents built on Agent Platform can be delivered to employees through the Gemini Enterprise app.
How agents are built
- ADK: Google's open-source framework to build, debug and deploy agents and multi-agent systems, available in Python, TypeScript, Go, Java and Kotlin.
- Agent Studio: a low-code visual builder whose agents can be exported to ADK for full-code work; Agent Garden has prebuilt templates.
- Models: more than 200 in Model Garden, including Gemini and Anthropic's Claude models.
- Tools: your APIs, MCP servers and search over your content with Agent Search.
How agents run and are governed
| Component | What Google says it does |
|---|---|
| Agent Runtime | A fully managed runtime to deploy, operate and scale agents, with sub-second cold starts and support for agents that run for days. |
| Sessions and Memory Bank | Conversation history, and long-term memories generated and curated from conversations. |
| Agent Sandbox | A hardened environment for model-generated code and computer-use tasks. |
| Agent Identity | A unique cryptographic ID for each agent, with an auditable trail mapped to authorization policies. |
| Agent Registry and Agent Gateway | One index of agents, tools and skills, and a gateway that applies security policies and Model Armor protections. |
| Evaluation and observability | Simulation, evaluation on live traffic, traces and dashboards. |
What it costs
| Item | Google's list price |
|---|---|
| Agent Compute | $0.085 per vCPU-hour, after 50 free vCPU-hours a month per account. |
| Agent Memory | $0.009 per GiB-hour, after 100 free GiB-hours a month; idle time between turns is not billed for Runtime. |
| Sessions and Memory Bank | Storage at $0.30 per GiB a month; reads at 1 vCPU-hour per 3 million and writes at 1 vCPU-hour per 1 million. |
| Agent Gateway | Billed in vCPU-hours: 1 vCPU-hour per 15,000 API calls or authorization requests. |
| Models | Billed per token at each model's own rates. |
Where agents can run
Google lists 23 regions for Agent Runtime, Sessions, Memory Bank, sandboxes and Agent Gateway, including Frankfurt, London, Mumbai and Singapore. In the Middle East only Tel Aviv (me-west1) is listed, not Doha (me-central1) or Dammam (me-central2), as of November 2026. Data at rest stays in the chosen location, while model processing follows the location configured for each model.
How to start
Pick the job
One agent with a clear task, the data and systems it needs and a way to check its answers.
Choose model and region
A model from Model Garden and a region that fits your data rules.
Build with ADK
Instructions, tools and sub-agents in code, tested locally.
Deploy and govern
Agent Runtime with Sessions and Memory Bank, Agent Identity and Gateway policies.
Evaluate and launch
Evaluations and traces before launch, then a monthly review of quality and cost.
What to check before you start
- Data and law: where your data is stored and where the model processes it, against your data rules and the law that applies to you, such as the UAE's personal data protection law or the GDPR.
- Names: older guides and code use Vertex AI and Agent Engine names; in version 2.0.1 of the SDK the agent_engines module moved to the agentplatform package.
- Costs: estimate model tokens, runtime, sessions and memory together with realistic traffic.
- Language: test Arabic and English with your own data.
How Birbal AI helps
Birbal AI builds agents on Google Cloud with Vertex AI Agent Builder with engineers based in the UAE, and sets up Gemini Enterprise for the employees who use them. Birbal AI is an independent company, not part of Google.
Questions and answers
What is Vertex AI Agent Builder?
Google Cloud's suite of products to build, scale and govern AI agents in production, including the Agent Development Kit and Agent Engine. Since 22 April 2026 it has been part of Gemini Enterprise Agent Platform.
Is Vertex AI Agent Builder being discontinued?
Google says Agent Builder has become part of Gemini Enterprise Agent Platform and that Vertex AI services are now delivered through the Agent Platform. The capabilities continue under new names, such as Agent Runtime for Agent Engine.
What is the Agent Development Kit (ADK)?
Google's open-source framework for building, debugging and deploying agents and multi-agent systems, in Python, TypeScript, Go, Java and Kotlin.
How much does Agent Platform cost?
It is usage-based: model tokens at each model's rates, plus Agent Runtime at $0.085 per vCPU-hour and $0.009 per GiB-hour after a monthly free tier, and storage and operations for Sessions and Memory Bank.
Can agents run in the Middle East?
As of November 2026, Google lists Tel Aviv as the only Middle East location for Agent Runtime, Sessions and Memory Bank; Doha and Dammam are not listed.
Does Birbal AI build these agents for us?
Yes, with engineers based in the UAE. Ask for a free estimate and get a scope, timeline and cost range within 48 hours.
Sources
- Vertex AI Agent Builder documentation (Google Cloud)
- More ways to build and scale AI agents with Vertex AI Agent Builder (Google Cloud blog)
- Introducing Gemini Enterprise Agent Platform, 22 April 2026 (Google Cloud blog)
- Gemini Enterprise Agent Platform name changes (Google Cloud documentation)
- Agent Development Kit (Google)
- Agent Runtime (Google Cloud documentation)
- Gemini Enterprise Agent Platform pricing (Google Cloud)
- Supported locations for agents in Agent Platform (Google Cloud documentation)
Birbal AI is an independent company based in Abu Dhabi. Facts about third parties come from the sources above; check them for the latest position.