Vertex AI Agent Builder

cloud.google.com/products/agent-builder now 301-redirects to cloud.google.com/products/gemini-enterprise-agent-platform, with the page's own title reading 'Gemini Enterprise Agent Platform (formerly Vertex AI)'. Google's documentation says it plainly: 'Agent Builder has transitioned to become part of Gemini Enterprise Agent Platform.' This was announced at Google Cloud Next on April 22, 2026 as a platform-wide rebrand of Vertex AI, not a cosmetic rename of one product — Vertex AI's ML tooling, the old Agentspace agent layer, and Agent Builder's search/conversation tools were all consolidated under one umbrella and one console. Existing API endpoints and workloads keep running unchanged, per Google's own migration guidance, so nothing breaks for current users — but anyone evaluating 'Vertex AI Agent Builder' today is really evaluating this larger platform, and any pricing or docs page still bookmarked under the old name will bounce you to the new one.

Google organizes the platform into four pillars. Build: Agent Development Kit (ADK), a low-code Agent Studio canvas, a prebuilt-agent library (Agent Garden), Model Garden access to 200+ models (Gemini, Anthropic's Claude, Gemma and others), RAG Engine, and Vector Search. Scale: Agent Runtime (the hosted execution environment, formerly called Agent Engine), Agent Platform Sessions and Memory Bank for state/persistent memory, and sandboxed code execution. Govern: Agent Registry, Agent Identity, Agent Gateway (policy enforcement on tool calls), and content/semantic governance policies. Optimize: agent evaluation tooling (Multi-Turn AutoRaters, online evaluation), simulation for stress-testing agent behavior, and an observability/trace viewer. This is a materially bigger surface than the old 'Agent Builder' branding implied — it's now Google's full agent stack, not just a search-and-conversation builder.

ADK (the code-first agent framework at the core of the 'Build' pillar) is Apache-2.0 licensed and developed in the open at github.com/google/adk-python: 21.2k stars, 3.9k forks, 434 contributors, 82 releases (latest v2.7.1), and commits landing within the hour of this check. That's a real, currently-active open-source project, not an abandoned SDK wrapped around a hosted API — a legitimate strength if vendor lock-in to a specific model or a black-box agent format is a concern, since ADK itself is model-agnostic and has sibling SDKs in Java, Go and TypeScript. The gap: ADK being open source doesn't make the platform open source. Runtime, Sessions, Memory Bank, Vector Search, Agent Gateway and everything else it talks to are closed, metered Google Cloud services — you can self-host the agent logic, but production deployment on this platform means a GCP billing account.

There's no flat subscription. Google's own pricing page (docs.cloud.google.com/gemini-enterprise-agent-platform, confirmed live 2026-08-20) bills 'Scale' pillar usage (Runtime, Sessions, Memory Bank, Skill Registry, Agent Gateway) on three meters: Agent Compute at $0.085/vCPU-hour after a 50-hour/month free tier, Agent Memory at $0.009/GiB-hour after 100 GiB-hours free, and Agent Storage at roughly $0.000411/GiB-hour (~$0.30/GiB-month) after 1 GiB-month free. Generative AI usage (text/chat/code) is billed separately starting at $0.0001 per 1,000 characters, and Vector Search/Pipelines have their own separate meters. New customers get the standard $300 Google Cloud trial credit — that's a general GCP incentive, not something specific to this product. One thing worth flagging directly: Memory Bank, Sessions, and Skill Registry billing has not started yet — Google's pricing page states it 'will commence on September 1st, 2026' — so anyone testing those specific features today is using them for free in a way that will not continue past that date, roughly ten days from this check.

G2 and Gartner Peer Insights both host reviews, but under a mix of old and new product names that are actively being reconciled — Gartner Peer Insights still lists the product as 'Vertex AI' with 340 reviews, and G2 has separate listings for 'Google Vertex AI' and 'Gemini Enterprise Agent Platform'. Both sites blocked automated access during this check (G2 returned a CAPTCHA wall, Gartner a Cloudflare challenge page), so an aggregate star rating could not be independently verified — not established. Buyers should read reviews directly on those platforms rather than trust a secondhand summary, and should check which product name a given review is actually about, since the rename has split historical review volume across two listings.