Google Launches Gemini Enterprise Agent Platform, Replacing Vertex AI

Google has folded Vertex AI into a new, rebranded product: the Gemini Enterprise Agent Platform. Announced at Google Cloud Next 2026 this week, it consolidates model access, agent development, governance, and security into a single surface — and it's the company's most direct answer yet to OpenAI's Agent Kit and Anthropic's Claude Agent SDK.

The launch is paired with a $750 million fund for partners building on the platform. Google clearly wants this to be the default place enterprises build AI agents.

What's new

The Gemini Enterprise Agent Platform is an evolution of Vertex AI, not a replacement that breaks existing workloads. Customers who were already using Vertex for model training and deployment will see the same primitives — plus a set of new capabilities built specifically for running agents in production.

The Model Garden now offers first-class access to more than 200 models, including Google's latest: Gemini 3.1 Pro, Gemini 3.1 Flash Image, and the Lyria 3 audio model. Third-party models from Anthropic, Meta, and Mistral are also available through the same interface.

The bigger change is what sits on top of the models.

Agent Identity: every agent gets a passport

The most distinctive feature is Agent Identity. Every agent created on the platform is assigned a unique cryptographic ID, and that ID is attached to every action the agent takes — every tool call, every database write, every API hit.

Why this matters: until now, when an AI agent did something wrong, the audit trail usually ended at a human user or a generic service account. Agent Identity makes the agent itself a first-class actor, so compliance teams can trace decisions back to a specific agent version and the policy that authorized it.

The IDs are designed to map back to authorization policies that are "traceable and auditable." For regulated industries like finance, healthcare, and insurance, this is the feature that unblocks production deployment.

Agent Gateway and Model Armor

Two more pieces round out the governance story. Agent Gateway acts as what Google calls "air traffic control" for agent-to-agent and agent-to-data interactions — it enforces which agents can talk to which systems, logs the traffic, and can block calls that violate policy.

Model Armor is a prompt-layer defense. It's aimed at the class of attacks that have become real problems over the last year: prompt injection, tool poisoning, and sensitive data leakage through model outputs. Model Armor sits in front of model calls and screens both inputs and outputs.

Neither feature is novel on its own — specialized vendors have offered similar things for a while — but bundling them into the default enterprise platform changes the default security posture for anyone building on Google Cloud.

Partner agents and the $750M fund

Alongside the platform, Google announced a catalog of partner-built agents available inside Gemini Enterprise. Capgemini, for example, unveiled a Google Cloud AI Enterprise Hub the same day, aimed at helping Fortune 500s deploy agentic AI without building everything from scratch.

The $750 million agentic AI partner fund is the incentive structure behind all of this. It's targeted at global consulting firms, systems integrators, and ISVs, and explicitly covers "agentic AI prototyping, agent building and deployment, and upskilling." Translation: Google is paying partners to make Gemini Enterprise the path of least resistance for enterprise buyers.

The competitive picture

This is Google's cleanest shot yet at the OpenAI and Anthropic enterprise stack. OpenAI's Agents SDK and Anthropic's Claude Agent SDK are both developer-first, opinionated frameworks. Google's pitch with Gemini Enterprise is different: don't write framework code, compose agents in a platform that handles governance, identity, and deployment for you.

Whether that resonates depends on the buyer. Engineering-led organizations tend to prefer SDKs. Enterprise IT, compliance, and security teams almost always prefer platforms. Google is explicitly targeting the second group.

Expert take

Industry analysts at The Next Web described the launch as Google's "full-stack bet against OpenAI and Anthropic" — chips, models, and platform in one coordinated announcement. Infosecurity Magazine highlighted the Agent Identity feature specifically, calling it an early signal that cryptographic agent identity could become a baseline requirement for enterprise AI.

There's also a practical read: Vertex AI had a reputation among developers as powerful but sprawling. Collapsing it into Gemini Enterprise with a cleaner surface area is also a usability bet.

What's next

Gemini Enterprise is available now to Google Cloud customers. Existing Vertex AI workloads keep running unchanged, and Google is providing migration paths for teams that want to adopt the new agent-specific features.

The features to watch over the next quarter are Agent Identity adoption in regulated industries, and whether the A2A (agent-to-agent) protocol Google is pushing inside the platform gets picked up by OpenAI or Anthropic. If it does, it starts to look like a real standard. If not, it's a Google-only play.

Bottom line

Google stopped trying to sell AI infrastructure as a set of parts and started selling it as a platform. For enterprises that have been stuck choosing between a half-dozen agent frameworks, the pitch is simple: pick one platform, get identity, governance, and 200+ models in the same place. If it ships cleanly, it will pull serious workloads off OpenAI and Anthropic's enterprise tiers.