Google's Gemini Enterprise Agent Platform Merges Agent Building, Runtime and Governance

Google Cloud has introduced Gemini Enterprise Agent Platform, a product that pulls AI agent creation, deployment, runtime operations and governance into a single surface. According to a dev.to report on the April 23, 2026 announcement, the platform is an evolution of services previously tied to Vertex AI, and it signals a change in how Google sells Gemini to businesses: less as a model that answers prompts, more as a stack for agents that carry out defined work across company systems.

What the platform includes

Google describes three connected areas: building agents, running them with context and tools, and observing and controlling their behavior after deployment.

Agent Studio offers a low-code interface for constructing agents, while an upgraded Agent Development Kit serves developers who prefer code. A reworked Agent Runtime is built for long-running, stateful agents, and Memory Bank gives those agents persistent context, meaning they can retain relevant information across a multi-step task rather than treating every interaction as brand new.

Around those building blocks sit operational tools: Agent Identity, Agent Registry, Agent Gateway, Agent Simulation, Agent Evaluation, Agent Observability and Agent Optimizer. Google presents these as the identity, security, governance and evaluation groundwork required to run agents at scale. A Gemini Enterprise app acts as an employee-facing front end, letting organizations deliver configured agents to staff in a controlled environment.

Model access through Model Garden

Model Garden anchors the platform's reach. Google lists Gemini models including Gemini 3.1 Pro, Gemini 3.1 Flash Image, Lyria 3 and Gemma 4, alongside third-party options such as Claude Opus, Sonnet and Haiku. As dev.to notes, that multi-model catalog matters for teams that want to benchmark models against a particular task instead of tying an entire application to one model family.

Part of a wider Gemini rollout

The Agent Platform arrives alongside a set of related updates. A Gemini API public preview through Google AI Studio includes Gemini 2.5 Pro Preview, with billing enabled in some previews and, in certain cases, a route into Vertex AI. Antigravity 2.0 adds a desktop application, a command-line interface and an SDK for managing and deploying agents across Google AI Studio, Android and Firebase, with Managed Agents in the Gemini API and an integration path into the Agent Platform. There is also native Android support in AI Studio, an Interactions API for Gemini, Workspace integration, a Gemini app for macOS, and Gemini features in Gboard on Android.

Google has further signposted a planned Gemini Enterprise for Customer Experience line, with CX Agent Studio, Agent Assist, CX Insights and Commerce agents, pointing at employee support, customer interaction and commerce as the areas where it expects demand.

What remains unresolved

The announcements do not fully specify timing, regions or exact availability for every component, and pricing is unclear across products and tiers, particularly for smaller organizations. The dev.to report also cautions that the platform does not remove the need to define reliable processes, permissions and human oversight before an agent enters a customer or operational workflow. A sensible early evaluation, it suggests, establishes what data an agent may access and what it must not, which actions require human approval, how incorrect output will be tested and handled, and whether a low-code build, API integration or managed agent approach fits the workflow in question.

Why it matters

The launch marks a shift in how enterprise AI is packaged. Google is competing less on raw model quality and more on the surrounding plumbing — identity, observability, evaluation and governance — that determines whether agents can operate inside real businesses. Consolidating those controls into one platform lowers the barrier between isolated AI experiments and repeatable, governed workflows, and puts Google in direct competition with other agent platform vendors chasing the same enterprise budget. The caveat is equally significant: persistent, tool-using agents that act across company systems raise the stakes for permissions and oversight, and buyers currently lack confirmed pricing and rollout details. For most organizations, the practical response is a narrow pilot on a well-defined workflow rather than a broad program designed around features that may not yet be available in a given region or plan.