At Google Cloud Next, Google relaunched its enterprise AI stack as Gemini Enterprise—an end-to-end system for building, deploying, governing, and using autonomous agents at scale. The headline piece is the Gemini Enterprise Agent Platform, an evolution of Vertex AI that bundles models, development tools, runtime, identity, security, and observability under one control plane.
The portfolio splits along three audiences: Agent Platform for developers, the Gemini Enterprise app for knowledge workers, and an Agent Gallery for discovering and deploying validated third-party agents from partners including Adobe, Oracle, Salesforce, ServiceNow, and Workday.
The Agent Platform: Build, Scale, Govern, Optimize
Google organizes the platform around four verbs:
Build
An enhanced Agent Development Kit (ADK) with a new graph-based framework for organizing agents into networks of sub-agents, plus richer ecosystem integrations and full MCP support. Low-code Agent Studio lives alongside ADK so technical teams and non-developers can ship to the same runtime.
Scale
An improved Agent Runtime tuned for fast, persistent execution; agent-to-agent (A2A) orchestration for delegating tasks between agents; and a new Memory Bank plus Memory Profiles that give agents long-term contextual memory across sessions—recalling user preferences and prior history without re-priming on every run.
Govern
This is where Google leans hardest. Every agent gets a unique cryptographic Agent Identity for traceability and auditing. Agent Gateway acts as air traffic control between agents and data sources. Model Armor defends against prompt injection, tool poisoning, and sensitive-data leakage. The pitch: the same level of oversight and auditability you expect in payroll or quarterly financial reporting, applied to agent fleets.
Optimize
Agent Simulation lets teams stress-test agents against real-world scenarios before they ship. Agent Observability provides a real-time view into safety and performance once they're live.
The Gemini Enterprise App: a Front Door for Knowledge Workers
Anything built on Agent Platform surfaces instantly in the Gemini Enterprise app. New capabilities include:
- Agent Designer — natural language creation of schedule- and trigger-based agents, including those triggered by other business apps via partner connectors.
- Inbox — a unified hub for monitoring active agents (especially long-running ones), adjusting configurations, and receiving real-time status alerts via email and chat.
- Projects — a shared workspace where teams and agents accumulate persistent memory on a topic, so context outlives any individual team member.
- Canvas — co-editing in Google Docs and Slides, with new Microsoft 365 interoperability for Office-format export.
The Agent Gallery: Validated Third-Party Agents
Rather than push every customer to build from scratch, Google is bringing the full Google Cloud Marketplace catalog of partner-built agents directly into the Agent Gallery inside the app. Each one is validated by Google Cloud for security and interoperability, and surfaces through a request-and-approval gateway that lets employees ask for what they need while IT signs off. Salesforce, ServiceNow, Adobe, Workday, and Oracle agents are explicitly named as launch partners.
How the Pieces Fit Together
Google's framing is that you no longer need to choose between a developer platform and a knowledge-worker product. Pro-code agents from ADK and no-code agents from Agent Designer are both managed under the same identity, security, and audit model. IT gets a single control plane; developers get the lower-level building blocks; everyone else gets a shareable workspace.
For interoperability, Google is leaning on open protocols: A2A for agent-to-agent communication and MCP for tool and data integration. The platform is positioned as multi-AI rather than single-vendor, which matches Google Cloud's broader Next '26 messaging on multicloud and multi-AI strategy.
Why It Matters for Web Developers
The Gemini Enterprise relaunch is the clearest sign yet that the enterprise agent market is consolidating around governance + runtime + marketplace as the core unit, not standalone models or standalone IDE tooling. For web developers building agent products, the practical implications are concrete: shipping into the Agent Gallery is now a viable distribution channel; ADK's graph-based framework is a real alternative to LangGraph for sub-agent orchestration; and Memory Bank means you can stop building per-app memory layers if you're already on Vertex. For internal tools, the combination of Agent Identity, Agent Gateway, and Model Armor gives security teams enough surface area to actually approve agent rollouts—which has been the bottleneck on enterprise agent adoption for the past year.