Google launched the Gemini Enterprise Agent Platform at Cloud Next 2026 on April 23, 2026. It is the most complete enterprise-agent stack Google has shipped, built on top of Vertex AI and deeply integrated with Google Cloud data and security.
The six components
- Agent Studio: low-code and pro-code agent authoring with versioning.
- Agent-to-Agent Orchestration: agents can call other agents, passing scoped context and tool access.
- Agent Registry: catalogs deployed agents; permissioned discovery; enforced lifecycle policies.
- Agent Identity: cryptographic identity for each agent. Supports audit, access control, and cross-tenant scoping.
- Agent Gateway: single managed entry for external invocations, rate limiting, and policy enforcement.
- Agent Observability: tracing, step-level logging, and cost attribution per agent run.
Why the stack matters
Frontier-model access is now a commodity for enterprises. The differentiator is what you wrap around the model: identity, registry, policy, audit, observability. Google is the first hyperscaler to ship all six as a single integrated product rather than six separate services.
Comparison:
| Stack | Identity | Registry | Orchestration | Observability |
|---|---|---|---|---|
| Gemini Enterprise Agent Platform | Yes (native) | Yes | A2A | Full tracing |
| AWS Bedrock Agents | IAM, partial | No formal registry | Limited | CloudWatch-based |
| Azure AI Foundry Agents | Entra ID | Yes (model catalog) | Limited | Azure Monitor |
| Anthropic Claude Code + Skills + MCP | Per-session | No registry | Tool-level | Session-level |
Google’s advantage is platform cohesion. The trade-off: vendor lock, since the Platform is cloud-native to GCP.
Integration with TPU 8i
Agent workloads are latency-sensitive. Gemini Enterprise Agent Platform is expected to progressively migrate to TPU 8i as capacity comes online. The stated design target of serving “millions of concurrent agents” on 8i-class silicon lines up with the agent-platform pitch.
Pricing
Not fully disclosed at launch. Usage-based billing across:
- Gemini model inference (per token, standard Vertex AI rates).
- Agent Platform service fees (per 1M agent-run events).
- Storage (registry, observability data).
Expect competitive pressure on per-agent-run pricing as AWS and Azure respond.
Ecosystem impact
- Agent SIs and consultancies: $750M Google partner budget directly subsidizes SI resellers of Gemini Agents through 2027.
- Wiz integration: Agentic Defense ships the first security-agent implementation on the Platform.
- Enterprise customers already on Vertex AI: upgrade path is mostly config, not net-new deployment.
Related
Sources
Primary and corroborating references used for this news item.