Google DeepMind is making a decisive move to claim territory in the enterprise agent market. The company announced new Managed Agents capabilities in the Gemini API this week, introducing background task execution and remote Model Context Protocol (MCP) support—features designed to let developers build reliable, production-grade autonomous systems without managing infrastructure. The timing is strategic: as OpenAI accelerates its agent narrative around GPT-4o and Anthropic's Claude gains traction in tool-use scenarios through MCP adoption, Google is positioning Gemini as the infrastructure backbone for developers who need agents that can run asynchronously, handle complex workflows, and integrate with third-party services at scale. This announcement represents Google's clearest signal yet that it's treating the agent layer not as a consumer feature but as a platform play.
The Managed Agents framework lets developers offload the operational burden of agentic AI. Instead of provisioning containers or managing state, developers can invoke agents through simple API calls; Google handles scaling, error recovery, and execution monitoring. The remote MCP support is particularly significant because it signals Google's willingness to work within an open standard ecosystem rather than forcing proprietary lock-in—a subtle but crucial competitive move against Anthropic's emerging dominance in the tools-and-context space. Developers can now orchestrate complex multi-step workflows where an agent schedules background jobs, waits for external APIs to respond, and chains results together without human intervention. Simultaneously, Google's expansion of third-party app connections in AI Mode Search—allowing users to securely link Gmail, Drive, and other services—creates a consumer-facing complement to the API layer, letting end users access agent-like capabilities without code.
What's notably absent from Google's messaging is clarity on how Gemini, Vertex AI, and various DeepMind projects fit into a unified story. Industry observers note that Google's fragmented AI go-to-market strategy—spreading capabilities across consumer Search, enterprise APIs, and research divisions—remains a vulnerability when competing against OpenAI's singular focus or Anthropic's cohesive positioning. Pricing and availability details for Managed Agents suggest Google is betting on developer adoption as the primary lever, but whether that's enough to shift momentum away from competitors who've already shipped agent interfaces remains an open question. The real test will be whether enterprises actually adopt this infrastructure or continue building on top of existing LLM APIs they already understand.