Google’s Managed Agents API promises single-call implementation at the expense of artificial layer management
At Google I/O, the company unveiled Managed Agents in its Gemini API — a service that promises to collapse weeks of agent deployment work into a single API call. It’s also a sign that Google believes its ecosystem, including the recently introduced Antigravity CLI, is ready to own the ultimate rendering layer.
Before a single agent is written, teams are already spending days on unpleasant work: setting up kill zones, managing sandboxes, infrastructure for wiring tools. Model providers like Anthropic have introduced platforms to handle much of that work — but Google’s approach is different.
Google said in a blog post that the Managed Agents in Gemini API summarizes “remove complexity so you can focus on your product experience and agent behavior.” This service is available in preview with new custom templates in Google AI Studio.
The growth presented a real architectural question: should agent management reside in the abstraction layer – embedded in the model or its harness – or in the infrastructure layer, as a separate runtime?
Comparing the Google method
Until recently, agent orchestration has relied on frameworks that sit above the model, directing agents and allowing teams to manage routing and execution separately. That layer is now being taken over by the platforms themselves.
Recent platforms such as Claude Managed Agents embed orchestration in the model layer instead of a separate runtime platform. The idea is that the model owns the logic and orchestration layers, and businesses have control over the execution.
AWS, with new capabilities in Bedrock AgentCore, adds a managed harness that brings together the previous functions of dispatch agents. Google’s approach goes further, developing the model, wires, and sandbox together and running everything in a secure environment managed by Google.
Ramp’s René Sultan, quoted in Google’s announcement, said the trend is real: "The real change with Gemini Managed Agents is that the agent runtime is moving to the platform. With a sandbox, infrastructure and execution loop managed by you, developers can focus on generating domain-specific behavior for the agent and iterating at a completely different pace."
The new reality of orchestration
Businesses starting out with agents may find the platform offerings from Anthropic and Google strong, especially since they remove much of the complexity of deploying agents while retaining some control. Google, however, is aiming for an integrated system upwards, while Anthropic is betting on the model layer as the orchestration plane, and AWS is focusing on authorization.
But this comes with some risks, according to XYO founder and CEO Arie Trouw.
“An additional risk is that developers will change what used to be deterministic services to what will now be possible services, which can present unintended consequences for users, or serious data corruption,” Trouw told VentureBeat in an email. “This is a classic example of having an amazing hammer and everything starts to look like nails. I’ve seen this pattern over and over again as an engineer and business founder myself over the past few decades.”
