At Google I/O, the company unveiled managed agents in its Gemini API — a service that promises to condense weeks of agent deployment work into a single API call. It also signals that Google believes its ecosystem, which also includes the newly launched AntiGravity CLI, is ready to deliver end-to-end ownership of the execution layer.
Before a single agent has been written, teams are already spending days on non-glamorous work: setting up execution environments, managing sandboxes, wiring tool call infrastructure. Model providers like Anthropic have launched platforms to handle that work — but Google has a different approach.
Google said in a blog post that managed agents in the Gemini API “remove complexity so you can focus on your product experience and agent behavior.” The service is available in preview through a new custom template in Google AI Studio.
The development has introduced a real architectural question: should agent management live at the execution layer – embedded in the model or its harness – or at the infrastructure layer, as a separate runtime?
Comparing Google’s Approach
Until recently, agent orchestration relied on frameworks that sat on top of the model, directing agents and letting teams control routing and execution separately. That layer is now being absorbed by the platform itself.
Recent platforms such as cloud managed agents embed the orchestration at the model layer rather than on a separate runtime platform. The idea is that the model owns the logic and orchestration layers, and enterprises have control over the execution.
AWS, through new capabilities on Bedrock AgentCore, adds managed harnesses that tie together advance functions to deploy agents. Google’s approach goes even further, customizing models, harnesses, and sandboxes together and running everything in a secure Google-managed environment.
Ramp’s René Sultan, quoted in Google’s announcement, said the change is solid: "The real change with Gemini managed agents is that the agent runtime moves into the platform. With the sandbox, infrastructure, and execution loop managed for you, developers can focus on producing the agent’s domain-specific behavior and iterating at a completely different pace."
new orchestration reality
Enterprises starting out with agents may find Anthropic and Google’s platform offerings stronger, especially since they remove much of the difficulty of deploying agents while maintaining some control. However, Google is emphasizing a more vertically integrated system, while Anthropic is betting on the model layer as an orchestration plane, and AWS focuses on authorization.
But according to Ari Trouw, founder and chief executive of XYO, it also brings some risks.
“An additional risk is that developers will turn what were previously deterministic services into probabilistic services, which could lead to unexpected consequences for users, or in the worst case, data corruption,” Trouw told VentureBeat in an email. “This is a classic example of having a wonderful hammer and making everything look like a nail. I’ve seen this pattern over and over again as a developer and business founder over the last few decades.”
<a href