With the recent announcement of Agent Compute Engine, I'm curious about Google's long-term plans for Gemini Enterprise Agent Runtime (formerly Vertex AI Agent Engine).
One concern I have with the current Agent Runtime architecture is that sessions are logically separated, but the underlying compute environment and agent identity can be shared across users.
This makes secure multi-user agent design challenging, especially when users have different permissions.
You still need to enforce authorization and session isolation at the application or tool layer, rather than relying on infrastructure-level isolation.
In contrast, Amazon Bedrock AgentCore Runtime and Google's new Agent Compute Engine seem to take a more session-centric approach to compute isolation.
This raises a few questions:
Is Google planning to integrate Agent Compute Engine into Agent Runtime?
Will Agent Runtime eventually support per-session compute isolation?
Or are these intended to remain separate products for different use cases?
I…