New Features Announced for Gemini Enterprise Agent Platform

Google Cloud unveiled new features for the Gemini Enterprise Agent Platform to make it easier and safer to scale enterprise AI agents. This update focuses on strengthening agents' long-running capabilities as well as their security and management capabilities.

The key features are as follows:

• Agent Memory Bank: Automatically extracts and maintains conversational context through a structured schema. This allows agents to remember user preferences and past decisions, providing a personalized experience without losing context.
• Agent Runtime: Enables agents to run continuously for up to 7 days. This allows complex asynchronous tasks, such as sales sequences or onboarding processes, to be performed without human intervention.
• Agent Identity: A new IAM type based on open standards that applies the principle of least privilege to agents. It prevents token theft and provides traceable auditing for all agent actions.
• Agent Gateway: A control point that centrally controls and manages all interactions within the agent ecosystem. It integrates with Model Armor to protect the system from prompt injection and data leakage.