Google DeepMind announced expanded capabilities for Gemini API's Managed Agents, introducing support for the Gemini 3.6 Flash model alongside a new hooks framework designed to connect external applications and services. The addition of hooks represents a significant developer convenience layer, allowing agents to securely invoke third-party APIs and integrate with business tools without building custom middleware. For instance, a Managed Agent could now directly interact with enterprise software like Salesforce or Slack through pre-defined hooks, reducing integration work that previously required boilerplate code. The 3.6 Flash variant offers faster inference speeds and lower costs compared to larger Gemini models, making it practical for resource-constrained deployments. Google is positioning these updates as production-ready, addressing long-standing pain points around agent reliability and deployment complexity.
The hooks framework integrates with Google's broader push to embed AI capabilities across its consumer products. Google Search's AI Mode now accepts connections to external applications, allowing developers to build agents that answer queries by pulling real-time data from connected services. This creates a two-way ecosystem: developers get better tooling to build agents, while Google gains richer integrations that make Search's AI features more powerful. A practical example would be a travel booking agent that connects through hooks to airline and hotel reservation systems, enabling users to check availability and complete bookings within Search's interface. This approach mirrors but differs from OpenAI's agent implementations, which rely more heavily on function calling at the model level rather than a managed platform handling orchestration.
The competitive landscape for agentic systems remains fragmented. While OpenAI has emphasized model-level function calling and reasoning capabilities, Google's strategy emphasizes managed infrastructure and ecosystem integration, particularly leveraging its Search dominance. The Managed Agents platform abstracts away orchestration complexity, appealing to enterprise developers who need reliability guarantees. Combined with Search integration announcements and expanded app connections, Google is betting on developer adoption through integrated tooling rather than raw model capability. Meta remains largely absent from this announcement cycle, continuing its focus on Llama model releases for on-premise and cloud deployment rather than managed agent platforms.