Google Unveils Agent Development Kit for Rapid AI Agent Prototyping and Deployment in Enterprises

Over the past year, businesses have witnessed a surge in platforms designed for creating AI agents, often requiring minimal coding. As agentic ecosystems from various organizations continue to grow, it's no surprise that major model providers are now developing comprehensive platforms for agent creation and management.

In response to this trend, Google recently announced an expansion of its agentic offerings, positioning itself to compete with other agent-building platforms. Google's new Agent Development Kit (ADK) and additional features aim to provide users with greater control over agent behavior.

Google's ADK is designed to streamline the development of multi-agent systems on Gemini models. The company boasts that users can "build an AI agent in under 100 lines of intuitive code" using ADK. The platform also supports the Model Context Protocol (MCP), a data connection protocol developed by Anthropic, which standardizes data movement between agents.

Google claims that ADK will enable organizations to:

• Influence how agents think, reason, and collaborate through orchestration controls and guardrails
• Engage in human-like conversations with agents, thanks to ADK's unique bidirectional audio and video streaming capabilities
• Kickstart development with a collection of ready-to-use sample agents and tools
• Select the most suitable model for their agent from Google's Model Garden
• Choose the preferred deployment target, whether it's Kubernetes or Google's Vertex AI
• Deploy agents directly to production via Vertex AI

While ADK is optimized for Gemini models, Vertex AI provides access to models from other providers such as Anthropic, Meta, Mistral, AI21 Labs, CAMB.AI, and Qodo. Google states that developers can use ADK to ground agent and application responses to various data connectors.

In addition to ADK, Google introduced Agent Engine, a managed runtime dashboard that complements ADK and offers enterprise-grade controls.

During a briefing with reporters, Google explained that Agent Engine enables organizations to progress from concept to training and ultimately to production. It manages "agent context, infrastructure management, scaling complexities, security, evaluation, and monitoring."

Agent Engine can be integrated with ADK but is also compatible with other frameworks like LangGraph or CrewAI. With support for both short-term and long-term memory, users can maintain context for agents and customize the amount of information from past interactions that agents can access.

Agent Engine also allows enterprises to assess the agents' behavior and performance during real-time usage.

For companies seeking additional assistance in building agents, Google launched Agent Garden. Similar to a model garden, Agent Garden is a library of pre-built agents and tools that users can leverage to model their own agents.

One of the primary concerns for many organizations when it comes to agents is security and trust. To address this, Google's ADK and Vertex AI offer various configurations for enterprises, including:

• Controlling agent output with content filters, defined boundaries, and prohibited topics
• Implementing identity controls with agent permissions
• Setting secure parameters to limit sensitive data access and prevent leaks
• Establishing guardrails, such as screening inputs before they reach the models that run agents
• Enabling auto-monitoring of agent behavior

The launch of ADK and Google's other agentic AI offerings intensifies the competition among agent platform providers. While enterprises could previously build agents using Google's AI services, the introduction of these new tools positions Google more directly against other agent providers. As technology companies increasingly offer all-in-one agent-building platforms, Google must demonstrate that its comprehensive solution, optimized for Gemini models and Vertex AI, is the superior choice.

In March, OpenAI released Agents SDK, which allows users to build agents using open-source tools and non-OpenAI models. Agents SDK also provides configurable enterprise security and guardrails. Amazon's Agents on Bedrock, launched in 2023, enables organizations to create agents in a single platform, and Bedrock was updated last year to include orchestration capabilities.

Emergence AI, a newcomer in the field, recently introduced an agent builder platform that allows users to create any AI agent they need on the fly. Users simply specify the task they need to accomplish, and AI models generate the agents to complete it.