Google’s new agent framework

Plus: A huge list of useful MCP servers, the 5 warning signs of agent over-engineering, and more

Author

Michael Cunningham
April 10, 2025

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

“Please stop with these corny jokes at the beginning of every issue. We’re trying to print serious news here.” —my co-founder agent

In today’s issue…

  • Google’s new multi-agent framework

  • Nvidia’s $2,000 agent sessions, available for free

  • A huge list of useful MCP servers

  • The 5 warning signs of agent over-engineering

  • More than half of companies already deploy agents

…and more

📰 NEWS

Source: Wikipedia

Agent Development Kit (ADK) is Google’s new open-source competitor to Microsoft’s AutoGen. Along with much of the functionality which is standard to other multi-agent frameworks, ADK comes with a variety of integrations to Google Cloud for easy deployment.

The company’s other major agent move was the release of Agent2Agent (A2A), a protocol for communication between agents from any vendor. A2A is a collaboration with an eye-popping list of top tech and consulting firms, clearly intended to stake Google’s claim in the race to establish standards for agent-agent and agent-tool communication.

A new agentic feature by the company can search Amazon—and even other websites if it is not available there—for a desired product, show options to the user for approval, and enter the user’s payment info to purchase it. It is currently being beta-tested with select users.

If you find Building AI Agents valuable, forward this email to a friend or colleague!

🛠️ USEFUL STUFF

Source: Nvidia

Videos on a wide range of agentic and enterprise AI topics from Nvidia’s GTC conference are now available online at no cost, provided viewers enter some information.

This curated list of model context protocol (MCP) servers gives builders the ability to integrate their agents with hundreds of different apps and services, from Slack and Teams to YouTube transcripts to Apple reminders, and more.

Stagehand is a framework by Browserbase for quickly building agents that intelligently automate tasks involving web browsing. A major new update adds several new functionalities, including integration with OpenAI and Anthropic’s computer use models.

A new suite of MCP servers by Amazon Web Services (AWS) give agents direct access to best practices, documentation, knowledge bases, and more, enabling users to create agents that intelligently build and control AWS systems.

“Handoffs” are a commonly used method for transferring control between agents in multi-agent frameworks such as OpenAI’s Agents SDK and LangGraph Swarm. In this video, LangChain CEO Harrison Chase shows what they are and how to use them.

💡 ANALYSIS & RESEARCH

Source: Freepik

This article makes the case that many current agent systems are needlessly complex, and provides excellent tips on how to recognize and avoid this trap.

Jacob Liberman, director of product management at Nvidia, discusses how agents are already beginning to be integrated into major enterprises’ operations.

The current agent paradigm involves slotting AI into existing processes to speed them up, but the author of this piece argues that the real value will be in finding entirely new processes that would have been impossible without them.

Less than a year after they went mainstream, over half of companies are using agentic AI in their workflows, according to a new survey, with 35% more planning to integrate them by 2027.

Agents represent a significant attack surface for scammers when not properly protected—and careless companies are already losing money.

Thanks for reading! Until next time, keep learning and building!

What did you think of today's issue?

Login or Subscribe to participate in polls.

If you have any specific feedback, just reply to this email—we’d love to hear from you

Follow us on X (Twitter), LinkedIn, and Instagram