Building custom tools for agents | The Agent Factory Podcast

How do you build an agent that can actually do things? This episode of The Agent Factory is a masterclass in crafting custom tools—the essential bridge between your agent's AI brain and the real world. Forget generic APIs; we're diving deep into the code and architectural patterns you need to give your agents specialized skills. Join hosts Smitha and Amit Maraj as they build "FinWise," a powerful financial agent from the ground up, demonstrating exactly how to design, build, and deploy a suite of custom tools for a complex task. 🛠️ By the end of this episode, you'll have a practical guide to the four essential patterns for building more capable and reliable agents. In this episode you will learn: 1️⃣ Agent industry pulse: We'll quickly cover the latest trends that make powerful custom tools possible. 2️⃣ We explore how Agentic RAG allows agents to intelligently fetch live data, look at new models like Kimi K2 and Qwen 3 that are optimized for tool use, and introduce new frameworks from Google like the GenAI Processors library and the experimental Opal tool that simplify how developers build these sophisticated systems. The Factory Floor The heart of this episode is the practical build. We'll walk you through the creation of our "FinWise" agent to showcase four essential and distinct patterns for creating custom tools with the Agent Development Kit (ADK): - Simple function tool: The "go to" pattern for creating basic, stateless tools like a real time stock price checker. Perfect for direct API calls. - Agent as a tool: A clever architecture for complex tasks. Learn how to grant an agent specialized skills (like Python code execution) by wrapping a "specialist agent" as a reusable tool. - MCP (Model Context Protocol) tool: The ideal pattern for heavy, computational services. See how to build a portfolio optimizer as a standalone, discoverable server that any agent can connect to. - Authenticated tool: The critical pattern for security. Learn to build tools that h