Chris demonstrates how he built a powerful AI agent for his daily planning app, Ellie, in just 48 hours by combining large language models with specialized tools and automation platforms like Zapier MCP. The agent efficiently manages tasks, calendars, and multi-user coordination through a chat interface, showcasing how accessible and practical AI agents have become for real-world applications.
In this video, Chris demonstrates how he built a useful AI agent for his daily planning app, Ellie, in just 48 hours. Ellie combines to-do lists and calendars, and the AI agent can interact with it through a chat interface. The agent can perform tasks like finding the best time for events by checking multiple calendars, creating tasks, and even time-boxing activities. Chris emphasizes that the agent was built with surprisingly little code and explains the process step-by-step, making it accessible for developers and non-developers alike.
The core concept behind the agent is that it uses a large language model (LLM) combined with a set of tools specific to Ellie, such as creating tasks, searching calendars, and updating tasks. The agent operates in a loop where it decides whether it needs to call a tool to fulfill a user request, executes the tool, and then uses the results to respond. Chris walks through how he implemented this by first creating a simple chat UI, then defining tools with clear descriptions and parameters, and finally coding the logic to execute these tools and handle the interaction loop safely.
Chris progressively added more complex tools to the agent, including searching and updating tasks, and accessing calendar events from Google, Outlook, and Apple calendars. He also tackled multi-user coordination, enabling the agent to find mutual availability between two users by refining the system prompt to guide the agent’s workflow. Although some tasks like time-boxing the day were challenging for smaller models, Chris overcame this by integrating more powerful LLMs as specialized tools, showcasing how chaining LLMs can handle complex workflows.
A major breakthrough came when Chris integrated Zapier MCP, an automation platform that connects thousands of apps, into the agent. This allowed the agent to access external tools like Slack without building direct integrations. For example, the agent can search Slack messages and create tasks in Ellie based on those messages, or send summaries back to Slack channels. This integration significantly expanded the agent’s capabilities with minimal additional code, making it far more powerful and versatile.
Chris concludes by reflecting on how AI agents have become practical and affordable enough to deploy in real-world applications. He plans to open a closed beta for Ellie users to test the agent and eventually bring it to the iPhone app with voice capabilities. He encourages developers to build AI agents for their own apps and hopes the video demystifies the technology for everyone. Overall, the project showcases how combining LLMs with tool-based architectures and automation platforms can create highly functional AI assistants quickly and efficiently.