Build an AI Agent with Temporal [video]

Senior Staff Solutions Architect Steve Androulakis uses Temporal to build a durable AI agent in this five-minute video.

The agent is first given a description of the goal and the tools it has at its disposal. The agent uses the LLM to decide which tool to use. If the LLM gives a bad response at any point, Temporal helps the agent self-heal by retrying the LLM until it succeeds. The agent allows for amending of parameters such as flight dates using natural language. The agent is not hard-coded to this use case. The Workflow supports dynamic agent goals and tool definitions, so it can essentially work toward any goal so long as the correct tools and goals are provided. Temporal handles the human-in-the-loop confirmation steps via signals. The React JS UI simply queries the Temporal Workflow to display the conversation history. The workflow is coded using Temporal’s Python SDK.