How to Build an AI Agent in UiPath (Step-by-Step Tutorial)

About this video

Kevin, in a video sponsored by UiPath, builds an AI agent from scratch that takes a request, evaluates options, and selects the best fit — a pattern he says also applies to routing support tickets, recommending products, or selecting options in a workflow. The example is a cookie recommender: for ‘I want something chocolatey, but not too sweet’ from options including chocolate chip, double chocolate, oatmeal raisin, and sugar cookie, it should pick chocolate chip. Starting in VS Code with Git Bash and Node.js version 18 or later, he installs the UiPath CLI (version 1.1.1 shown), installs the agent tool, optionally installs UiPath skills for coding tools such as Claude, Copilot, Cursor, Gemini, or Codex, and logs in, noting a free community edition is available. He creates a ‘Cookie Agent’ folder, a ‘Cookie Selector’ solution, and a project inside it, then configures agent.json and entrypoints.json, stressing the input and output schemas must match in both. After validating (result: success) and uploading, he opens the designer URL in Studio Web and tests via Debug: the chocolate request returns chocolate chip and ‘classic with raisins and oats’ returns oatmeal raisin. He also notes the same agent can be built visually in Agent Builder and orchestrated in larger workflows with UiPath Maestro.

Key takeaways

• AI agents in UiPath are framed for tasks requiring judgment — evaluating information and choosing between options — versus traditional automation for steps that are the same every time.
• Setup requires Node.js version 18 or later, installing the UiPath CLI and agent tool, and authenticating the CLI to a UiPath account before creating anything.
• An agent lives inside a solution: create a working folder, a solution, and a project inside the solution before configuring it.
• The agent is defined in agent.json and entrypoints.json, covering model settings, system prompt, and input/output schemas, and the schemas must match in both files.
• Validate from the command line before uploading to Studio Web, then test in Debug with sample JSON and refine the system prompt; the same build can also be done visually in Agent Builder.

Episode details: published July 2, 2026 · 11:16 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Kevin Stratvert’s own claims, shared for education — they are not typical or guaranteed.

← Previous: ChatGPT Tutorial for Beginners: How to Actually Get Work Done with AI · Next: Ep. 03 – AI Assistant in Adobe Creative Cloud →

All Kevin Stratvert videos · Kevin Stratvert profile · Kevin Stratvert on YouTube