Build Your First UiPath Automation (AI + RPA + Human Approval)

About this video

A long step-by-step build of a first UiPath agentic automation in Studio Web (free Community Edition): an invoice approval process combining RPA, an AI agent, and human approval. Kevin outlines the platform — Studio for building, Agents/Agent Builder, Maestro for orchestration, and Orchestrator as the post-publish control room — and notes the platform was named one of Time’s Best Inventions of 2025 (creator’s framing). The build: (1) a start event using a Google Drive connector event (’new file created’) watching an Invoices folder, filtered to PDF file extension, outputting file name, ID, and URL variables, verified with Debug (F5); (2) an RPA task ‘Download and extract invoice details’ with a file-URL input and four outputs (invoice date, number, amount, vendor name) — it uses Get File/Folder, then Document Understanding (enabled once via Admin Center > tenant > Services; part of UiPath IXP) with the predefined invoices extractor, assigning extracted values to the outputs, including fixing a data-type error by switching invoice amount to ’number with decimal’; (3) an autonomous ‘invoice decision agent’ built in Agent Builder with matching inputs/outputs (status and reason), a system prompt encoding the approval rules and a user prompt with curly-brace placeholders, tested with evaluation sets — a $480 invoice expected to be approved as within limit and a $10,000 invoice expected to be flagged Review as over limit (the auto-approval threshold applied in testing is $500); (4) a gateway branching on agent status: approved invoices are moved via a Google Drive connector activity to an Approved folder, while Review invoices go to a human-in-the-loop Action App task (simple approval template fed the invoice details, status, and reason); a second gateway routes the human’s decision — approve moves the file to Approved, reject moves it to a Rejected folder. He end-to-end tests both paths (enabling Actions/Action Center first for the review path), then publishes the solution to Orchestrator, where it runs automatically whenever a new invoice lands in Drive, with Maestro available to monitor performance.

Key takeaways

• An agentic process combines three layers: RPA robots for deterministic work, AI agents for rule-based reasoning, and human-in-the-loop approvals for exceptions
• Trigger processes from connector events (e.g., a new PDF in a Google Drive folder) and pass data between steps with typed input/output variables
• Document Understanding (IXP) with the predefined invoices extractor pulls date, number, amount, and vendor from PDFs without custom training — enable it once per tenant
• Test agents before publishing with evaluation sets: sample cases with expected status/reason (e.g., $480 approve vs. $10,000 review) verify consistent decisions
• Gateways branch the flow on agent status or human action; publishing deploys the whole solution to Orchestrator to run, schedule, and log automatically

Episode details: published December 18, 2025 · 53:24 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: The Future of AI - Inside The Studio w/ Kevin | Episode #1 · Next: Grammarly Tutorial: How to Improve Your Writing Instantly →

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