---
title: "Meet Atlas, our automated testing agent"
slug: video-meet-atlas
section: Video Library
status: stable
description: "Every day, the software behind Globus, the platform that runs our programs, gets updated."
last-updated: 2026-10-07
---

# Meet Atlas, our automated testing agent

Video: Meet Atlas, our automated testing agent (2:03) — /docs/video-meet-atlas

- Answers: What is Atlas and how does automated testing work in Globus
- Audience: engineering; program scope: all programs
- Visibility: staff; last verified 2026-11-06; owner platform-team

Chapters:
- 0:00 Software changes every day — /docs/video-meet-atlas?t=0
- 0:27 Meet Atlas — /docs/video-meet-atlas?t=27
- 0:39 Step 1: people write the checklist — /docs/video-meet-atlas?t=39
- 0:55 Step 2: Atlas runs the checks — /docs/video-meet-atlas?t=55
- 1:11 Step 3: Atlas reports back — /docs/video-meet-atlas?t=71
- 1:26 What Atlas will never do — /docs/video-meet-atlas?t=86
- 1:46 Early but already useful — /docs/video-meet-atlas?t=106

## Transcript

Every day, the software behind Globus, the platform that runs our programs, gets updated. New features, small fixes, little improvements, and every single change carries a tiny risk. Something that worked yesterday might quietly break today. A form that won't save. A button that does nothing. Checking all of that by hand every single day would take a small army of people. So we built a helper. Meet Atlas. Our automated testing agent. Think of it as a very careful inspector. One that never gets tired. Here's how it works. First, people write the checklist. Our team describes, in plain language, what should happen. For example, when someone updates their phone number, it should be saved. And Atlas only ever uses the checks that a person has approved. Then Atlas gets to work. It opens a real web browser, signs in with its own practice account, and clicks through the platform just like a person would. It types, it submits, and then it double-checks that the information was really saved. When it finishes, Atlas reports back. It runs every weekday morning and again after each new update goes out. It posts a short summary to the team chat and keeps a full record with screenshots of exactly what it saw. Now the important part, what Atlas will never do. It only works in our practice environment. Never on live systems and never with real applicants' information. It can't approve, block, or release anything. If it finds a problem, that's only a suggestion. A person always makes the final call. Atlas is still early in its journey, but it has already found real bugs before they reached anyone. The idea is simple. The robot checks, the evidence speaks, and people decide. That's Atlas.