MyLabelDesk is a B2B platform for record labels. It had Sentry running in production, and Sentry was quiet. Meanwhile audio uploads outside the creation wizard never reached the server, and the export buttons did nothing when clicked. Neither bug threw an exception. Our browser agents found both, and we nearly dismissed the first report as the agent's mistake.
The setup
The platform is a workflow tool for the music industry. Its core flow spans 11 steps, from track upload through internal review, signing decisions, artist intake, contract management, and distribution to services like Spotify and Apple Music.
It is exactly the kind of app that is hard to test: multi-step workflows with role-based access, real-time updates, file uploads to cloud storage, and integrations with three external services. The team had Sentry set up and it was catching runtime errors.
What we ran
We ran 25 AI agent test sessions with Aiqaramba. Each agent got a plain-language description of a scenario, such as "create a track without a file, then add the audio from the track page". No selectors, no scripts. The agents worked through the app in Chrome and Firefox and reported what happened after each action.
Fifteen journeys covered the full 11-step workflow end to end, most of them run twice across Chrome and Firefox. Across those 25 sessions the agents reported 20 bugs. Two of them were blockers.
Bug 1: uploads that never left the browser
The first report on this one confused us. An agent had created a track, gone to the detail page to attach the audio file, and reported that the upload "completed without visible result". We read that as the agent being sloppy about checking its work, and nearly filed it as a prompt problem.
Then a second agent, in Firefox, reported the same thing. A third had opened the browser console and pasted the actual error: the request to the upload endpoint was blocked by the browser's CORS policy. The backend function that hands out upload URLs did not answer the browser's preflight request, so the browser refused to send the real one.
From the user's side it looked like this. Click upload. Nothing happens. No error, no spinner, no file. Click again. Nothing. Eventually you assume you did something wrong, or you leave.
Uploading inside the creation wizard worked fine, and that is why nobody had noticed. A person testing by hand goes through the wizard. Every session that added audio afterwards hit the block.
Sentry stayed quiet because the request never reached the server. There was no exception to report.
Bug 2: export buttons that did nothing
The distribution step has export buttons that send track packages to distributors. When agents clicked them: no download, no success message, no error, no loading indicator. The buttons accepted the click and nothing else happened.
The likely cause was that the test tracks lacked metadata the export needs, such as an ISRC or artwork. Requiring that is fair, but the buttons gave no hint of what was missing, so a label manager with a release date had no way to find out. A separate "Export data" button in the settings returned a 400 error.
The distribution export buttons themselves threw nothing for Sentry to catch.
Where error monitoring stops
Sentry, Datadog and Bugsnag are good at what they do: reporting errors the code throws. Both critical bugs here were failures in which nothing threw. The user clicked, and the product silently did not do the thing.
The grey zone is the Supabase 406 errors. They showed up in the network requests of 10 of the 25 sessions, and Sentry could have caught them with the right instrumentation. Nobody had set that up. The app kept working with degraded data, so without someone clicking through the affected screens they went unnoticed.
A quiet error dashboard meant nothing was crashing. It said nothing about whether a label could get a track to a distributor.
The full breakdown: 20 bugs by severity
- 2 critical: the CORS upload block and the dead export buttons
- 1 major: no user feedback at all when an upload fails
- 5 moderate: including the Supabase 406 errors across 10 sessions, a broken settings section, and review states that only updated after a manual page refresh
- 12 minor: confusing placeholder text in forms, inconsistent search results, and UI elements that did not respond on the first click
What made the agents effective here
Whole workflows. The scenarios followed the 11-step flow, including actions that depend on earlier steps. The upload bug only appears if you create the track first and add the file later. A unit test does not do that, and a person clicking through by hand rarely does.
Two browsers at once. Chrome and Firefox ran in parallel, so the second browser cost no extra time. It also gave us the second report that made us take the first one seriously.
Reading the console. The agents see the page, but they can also read the browser console, network responses and DOM state. On the upload bug the page showed nothing. The console showed the blocked request, the endpoint and the error.
The result
After the run, the team had a prioritised report with reproduction steps, console output and screenshots for each issue. The two critical bugs were fixed within days. Both had been live in production with real users working around them.