You can no longer watch The Pragmatic Engineer Podcast as a video within the Spotify app. I have made the decision to quit publishing video on the platform. It was not a difficult call. It came after a series of reliability issues that made it clear that reliability is taking a back seat at Spotify.
The Outage Timeline
For the first eighteen months, things worked fine, albeit with some annoyances. The Creators portal was buggy and often required me to sign in via email code every single Wednesday. But from late May, things got worse. On three separate occasions within five weeks, the video processing pipeline failed.
Outage 1: The episode would not process for over two hours. The Creator portal showed NaN% values. I emailed the team. They confirmed it was a bug in the metadata pipeline and promised to improve communication.
Outage 2: Four weeks later, the entire platform had issues. Spotify does not maintain a public status page, so it was impossible to know the severity. I did not include a Spotify link in that week's announcement.
Outage 3: The publishing pipeline failed again. After two hours of waiting, I sent the announcement without a link. I told the team I was considering leaving. I asked for the incident review.
The Response
They apologized, but the incident review never arrived. I waited three weeks. Nothing. No transparent post-mortem. No improvement in communication.
When I finally offboarded from Spotify to an RSS feed, the Creators portal became even buggier, showing errors all over the dashboard. It was as if no one had tested the offboarding flow. Eventually, a few days later, I received the incident report for Outage 3. But the timeline did not add up. It claimed the team was alerted internally, but my email confirmed I had notified them 17:30. I complained, and they updated the report.
The "AI Psychosis"
This experience highlights a larger issue. Spotify's leadership is focused on AI adoption. They recently announced they ship 4,500 production deploys a day and that 73% of PRs are AI-assisted. Engineers are running 5 to 10 Claude sessions at once.
This is the definition of "AI psychosis" in the corporate sense. It is the rush to adopt AI tools at the cost of reliability and user experience. I met Spotify's Head of Technology in March, who told me reliability was a priority. But the actions show otherwise.
Max Kanat-Alexander, a distinguished engineer at Capital One, once wrote about how software succeeds by "sucking less" with every release. If you release frequently but only add new features instead of fixing what sucks, users will eventually leave. I got tired of the poor reliability, the errors, and the sense that Spotify just does not care about improving the product for creators.