On February 12, Spotify co-CEO Gustav Söderström said during an earnings call: “Our best engineers haven’t written a line of code since December.”
He added that AI has significantly accelerated the company’s coding and deployment speed, and that Spotify released more than 50 new features throughout 2025.
The remark spread quickly and sparked strong reactions in the developer community — some were excited, some were alarmed, and some said it couldn’t be true.
It is worth taking a closer look at what Spotify is actually doing.
What they are using: the Honk system
Spotify has an internal system called Honk, which integrates Claude with a workflow pipeline.
Here is roughly how it works:
An engineer opens Slack on their phone and sends a message to Honk:
Fix the xxx feature in the iOS app. Steps to reproduce: xxxx
Or: Add an xxx feature to the push notification email for users
After receiving the task, Honk:
- Calls Claude to understand the request and generate code
- Automatically builds and runs tests
- Packages everything into a new version of the app
- Pushes the result back to Slack
Before arriving at the office, the engineer can see the completed version on their phone, confirm it is correct, and merge it directly into production.
How this differs from vibe coding
Many people describe this kind of workflow as “vibe coding” — the idea that you can just say a few words and code appears.
Spotify’s approach is closer to an industrialized process: there are clear requirements, test coverage, review confirmation, and a deployment pipeline. Engineers are not casually vibing; they are making architectural decisions, defining requirements, and controlling quality — they have simply outsourced the typing step to AI.
This distinction matters. Code quality is not guaranteed by vibes; it is guaranteed by the automated test suite, the engineer’s final sign-off, and the engineering infrastructure behind Honk.
How to understand the METR study that said AI slows programmers down
There is an apparent contradiction.
We previously reported on a METR study: experienced programmers using AI tools completed tasks 19% slower than those who did not use AI.
But Spotify’s situation differs from the METR test scenario:
- METR tested writing new, unfamiliar tasks under AI tools, where engineers had to constantly correct AI errors
- Spotify’s Honk is designed for highly structured internal tasks with abundant context: a mature codebase, comprehensive internal documentation, and standardized processes
Honk works well under these conditions because Claude has enough context to leverage.
What this means for engineers
Spotify is not reducing its number of engineers. It is actually accelerating its product release cadence — behind those 50-plus new features, people still need to figure out what those features should do and how they should look.
This is probably the most realistic picture of AI coding today: engineers are not disappearing; their work is shifting to a higher level of abstraction — from writing code to defining what the code should do.
What this shift means for engineers’ long-term career development remains an open question. But the Spotify case shows us that, with the right infrastructure in place, this shift is already happening in real companies.
Sources: Spotify says its best developers haven't written a line of code since December,CocoLoop、 thanks to AI（TechCrunch）；Spotify's AI Coding Shift: Honk and Claude Code Explained（Let's Data Science）