Spotify's top devs haven't written code since December

While most people are half-asleep scrolling feeds on their morning commute, Spotify’s top engineers are apparently deploying software updates to millions of users straight from their phones.

During Spotify’s fourth-quarter earnings call, co-CEO Gustav Söderström dropped a jaw-dropping stat: the company’s best developers have not written a single line of code since December. Instead, they are using generative AI to handle the actual typing.

Commuting into production

The secret weapon behind this shift is an internal platform dubbed "Honk," which integrates Claude Code for real-time, remote code generation and deployment.

Söderström walked analysts through a wild scenario. An engineer riding the train to work can open Slack on their smartphone, tell Claude to fix a specific bug or add a new feature to the iOS app, and let the model get to work.

Once Claude finishes, a fresh build of the app is pushed directly back to the engineer's phone via Slack. The developer can review it and merge it into production before they even step foot inside the office.

Spotify said this workflow has sped up development tremendously. Throughout 2025, the streamer shipped more than 50 new features and app updates. In recent weeks alone, it pushed out AI-powered Prompted Playlists, Page Match for audiobooks, and About This Song.

Subjective data as an AI moat

Beyond automated coding, Söderström argued that Spotify holds a proprietary data advantage that standard web scrapers cannot touch.

Unlike factual resources like Wikipedia, music is inherently subjective. There is no single correct answer to a prompt like "workout music."

Americans generally lean toward hip-hop, though millions work out to death metal. Many Europeans prefer EDM, while a large chunk of Scandinavians opt for heavy metal. Spotify is actively collecting this nuanced, taste-driven dataset at a scale nobody else has, and Söderström noted that their models improve with every retrain.

As for the flood of AI-generated music, Spotify said it allows artists and labels to tag how tracks were produced in the metadata, while platform moderation continues to sweep for low-quality spam.

Why it matters

Spotify just gave the tech world a glimpse of what software engineering looks like when coding agents mature from smart autocompletes into autonomous operators.

We think Spotify's approach is brilliant because it bridges Slack, Claude Code, and mobile delivery into a seamless pipeline. Engineers no longer need to be anchored to a physical desk and IDE to deliver actual code. They have essentially graduated from manual coders to high-level system reviewers, directing agentic workers with natural language.

It also ratchets up the pressure on every engineering organization. If Spotify can ship 50 features a year and push updates from a subway commute, companies tethered to conventional review-and-merge rituals will struggle to match that shipping velocity.

The real challenge, in our view, lies in risk management. Giving engineers the ability to tap "merge to production" on a phone while walking into work demands bulletproof automated testing and ironclad guardrails. For engineers everywhere else, the takeaway is crystal clear: learning how to orchestrate AI agents will soon matter far more than how fast you can type syntax.