Spotify’s Best Developers Haven’t Written Code Since December — Here’s How

February 15, 2026 • 11 min read

During Spotify’s Q4 2025 earnings call on February 12, 2026, co-CEO Gustav Söderström made a statement that sent shockwaves through every engineering org on the planet. The company’s most senior, most experienced developers — the people you’d expect to be writing the hardest code — haven’t written a single line of code since December.

Not because they were on vacation. Not because they were in meetings. Because Spotify built a system that made manual coding unnecessary for them.

The system is called Honk. It’s built on top of Claude Code. And it has fundamentally changed how one of the world’s largest engineering organizations ships software.

“Our best developers haven’t written a single line of code since December. They use an internal system we call Honk, and it’s just the beginning.”

— Gustav Söderström, Spotify co-CEO, February 12, 2026

Within 48 hours, the story was everywhere. TechCrunch, Fast Company, Slashdot, TechSpot, Android Authority — every major tech outlet picked it up. It’s the single biggest real-world validation of agentic engineering at enterprise scale. And it arrived one day before Fortune published their own bombshell: “OpenAI and Anthropic spark coding revolution as developers abandoned traditional programming.”

Let’s break down exactly how Honk works, what Spotify shipped with it, and what this means for every developer reading this.

What Is Honk and How Does It Actually Work?

Honk is Spotify’s internal developer tooling system, built on top of Anthropic’s Claude Code. The name is charmingly Spotify — the company has always had a culture of playful internal naming — but the system underneath is dead serious infrastructure.

Here’s the workflow Söderström described on the earnings call:

An engineer is on their morning commute. They open Slack on their phone. They tell Claude — via Honk — to fix a specific bug in the iOS app, or to add a particular feature. Claude Code does the work: reads the relevant codebase, understands the architecture, writes the implementation, runs tests, and pushes a new build version back to the engineer on Slack. The engineer reviews it on their phone. If it looks good, they merge to production.

All of this happens before they walk through the office door.

Think about what that means in concrete terms. The traditional development cycle for a bug fix in a large mobile app — reproduce, diagnose, implement, write tests, code review, merge, deploy — is measured in hours at best, days at worst. Honk compresses that into the duration of a train ride.

The key insight is that Honk doesn’t replace engineers. It replaces the mechanical act of typing code. The engineer still decides what to build, how the architecture should work, and whether the output meets the bar. They’re doing the parts of engineering that actually require human judgment. The parts that don’t — translating a well-understood requirement into syntax — that’s what Claude handles.

The Results: 50+ Features in 2025

Söderström didn’t just talk about Honk as a future bet. He pointed to concrete output. Spotify shipped over 50 new features in 2025, and the velocity was visibly accelerating through the second half of the year as Honk adoption grew internally.

Some of the headline features that shipped during this period:

• AI-Powered Prompted Playlists — Users describe a mood, activity, or vibe in natural language, and Spotify generates a playlist. This required deep integration between the recommendation engine, NLP processing, and the playlist creation system. Multiple services, multiple teams, shipped fast.
• Page Match for Audiobooks — A feature that syncs your position across the audio and text versions of a book. Technically complex — requires alignment between two entirely different content formats in real-time.
• About This Song — Rich contextual information about tracks, surfacing songwriting credits, production details, and cultural context. Required integration with metadata systems, rights databases, and content generation.

These aren’t trivial features. They’re the kind of cross-cutting, multi-service implementations that traditionally take quarters to ship at a company of Spotify’s scale. The fact that they were shipping this volume while their senior engineers weren’t manually writing code is the most concrete proof point we’ve seen that agentic engineering works at scale.

What This Means for the Industry

Every company talks about AI improving developer productivity. Most of the data so far has been self-reported surveys, GitHub Copilot completion rates, or startups claiming 10x speedups on greenfield projects. Spotify is the first major tech company to say, publicly, on an earnings call, that their best engineers have stopped writing code entirely — and that the results are better.

That changes the conversation. Earnings calls aren’t blog posts. They’re regulated disclosures to investors. When a co-CEO says this on a call with analysts, it means the company has the internal data to back it up and is confident enough to attach their stock price to the claim.

The timing matters too. One day after Söderström’s remarks, Fortune published a piece titled “OpenAI and Anthropic spark coding revolution as developers abandoned traditional programming.” This wasn’t coordinated. It’s convergent evidence. Multiple independent signals pointing at the same conclusion: the way professional software gets built changed in late 2025, and by early 2026, the results are impossible to ignore.

The Pattern: Engineers as Architects, Not Typists

If you’ve been following the AI coding space, the Spotify story fits into a pattern that’s been building for months. Andrej Karpathy coined “agentic engineering” on February 5, 2026, describing a paradigm where developers orchestrate AI agents rather than write code themselves. One week later, Spotify confirmed they’ve been operating in exactly that mode since December.

The convergence isn’t coincidental. It’s the same underlying shift, observed from two different angles.

Karpathy described the theory: developers become architects and reviewers, defining what needs to be built, decomposing it into tasks, and evaluating the output. Spotify demonstrated the practice: senior engineers describe changes to an AI agent via Slack, review the results, and merge to production.

The role of the engineer hasn’t been eliminated. It’s been elevated. The most valuable engineering skills in this new paradigm are:

• System design — Knowing how components should fit together, what trade-offs to make, and where the failure points are
• Task decomposition — Breaking complex goals into discrete, well-defined work items that an agent can execute independently
• Code review — Evaluating AI-generated implementations for correctness, performance, security, and maintainability
• Context management — Ensuring agents have the right information about your architecture, conventions, and constraints
• Quality judgment — Deciding when output is good enough to merge and when it needs another iteration

Notice what’s not on that list: typing speed, syntax memorization, boilerplate fluency. The skills that defined “senior developer” for the last 30 years are being unbundled from the skills that actually create value.

How to Set Up a Similar Workflow

You don’t have Spotify’s resources. You probably don’t have an 8,000-person engineering org or the budget to build a custom Slack-to-CI/CD pipeline. But the core pattern — describe work to an agent, let it execute, review the output, merge — is available to any developer right now.

The underlying technology isn’t proprietary. Honk is built on Claude Code. You have access to Claude Code. The gap between Spotify’s workflow and yours isn’t the AI. It’s the infrastructure around it — the context management, the session organization, the review pipeline.

Here’s how to approximate the Honk pattern at an individual or small-team scale:

1. Set up persistent project context. Create a CLAUDE.md file in your project root that documents your architecture, coding conventions, API patterns, and testing approach. This is the equivalent of Honk’s deep integration with Spotify’s internal docs. Every Claude Code session reads this file automatically, so your agent starts every task with full context about how your project works.

2. Use the Slack pattern for async work. You may not have a custom Slack bot, but the principle still applies: define the task in plain language, let the agent work, review when it’s done. Open a Claude Code session, describe the bug fix or feature, and let it run. Come back when it’s finished. You don’t need to watch it type.

3. Review, don’t rewrite. When the agent delivers output, your job is to evaluate — not to rewrite. Check the diff. Run the tests. Look at the architectural decisions. If something’s wrong, tell the agent what’s wrong and let it fix it. The moment you start manually editing AI-generated code line by line, you’ve broken the pattern.

4. Run multiple agents in parallel. Spotify’s senior engineers aren’t sitting idle while Honk works on one task. They’re orchestrating multiple streams. You can do the same: one session refactoring a module, another writing tests, a third implementing a new endpoint. The bottleneck shifts from your typing speed to your ability to decompose and review.

5. Build your own review pipeline. At minimum: agent completes work, you review the diff, you run tests, you merge. At scale: agent completes work, automated tests run, linter checks pass, you do a final architectural review, you merge. The more you can automate the verification steps, the more agents you can orchestrate simultaneously.

Organizing the Chaos with Beam

The practical challenge with running an agentic workflow isn’t the AI. It’s the terminal chaos. When you have three Claude Code sessions running in parallel — one per service, one for tests, one for infrastructure — you need to know which session is doing what, switch between them instantly, and maintain context across your entire project landscape.

This is where workspace organization becomes the difference between a productive agentic workflow and a disoriented mess of terminal windows.

Each workspace maps to a service or domain. Each tab within that workspace maps to a specific agent session or role. The naming gives you instant recognition. The keyboard shortcuts give you instant access. The saved layouts mean you don’t rebuild your orchestration environment every morning.

Project memory files (CLAUDE.md) persist per workspace, so when you switch from your iOS workspace to your backend workspace, the agent in each context already knows the architecture, conventions, and constraints of that specific service. No re-explaining. No context loss.

This is the same principle behind Honk — deep, persistent context that travels with the engineer — implemented at the individual developer level. You’re not Spotify. But your workflow can operate on the same pattern.

Key Takeaways

The Spotify revelation is the biggest real-world validation of agentic engineering we’ve seen. Here’s what every developer should take from it:

• This is not hype — it’s an earnings call. Söderström made these claims to investors on a regulated disclosure. Spotify has the internal data and is betting their credibility on it.
• The underlying technology is accessible. Honk is built on Claude Code. You have access to Claude Code. The gap between Spotify’s workflow and yours is infrastructure and discipline, not technology.
• Senior engineers aren’t being replaced. They’re being amplified. Spotify’s best developers stopped writing code so they could focus on architecture, review, and orchestration — the highest-value engineering work.
• 50+ features in 2025 is the output metric. The argument isn’t theoretical. Spotify shipped AI-Powered Prompted Playlists, Page Match, About This Song, and dozens more. Users can see the results.
• The commute-to-merge pipeline is real. An engineer can describe a change on their phone, get a completed implementation back, review it, and merge — all before arriving at the office. The entire development cycle compresses into minutes.
• Context management is the new competitive advantage. Honk works because it has deep integration with Spotify’s codebase and documentation. For your own workflow, CLAUDE.md files and organized workspaces are the equivalent.
• This is, in Söderström’s words, “just the beginning.” If Spotify’s senior engineers stopped writing code in December 2025, imagine where they’ll be by December 2026. The question isn’t whether to adopt this pattern. It’s how fast you can get there.