Spotify scales 1M automated code changes with OpenRewrite

If you’re a platform/devx engineer, you probably know Backstage, the open source developer portal by Spotify used across many enterprises. It brings together docs, systems, components, compute resources, and teams, with templates to automate repetitive workflows in a structured way. Huge thanks to the Backstage pro Juan Pablo Garcia Ripa for the deep dive today. Was awesome to get more context on how things actually work under the hood. I’ve been looking for a way to contribute and decided to jump in on improving their new plugin migration to the new frontend system: https://lnkd.in/gr78Ve77 If you’re planning a Backstage migration or want to contribute, feel free to DM me. Would love to partner with more Backstage community devs and make these migrations easier for everyone. Sharing what I kicked off below 👇🏼

I just read the latest ByteByteGo breakdown on how Spotify ships to 675 million users, and it’s a brilliant reminder of what actually makes a high-performance engineering culture work. At that scale, Spotify relies on "Golden Paths" - standardized, opinionated workflows that take the guesswork out of building and deploying. It’s about reducing cognitive load so engineers can focus on solving actual problems rather than fighting with infra. Reading it felt like a mirror to my own journey at JUSPAY. Over the last few years, I’ve had the chance to lead several teams through the transition to Trunk-Based Development and Canary Deployments. Whether it was the Payment Page team in 2022 or the AI Infra team today, the challenges were the same: - Moving away from "merge hell" and frozen sandbox branches. - Solving the high-stakes risk of shipping in environments with no Dev-stage. - Dropping deployment issues by 90% simply by building our own "Golden Path." It’s one thing to read a case study about Spotify; it’s another to implement those principles in the middle of 20+ PRs a day and zero room for error. I’ve decided to write down the full story: the gritty details, the specific tools like Superposition we used to manage the chaos and the lessons learned across 4 different teams, in my first-ever Medium post!!! If you’re trying to scale a team without losing your sanity, I’d love for you to check it out: https://lnkd.in/dsXhNh_X #EngineeringLeadership #DevOps #TrunkBasedDevelopment #SpotifyEngineering #Juspay #ScalingSystems

Spotify's leadership just disclosed that their senior developers haven't manually written a line of code since December. Meanwhile our customers are drowning in validation work. The numbers tell the story - AI writes code 76% faster, but incidents per pull request jumped 23.5%. Change failure rates are up 30%. We're not in a productivity crisis. We're in a validation crisis. The bottleneck moved from creation to review, and most teams are still pretending the old playbook works.

Spotify devs stopped coding in December. Your team still build things by hand? Spotify just said their top engineers haven’t touched code since 2025. Not a typo. They shipped 50+ new features last year. AI did the heavy lifting. Engineers used a system called “Honk” to deploy features with Claude Code. No more waiting for reviews. No more endless sprints. Dev cycles run at the speed of a Slack message. This is not the end of the line. It’s the new starting point. Spotify is building datasets no LLM can copy. Unique user data, not just Wikipedia clones. Their models get smarter with every song, every playlist, every taste. That’s a moat you can’t buy. Here’s what I see: → AI is now the main engine for product velocity → Engineers are moving from code writers to problem solvers → Teams that wait will fall behind → The best features are built by humans who know how to use AI as leverage If your team is still building every thing by hand, you’re burning time you can’t get back. The game now is about speed, leverage, and building with intent. The opportunity is bigger than ever.

SPOTIFY SOLVED THE MICROSERVICES CHAOS PROBLEM. THEN OPEN-SOURCED THE SOLUTION Engineers wasted hours just figuring out what existed, docs were everywhere and nowhere, new hires spent weeks mapping the system before writing a single line of code so spotify built an internal tool to fix it then gave it to the world backstage is an open-source developer portal framework...it puts your entire engineering ecosystem in one place a software catalog that tracks every service, library, pipeline, and ml model with ownership, with status, with docs attached software templates so new projects start with your org's standards already baked in. no boilerplate debates. no inconsistent setups techdocs so documentation lives next to the code...engineers actually maintain it when it's not buried in notion or confluence spotify's internal chaos became a 32k star cncf project...1800+ contributors 4.5k companies running it in prod the tool that saved spotify's infra is sitting on github for free → https://lnkd.in/gq9yanir

Spotify has 9,000+ employees across 20 countries. And somehow, things actually get done. Most people credit the culture. The real answer? The structure underneath it. They called it the Squad Model. Six to twelve people per squad, each owning one product area end to end. Decisions made inside the squad. Accountability stays there, too. Spotify built visibility into everything. Who's carrying too much, where things are slowing down, and what's at risk before it becomes a problem. The autonomy only worked because the clarity existed alongside it. So when other companies try the "Spotify model", they borrow the flat structure, the squad names, the no-manager energy. They leave behind the operational discipline that made it sustainable. And then it falls apart and everyone blames the model. But decentralized teams aren't chaos, they're just a different kind of precision. What does your team structure actually look like? p.s. we have been thinking about this a lot while building UnitLook. Worth a look if you are figuring out the same thing: https://lnkd.in/g_H9sv-r

Stripe's Minions merge 1,000+ PRs a week. Ramp attributes about 50% of merged PRs to agents, and Spotify's Honk has shipped 1,500+. I tried to map what the background agent stack in production actually looks like: The interesting part sits in the layer above the coding agents, the one that decides when agents run, where they run, what they can touch, and how hundreds of them coordinate. The teams catching up now are buying the infrastructure instead of building it, because the stack has matured enough that they can. Benjamin Stark and I put together a cheat sheet to organize all of this. Primitives, trigger types, 80+ vendors, real company stacks. All on one page. Comment "agents" and I'll send you the full cheat sheet from Ona.

Deep Engineering Issue #44 is out now. Featuring Sandor Dargo, senior software engineer at Spotify R&D, on C++26 adoption decisions, why fallback plans need to be on every checklist, the compiler gap most teams underestimate, and keeping large C++ systems maintainable. Read it here: https://lnkd.in/gjUzqh-4 #CPlusPlus #CPP26 #SoftwareEngineering #DeepEngineering #SystemsProgramming #Maintainability

A founder burned $340K and the fix was a 45-minute pre-mortem. Spotify PMs figured this out in 2019. Before any sprint, they ran one exercise: imagine it's 90 days from now and the build is dead. Why? The same four killers surfaced every time. No clear owner. No defined user. Integration assumptions that collapsed on contact with reality. A data model built for demo day, not production. They turned those failure modes into scope constraints before a single engineer opened their IDE. Atlassian research shows pre-mortems catch 30% more critical scope gaps before build starts. McKinsey & Company found that rework from misaligned requirements accounts for 42% of total project cost in software builds under $500K. For non-technical founders, this gap is lethal. You can't catch what you can't name. Five questions. Answer them before you write a PRD, open Cursor, or book a dev agency. 1. Who owns decisions when you disagree? 2. Who exactly is the user? 3. What third-party integration has the least documentation? 4. What breaks in your data model when you add the second core feature? 5. What happens to your timeline if the first developer quits? Scope discipline is not a developer problem. It is a founder problem that developers inherit. Codalio runs this process automatically before any code gets generated. That's why it exists. https://codalio.com/