Case Study: Spotify Honk, the agent dropped into existing rails
Spotify's background coding agent, internal codename Honk, had 1,500+ AI-generated pull requests merged into production by November 2025, and was running at roughly 1,000 merged PRs every 10 days by QCon London 2026. The volume is not the lesson. The lesson is where the agent lives: inside Fleet Management, the code-transformation platform Spotify already ran, so the agent inherited targeting, PR creation, review, and merge infrastructure that had been in production for years. Among the eight case studies in this section, Honk is the strongest published evidence that the fastest path to background agents is swapping the agent into rails you already trust.
In 2025, Spotify replaced the deterministic migration script with an agent that takes a prompt. Everything around it, "targeting repositories, opening pull requests, getting reviews, and merging into production," stayed exactly the same. The blast radius of the change was one box in the architecture diagram, and every existing gate still applied. This is the same convergent shape Stripe and Ramp arrived at independently; see Case Study: Stripe Minions.
Spotify closes part 1 with an unusually plain list of open problems: agents are slow and unpredictable, which forces new validation and quality-control mechanisms; they need "robust guardrails and sandboxing to ensure agents operate as intended"; and running LLMs at this scale carries "significant computational expense." Then the line worth quoting: "We don't have all the answers yet." A team merging a thousand agent PRs every ten days says sandboxing and guardrails are unsolved. Take that at face value.