Spotify has built an AI coding agent called Honk to handle large-scale codebase migrations across many engineering repositories.
InfoQ’s presentation coverage says the work focuses on fleet-wide changes rather than one-off code suggestions. Jo Kelly-Fenton and Aleksandar Mitic describe architectural lessons such as decoupling CI verification runtimes from AI agents, managing automated pull request bottlenecks, and pushing standardization across thousands of repositories.
That makes the project more interesting than a typical coding demo. Large organizations often know what migration they need, such as updating frameworks or removing old patterns, but the work is repetitive and spread across many services. An agent can help apply the change, run checks, and open reviewable patches.
The hard part is operational, not just model quality. If the agent floods maintainers with poor pull requests or cannot verify changes reliably, it creates new work. Spotify’s experience shows that successful coding agents need workflow design, test infrastructure, and repository standards as much as prompt engineering.