Spotify has been refining its Fleet Management system to automate software maintenance tasks, addressing the manual work required for keeping codebases updated and secure. The system uses small code snippets to perform transformations across thousands of repositories, significantly reducing the time needed for tasks like bumping dependencies or updating configuration files. Since mid-2024, about half of Spotify's pull requests have been automated, helping maintain consistency across codebases. However, complex migrations still require human expertise, leading the team to explore AI solutions to simplify these processes. In February, Spotify began integrating AI coding agents into its Fleet Management system. These agents allow engineers to define and execute fleet-wide changes using natural language, replacing the need for complex migration scripts. An internal command-line interface (CLI) was developed to facilitate this process, enabling flexibility in switching between different AI models. The results have been promising, with agents generating over 1,500 pull requests for substantial changes, such as language modernization and migrating between UI components. The time savings for these migrations range from 60% to 90% compared to manual coding. The initiative isn't limited to migrations. Spotify's background coding agent can now be triggered via Slack or GitHub, allowing engineers and product managers to initiate tasks interactively. This feature has found enthusiastic users, as it enables quick documentation and proposal of changes without extensive setup. However, the introduction of AI agents presents challenges, including performance issues, unpredictability in outputs, and the need for robust validation mechanisms. Spotify acknowledges these issues and is actively working to enhance the system's reliability and safety.