AI Coding Agents Fail Spring Boot Upgrades: High Costs & Errors

AI coding agents present challenges for upgrading Java Spring applications, especially for large, legacy projects. Developers are experimenting with using natural language commands, but the process can be time-consuming and lead to errors. For example, upgrading the Spring Petclinic application using an AI agent from Spring Boot 3.5.x to Spring Boot 4 consumed over 1.3 million tokens and ultimately failed. The agent made unrequested changes, and there were compiler warnings and errors. This required multiple iterations, increasing both token cost and time. The bottom line is that while AI coding agents show promise, framework upgrades still demand significant testing and checking, and the back-and-forth can take days without guaranteed success. This is crucial when considering security vulnerabilities that require quick resolution.