It's a Feature, Not a Bug: Making Sure Code You Didn't Write Does What You Meant

AS

Andrew Stellman

O’Reilly author and AI educator

When an AI agent writes the code, it can do exactly what was asked and still miss what was meant. Andrew Stellman opens with a vibe-coded bus tracker that sent him to the wrong stop, then uses it to show how requirements, code, and tests can all agree while the original intent is lost. Drawing on his experience managing large development teams and on decades of quality engineering, he frames the problem as specification, delegation, and verification. He describes Quality Playbook, his open-source skill for Claude Code, Cursor, and Copilot, which reconstructs intent from documentation, project discussions, and users' reports, turns it into requirements, checks the code against them, and confirms findings with failing tests. Attendees will see findings from open-source projects such as Gson and the Linux kernel, including a long-standing defect found through user bug reports, and learn how to check that intent survived.