I Freaking Love Breaking My Own Software

Validation testing is a passion of mine. I enjoy finding bugs, figuring out why they occur, fixing them, and then repeating the process. This pursuit began during a job as an application analyst where I tested a new PACS system, documenting user-reported performance issues. My obsession with breaking software began when I worked as a clinician, feeling a sense of ownership when I found performance problems.

I still feel that rush when I discover issues before end users do. Today, I use various tools, including AI, to find edge cases. I ask AI to generate scenarios I may have overlooked and test them against applications. While AI assists in validation, I never ask it if my application is good. Instead, I want it to help me identify potential failures, challenge assumptions, and test the application's reliability.

This philosophy is crucial when building software for AI agents, as they can fail even if individual components seem to work correctly. To address this, I created ReliAgent, which examines API responses, parameters, and metadata for reliability issues before the result is passed downstream. However, this led to another challenge: ensuring the reliability tool itself is reliable. My obsession with breaking software remains unchanged, and I strive to find issues before users do.