While AI agents are getting better at writing code, they sometimes struggle with making the code faster. We have a number of tools at our disposal to do this, but some of them are not made with AI agents in mind. The profvis package is one such example. The interactive flame graph it produces is fantastic for human eyes, it allows us to easily see the broad picture while also drilling down into the details. This is what started the idea for this new package; debrief. Allowing the AI agent to read the profvis results as text, thus letting it make more informed decisions about how to optimize the code.
This is a companion discussion topic for the original entry at https://opensource.posit.co/blog/2026-06-22_debrief-0-1-0/