Summary
OpenAI engineer Rennie Song shows how they use their personal “dot,” an AI agent named Benny, to collect, triage, and act on product feedback for Dots itself.
The demo starts with a shared page that serves as a color-coded feedback tracker. Benny searches Twitter and Reddit for new posts mentioning Dots and OpenAI, and adds them to the table. Song then comments on a specific issue, such as broken computer connections and Bad Gateway login errors, and asks Benny to investigate in a cloud thread to find a root cause.
Song also phones Benny to talk through the themes. Benny groups the feedback into silent stalls, connection problems, and continuity issues such as lost context between voice chat and Codex, and says it prioritizes silent stalls because users lose trust when the agent goes quiet. It also notes it cannot give an exact ranking for the last two days, and does not guess. Song asks Benny to investigate the stalls and to fix the simplest issue, a greeting failure when a new dot requests a model the user cannot access. By the end, the tracker has grown by 15 items to 210, the stall investigation has found one concrete failure mode, and a pull request is underway for the greeting bug.
📺 Source: OpenAI · Published October 23, 2026
🏷️ Format: Workflow Case Study
