OpenAI demos engineer Rannie Song using a ChatGPT dot to triage Reddit feedback and trigger Codex pull requests
We’re using dots to build better dots. @rennsong, an engineer at OpenAI, uses her dot to gather feedback from social media and user reports, update a live tracker in ChatGPT Space, and investigate and fix bugs with Codex.
Context
Connecting ambient desktop agents that gather unstructured user complaints directly to coding agents that open pull requests illustrates how software teams are closing the loop between product triage and automated code patches.
OpenAI introduced dots inside its desktop workspace to act as persistent, specialized AI teammates that can monitor external sources, maintain shared tables, and collaborate with engineers. By pairing dots with Codex, product teams can hand off triaged bug reports directly into code generation tasks.
Deep dive
On October 23, the official ChatGPT account shared a short video featuring OpenAI engineer Rannie Song demonstrating how the team uses internal ChatGPT dots to build and improve dots themselves. In the clip, Song asks her dot in a macOS sidebar chat to open an external feedback page and update a structured Feedback Tracker table with the latest user reports from Reddit.
The video shows the purple dot agent automatically populating rows in the tracker with priority levels such as P1, product categories including connections and integrations, memory and continuity, voice, and reliability, alongside a concise summary and triage rationale for each issue. In one highlighted row covering broken computer connections, the agent tags a teammate directly inside the workspace comment field to investigate.
According to OpenAI, once the dot summarizes recurring feature requests and bug reports from community channels, Song hands the triaged items to Codex to draft and submit pull requests. While the video is a brief promotional workflow showcase rather than a new product launch, it highlights how persistent workspace agents and coding agents are being chained together inside OpenAI's own engineering loop.
What to expect next
Chaining social feedback ingestion directly to automated pull request generation should shorten turnaround times on routine UI bugs and user feature requests across software teams, while placing greater weight on automated regression test coverage, strict sandboxing against untrusted community text, and careful human code review before agent authored changes ship to production environments at enterprise scale across desktop and cloud workspaces.