Keep pulling the thread on Jason Liu.
An automation in Codex can find a flight booking ticket and reference number, check in for the flight, and text the boarding pass to a user's personal phone.
Codex can be instructed to create a pull request, monitor it until all tests pass, and then automatically send a direct message on Slack with the pull request link and preview URL.
A skill called "self-improve" can be built in Codex to review the most frequently used skills, read their session logs, and identify consistent user feedback to suggest improvements.
As of the week prior to the podcast, OpenAI allows users to submit their own plugins to the Codex plugin directory.
The most valuable skill for workers using AI is the ability to be dissatisfied with a result, learn the vocabulary to describe the problem, and articulate the necessary improvements to the AI.
The primary job remaining for knowledge workers is to understand what they dislike about something, put it into words, and communicate it to an AI.
OpenAI is moving many of the capabilities from the desktop Codex application into the cloud-based ChatGPT Work experience.
A key product goal for OpenAI is to enable work automations to continue running in the cloud even after a user closes their laptop.
ChatGPT Work is a different user experience view of the Codex experience, designed for work productivity tasks.
Codex displays Git history, pull requests, and code changes, while ChatGPT Work hides these details from the user.
For most organizational and scheduling tasks, the SOL model at a medium effort level is sufficient and does not require the highest intelligence.
For building complex prototype applications, such as a drum-learning app, it is best to use extra high or ultra effort levels in Codex.