Most people describe a task to AI from scratch every time.
OpenAI's new Record & Replay feature for Codex takes a different approach: you show Codex how you complete a task on your Mac once, and it turns that demonstration into a reusable skill.
That makes it useful for work that is repetitive, personal to your workflow, or simply annoying to explain step by step.
What It Does
You record yourself completing a workflow on your Mac. Codex observes the relevant actions and window content, then drafts a skill that explains:
- when to use the workflow
- which details can change each time
- the steps to complete it
- how to check that it worked
Later, start a new Codex thread and ask it to use that skill with the new input: a different file, date range, customer, issue, or report.
It is not a screen recording you watch back. It is a way to teach Codex the pattern behind a task.
Useful Things To Record
The best recordings are short, stable, and have a clear finish line.
Good examples:
- filing the same type of expense every week
- downloading and organizing a recurring report
- creating a correctly formatted issue in your team's tracker
- uploading and publishing a video with the usual settings
- booking a recurring parking space or travel arrangement
- moving data between the same internal tools
For example, instead of writing a long prompt explaining how your team creates a bug report, demonstrate it once. Next time, ask Codex: “Create a bug report for this issue using our usual workflow.”
How To Use It
- In the Codex app, open Plugins.
- Open the + menu and choose Record a skill.
- Give Codex a little context: what you are teaching and what may vary next time.
- Approve the recording when you are ready.
- Complete the task normally, without drifting into unrelated work.
- Stop the recording when the task is complete.
- Review the skill Codex creates and add any hidden preferences, such as naming rules or required checks.
The next time you need the task done, open a fresh thread, name the skill, and provide only what changed.
Make Your First Recording Better
Before recording, decide what the task should look like when it is finished.
Then keep the demonstration focused. Do not spend five minutes looking for a tab, changing unrelated settings, or handling private messages. Codex learns from the workflow you show it.
Tell it in advance which inputs will change. For example:
I am going to show you how I publish our weekly report.
The report file and publication date will change each time.
Keep the title format, folder location, and final review step the same.
Afterward, ask Codex to show you the drafted skill. Add the small decisions that matter but are easy to miss: which folder to use, how to name the file, who must approve it, and what counts as done.
When Not To Use It
Do not use Record & Replay for unstable or high-risk workflows until you have reviewed the generated skill carefully.
Avoid recording secrets, passwords, private customer data, or anything you would not want captured in a demonstration. Use realistic but non-sensitive inputs whenever possible.
It is also not the best answer when the task needs a full team integration, several coordinated skills, or a durable package shared across an organization. In that case, building a proper Codex plugin may be the better long-term move.
Who Can Use It Now
Record & Replay is currently available on macOS when Computer Use is available and enabled. Initial availability excludes the European Economic Area, the United Kingdom, and Switzerland.
So this is worth trying now if you work on a Mac and have a task you repeat often. Start with one low-risk, 2-to-5-minute workflow. The first goal is not automation perfection. It is removing one small, recurring piece of friction from your week.
Bottom Line
Record & Replay is interesting because it lowers the effort required to teach AI your way of working.
The practical test is simple:
If you can demonstrate a predictable task once, and you do it again next week, it is a strong candidate for a Codex skill.