OpenAI just shipped a new feature for the Codex app called Record & Replay, and it flips the usual agent-programming model on its head. Instead of writing instructions or prompts to teach Codex how to do something, you just show it once. Codex watches, learns, and turns your demonstration into a reusable, editable skill.
The feature is currently rolling out to select markets, with broader availability coming later.
Show once, reuse forever
The core idea is dead simple: you perform a task in front of Codex , filing an expense report, submitting a time-off request, navigating a multi-step internal tool , and Codex records the workflow. You control when recording starts and stops. Once you're done, Codex converts that demo into a skill: a structured, inspectable artifact that it can replay on demand.
Skills in Codex are already a first-class concept. Agent skills extend Codex with task-specific capabilities, packaging instructions, resources, and optional scripts so Codex can follow a workflow reliably. What Record & Replay adds is a new input method for creating those skills , one that requires zero writing. You demonstrate instead of describe.
What the skill looks like under the hood
The resulting skill isn't a black box. Once Codex generates it from your recording, you can inspect and edit it. Codex uses progressive disclosure to manage context efficiently: it starts with each skill's name, description, and file path, and loads the full instructions only when it decides to use a skill. This means the generated skill follows the same format as any hand-authored one , a
Don't miss what's next in AI
Join 300,000+ engineers and researchers who get the signal, not the noise.
- Full access to in-depth AI research breakdowns
- Be the first to know what's trending before it hits mainstream
- Daily curated papers, repos, and industry moves