OpenAI released Record & Replay for its Codex app on macOS, letting users demonstrate a workflow once and have the AI repeat it autonomously. The feature converts the demonstration into a reusable 'skill' that can execute tasks like uploading YouTube videos with metadata, thumbnails, and subtitles.
Key facts
- Record & Replay converts one demonstration into a reusable skill.
- Feature requires Computer Use to be enabled on macOS.
- Not available in EU, UK, or Switzerland.
- Version 26.616 adds bulk automations and thread handoff.
- Codex competes with Cursor ($9B+) and Claude Code.
The Record & Replay feature, announced in version 26.616 of Codex According to The Decoder, marks a step toward one-shot agentic automation. Users perform a task once while Codex records the sequence; the system then parses that recording into a skill that can be triggered repeatedly without further human input.
The feature requires Computer Use — OpenAI's vision-based agent capability that lets Codex see and interact with the screen — to be enabled. Computer Use has been available in the EU since June 16, but Record & Replay itself is withheld from the EU, UK, and Switzerland, likely due to regulatory uncertainty around autonomous systems under the EU AI Act.
What the release adds beyond recording
Version 26.616 also ships bulk actions for the Automations history, allowing users to manage multiple automated tasks at once, and thread handoff between local and remote hosts — letting a task started on one machine continue on another. These additions suggest OpenAI is building toward persistent, cross-session agent workflows that survive machine boundaries.
Codex is free to download but requires a paid ChatGPT account for substantive use. The app competes directly with Cursor, the AI-first code editor valued at $9B+ and backed by a16z, and with Anthropic's Claude Code, which also offers agentic coding capabilities. Cursor recently trained a GPT-size model from scratch with 10-20x more compute, signaling that the agentic coding market is entering a model-quality arms race.
One-shot recording vs. scripted automation
The key architectural difference between Record & Replay and traditional macro recorders is that Codex doesn't just replay mouse clicks — it builds an internal representation of the workflow as a 'skill,' meaning it can generalize to variations (different filenames, different metadata fields) without explicit programming. This mirrors the 'meta-skill evolution' pattern reported in multi-agent systems earlier this week [per gentic.news], where agents learn to compose skills without retraining.
The geographic restriction is the most telling signal. OpenAI is willing to ship agentic autonomy in the US and most of Asia but is blocking the feature in the EU, UK, and Switzerland — three jurisdictions with active or pending AI regulation. This is a real-world test of how regulation shapes product rollouts: the EU AI Act's high-risk classification for autonomous systems likely drove the decision.
Key Takeaways
- OpenAI's Record & Replay lets Codex learn a workflow from one demo and repeat it autonomously.
- The feature is blocked in the EU, UK, and Switzerland.
What to watch
Watch for OpenAI to extend Record & Replay to Windows and Linux, and whether the feature expands beyond macOS. The geographic restriction will be tested if EU regulators clarify the AI Act's stance on one-shot agentic skills — or if OpenAI decides to risk the regulatory friction. Also track whether Cursor or Claude Code ships a comparable one-shot recording feature in the next 60 days.
Source: the-decoder.com
