OpenAI has officially launched a highly anticipated Record & Replay feature for Codex, signaling a major leap in its “Computer Use” capabilities.

Instead of writing complex textual prompts or code to build an automation script, users can now teach the AI simply by demonstrating a task on their computer. Codex observes the manual sequence, maps the cross-application boundaries, and builds a robust, repeatable workflow.

🛠️ How It Works: “Show Once, Run Forever”

The feature changes the way users approach desktop automation, shifting from traditional robotic process automation (RPA) tools to visual, demonstration-based learning.

• Initiating the Capture: Users trigger the Record & Replay plugin directly from their system. While recording is active, you perform your routine task as you normally would—such as logging into an expense portal, extracting data from an invoice PDF, copying it into a spreadsheet, and pinging a Slack channel.
• Compiling into a “Skill”: Once you stop recording, Codex processes the captured screen interactions, window hierarchies, and inputs. It translates that visual timeline into a reusable automation package called a “Skill.”
• The SKILL.md Blueprint: Every recorded workflow is compiled into an inspectable, human-readable, and completely editable SKILL.md file. If a step changes or you want to insert variable inputs (like telling the AI to look for a new date or file on future runs), you can tweak the markdown file directly instead of re-recording the entire sequence.

🔒 Enterprise Scaling & Cross-Team Deployment

OpenAI is positioning Record & Replay as a massive budget-saver for corporate operations that historically rely on tedious manual labor or rigid legacy automation infrastructure.

A core highlight of the deployment model is Skill Sharing. Once a single employee records and refines a complex workflow—like a custom DocuSign contract pipeline or a multi-step Salesforce data entry routine—the resulting Skill file can be published directly to an organization’s centralized Codex registry. From there, the entire department or company can instantly run that same automation on their own machines.

To ensure safety across sensitive corporate databases, the feature ties into OpenAI’s newly launched App Permissions infrastructure. Workspace administrators can enforce “Always Ask” or “Important Actions” guardrails, forcing Codex to request explicit human confirmation before executing high-impact cross-app tasks (like submitting a financial transfer or deleting cloud entries).

📋 Availability and Core Limitations

Because the tool interacts deeply with system-level UI elements, the initial preview launch carries specific structural restrictions:

• OS Gatekeeping: The feature is currently exclusive to macOS users operating an active Computer Use environment.
• Ecosystem Footprint: Record & Replay plugs natively into Codex’s expanding environment, which features deeper integrations across terminal CLIs, VS Code, and major enterprise platforms like Slack, Gmail, Notion, and HubSpot.
• Geographic Rollout: Due to ongoing international regulatory evaluations, the feature is currently unavailable within the European Union (EU), with no timeline yet shared for a broader global release or specific separate pricing tiers.