OpenAI Retires Its Atlas Browser and Folds Codex Into ChatGPT Work, a Long-Running Automation Platform With Per-Hour Billing

Since OpenAI launched ChatGPT Work this week, the prior coverage has established the platform's existence and pricing structure. What the Ars Technica reporting adds is a clearer picture of exactly what got absorbed, what got deprecated, and where the unresolved friction lives.

Operator Is Dead. Codex Is Merging. Classic Chat Gets Demoted.

OpenAI's dedicated Operator web browser is being sunset, according to Ars Technica. It lasted fewer than nine months. The stated reason is that its core functionality, browser-based automation, is now built directly into ChatGPT Work via a built-in browser on the desktop client and an updated ChatGPT Chrome extension.

The Codex situation is messier. OpenAI describes Codex technology as "built in" to ChatGPT Work, but Codex still exists as a separate "view" inside the new ChatGPT app. On mobile, Codex is not available at all. The original ChatGPT desktop app, the basic conversational interface most users have known for years, is being rebranded "ChatGPT Classic" and pushed to a deprecated "quick chat" button.

What ChatGPT Work Actually Does

The central promise OpenAI is making: the platform can "stay with a project for hours if needed," according to Ars Technica. That directly addresses the complaint about the old Agent Mode in Operator, which tended to stall after a few minutes on longer tasks.

OpenAI wants users to test it on work they already understand, like analyzing a budget or preparing a sales meeting, so they can judge the output quality themselves. The company also claims it can run full workflow sequences, taking a task from customer research through a campaign brief to finished marketing assets.

Integrations cover Slack, Microsoft Teams, Google Drive, and SharePoint through custom-built plugins. The desktop version can also read and modify local files. Users can pull in a specific app manually using the "@" symbol, or let the system decide which tools to invoke on its own.

OpenAI says the platform will pause and wait for user approval before taking "important actions," though the source material does not define specifically which action types require a sign-off and which do not. That ambiguity matters for anyone considering running this on real business data.

The Cost Picture

ChatGPT Work billing follows the same structure as Codex. Subscription plans reach up to $100 per month, with usage included. OpenAI explicitly warns that complex tasks "may use more of your plan's included usage" than a standard chat request. What happens when a user hits the ceiling on a multi-hour autonomous task mid-run is not addressed in the available sourcing.

Scheduled Tasks: Cron Jobs for Everyone

The platform includes Scheduled Tasks, described by Ars Technica as a "souped-up version of cron jobs." These can trigger on a set schedule or when a monitored event occurs, continue running while the user is away, and be checked from a phone. For IT and operations workers, this is familiar territory. For the average office worker who has never written a cron job, it represents a genuine new capability, assuming the interface is as simple as OpenAI implies.

The Legitimate Concern About Autonomous Access

The strongest concern worth taking seriously: a tool that autonomously reads and modifies desktop files, accesses workplace platforms like Slack and SharePoint, and runs for hours without continuous human oversight is a significant attack surface. If an account is compromised, or if the model misinterprets a task scope, the potential for damage scales with the access granted. OpenAI points to its Compliance API and enterprise admin controls as guardrails, but how granular those controls are, and whether they are on by default or require active configuration, is not clear from the available sourcing.

Companies that rushed to deploy early AI tools without proper access controls have learned that lesson expensively. OpenAI's answer is "admin controls exist," which is a starting point, not a complete answer.

What Is Still Unresolved

The blurry line between ChatGPT Work and standalone Codex is the most concrete open question after this launch. OpenAI describes them as merged and as separate simultaneously. For developers who built workflows around Codex specifically, or enterprises that licensed it under specific terms, it is not yet clear whether the Codex "view" inside ChatGPT Work delivers identical functionality or a subset of it. That distinction will determine whether this consolidation is a genuine improvement or a quiet downgrade dressed up as a product launch.