OpenAI plans to test a unified ChatGPT experience that merges the Chat and Work tabs, allowing users to move from conversation to tasks involving files, plugins and web tools without switching modes; Codex will remain a separate area in the desktop app.

The design builds on features such as intelligent interfaces, persistent agents called Dots, plugins and Work’s access to files, apps and a browser, while keeping files, prompts and results together and reducing clicks.

By letting users assign tasks within everyday conversations rather than opening a separate agent mode, the unified interface could make AI agents easier to use and put them in more direct competition with standalone productivity and creative apps.

OpenAI will first make the update available to select users for feedback before expanding the rollout, and it notified users about the changes by email.

The move follows Anthropic’s September 2026 decision to combine chat and Cowork in Claude, with its unified experience rolling out to Pro, Max and Ultra subscribers before Enterprise; OpenAI has not said whether Anthropic’s change influenced its plans.

A shared ChatGPT and Codex allowance is scheduled to begin in December for ChatGPT Plus, Pro and Business users; OpenAI says most users should not have to wait longer for usage to reset than they do now.

OpenAI has not disclosed the shared pool’s size or how allowances will vary by plan, leaving heavy users without enough information to judge how the change will affect them.

Some subscribers are concerned that heavy Codex use could eat into allowance they would otherwise use for ChatGPT, and some say they may cancel; OpenAI’s reassurance about most users leaves the impact on heavier users unresolved.

Ahead of the change, OpenAI doubled ChatGPT Pro usage limits on October 29, 2026, for Pro $100 and $500 subscribers, raised limits for affected Pro $200 accounts and canceled a planned cut to that tier.

The adjustments follow complaints from some Codex users that their allowances drained unusually quickly; OpenAI executive Thibault Sottiaux rejected claims of undisclosed cuts, while Kingy AI said cache-efficiency problems were partly confirmed but a blanket reduction was unproven.

The shift reflects the heavier computing demands of agentic work involving research, code, files and browser use; the article cites estimates that Claude Cowork tasks can use five to 20 times the compute of ordinary chat and says Codex has more than five million weekly users.

The shared pool puts more of the trade-off between different kinds of computing use in customers’ hands: casual users may find Unlimited Instant reassuring, while subscribers doing lengthy Work tasks could face tighter limits as the test expands.