OpenAI plans to test one ChatGPT window for both chatting and agent work

One window instead of two

OpenAI wants ChatGPT to do more than answer questions. As Cryptopolitan on MSN reported, citing an email seen by BleepingComputer, the company will test a design that folds its chat assistant and its Work agent into one interface. A small group of users will get it first.

Today the two have different jobs. Chat is for quick questions and brainstorming. Work, an agent (software that carries out a multi-step task for you rather than just replying), takes on larger jobs: research, spreadsheets, reports, presentations and published Sites. You can upload files, use connected plugins, and watch and steer what the agent is doing.

In the merged version, all of that would live in a single conversation. You could start with a question and end with a finished deliverable without changing tools. Codex, OpenAI's coding product, would remain its own platform.

Anthropic got there first

OpenAI is not the first to try this. In a September 16 announcement, Anthropic said its Cowork agent mode and its chat mode were becoming one Claude. The combined experience began rolling out to Pro, Max and Ultra subscribers.

Enterprise customers followed on October 23. Team and Free plans have yet to receive it. Anthropic also added Claude Docs and Slides, which let people edit documents and presentations.

ChatGPT Work already offers similar abilities, including publishing Sites. The direction is the same at both companies: get people to stop treating an AI as a question box and start treating it as a coworker that finishes tasks.

Why the usage meter matters

Putting everything in one window makes cost harder to ignore. An agent job is not one reply. It can involve many steps, and each step uses computing power, so a long task can use far more of your allowance than a chat does.

OpenAI's current rules already point this way. Work and Codex draw from one shared allowance, and once it runs out you can buy credits. Speed costs extra too. On the $500 Pro plan, the Ultrafast setting uses included usage at eight times the standard rate, and after that it draws credits at six times the standard rate.

A further change arrives in December. Subscribers on Plus, Pro and Business will draw on one pool, counted in tokens (small chunks of text that models read and write), that covers both ChatGPT and Codex. Quick Instant chat will not count against it. Users will be able to see what is left, and when it resets, under Settings > Usage.

This follows OpenAI's October 29 move to double Work and Codex allowances for Pro subscribers at $100 and $500. Affected Pro $200 accounts also got higher limits.

Big forecasts, shaky trust

The money involved is rising quickly. Gartner expects spending on AI agents and assistants to climb from $29.2 billion in 2026 to $65.5 billion in 2027. It also projects total AI spending of $2.67 trillion this year, up 49.5%.

Actual adoption is far less advanced. An IDC study of 1,900 organizations found only 3.1% at the most mature stage of AI use, while 61.3% were still in the first two stages.

Trust in the meter is already being tested. Cryptopolitan earlier reported complaints from Codex users who felt their allowances were draining unusually fast. OpenAI executive Thibault Sottiaux rejected claims of undisclosed cuts. Kingy AI first rated the complaints "supported but unproven." It later revised that to "partly confirmed; blanket cut unproven" after acknowledging problems with cache efficiency, meaning how well the system reuses work it has already done.

The broader point is that as assistants take on bigger jobs, people will care about predictable costs nearly as much as about what the tools can do.