OpenAI spent 2025 shipping chat. On July 9 it shipped an agent that skips the chat and hands you the finished file instead. ChatGPT Work takes a goal, works on it independently for hours, and returns a document, spreadsheet, slide deck, or working web app rather than a wall of text you still have to turn into one yourself.
Codex, OpenAI's coding agent, got folded into the same desktop app in the process. The standalone Atlas browser is on its way out.
What ChatGPT Work actually does
Point it at an outcome, not a prompt. It gathers information across connected apps and workflows, breaks the project into steps, and works through them without needing you in the loop at every stage. The output is a finished artifact: a slide deck, a spreadsheet model, a written report, or a shareable web app, not a chat transcript you have to extract the useful part from.
It connects to the tools that outcome probably depends on: Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers, through plugins.
It's powered by GPT-5.6, OpenAI's three-tier model family that shipped the week before, with the heavier lifting presumably routed to Sol and the lighter steps to Terra or Luna.
One app instead of three
The bigger structural change is what happened to the app itself. Chat, Work, and Codex now live in a single ChatGPT desktop app, available today on Mac and Windows, for every plan including Free. If you'd been running a separate Codex app for coding tasks and ChatGPT for everything else, that separation is gone.
[!NOTE] OpenAI is also sunsetting the standalone Atlas browser, folding its agentic-browsing capabilities directly into ChatGPT rather than maintaining it as a separate product. If you'd adopted Atlas specifically, plan for that functionality to move, not disappear.
Alongside the merger, OpenAI opened a public beta of Sites, a feature for turning raw information into interactive web apps: dashboards, project trackers, launch calendars, internal portals, generated directly from a prompt rather than built by hand.
Rollout
ChatGPT Work is live now for Pro, Enterprise, and Edu users on mobile and web. Plus and Business users get it "over the next few days," which in practice means don't be surprised if it shows up mid-week rather than exactly on schedule.
The timing isn't subtle
This landed one week after Fortune reported Anthropic had overtaken OpenAI on annual revenue, driven substantially by Claude Code's enterprise traction. Coverage of the ChatGPT Work launch explicitly frames it as a response to that competitive pressure, and it's hard to read it any other way: an agent that ships finished spreadsheets and slide decks is a direct pitch at the same enterprise workflows Anthropic has been winning.
Whether merging three products into one app makes ChatGPT more useful or just more crowded is a question that won't have a real answer until people have lived with it for a few months. But the direction is clear. Both labs have concluded that the fight isn't about who answers questions better. It's about who finishes the work and hands you something you can ship.
Tagged with
Written by
DebuggerMe TeamThe DebuggerMe team builds developer tools, writes technical content, and helps teams ship better software.
Related Articles
All articles →Anthropic Overtakes OpenAI on Revenue, and Claude Code Is Why
Anthropic hit a $47 billion revenue run rate by May, passing OpenAI's self-reported $25-33 billion. Enterprise contracts and Claude Code's $2.5 billion annualized run rate explain the gap.
GPT-5.6: What Sol, Terra, and Luna Actually Mean for Developers
OpenAI shipped GPT-5.6 as a three-tier family: Sol, Terra, and Luna. Here's the pricing, the new caching rules, and which tier your workload actually needs.
Anthropic Claude 3.7 Sonnet Hybrid Reasoning
Anthropic's new Claude 3.7 Sonnet introduces hybrid reasoning, letting users toggle between fast responses and deep logical analysis.