OpenAI on July 9, 2026 discontinued its year-old Atlas web browser and folded its functions into a new agent called ChatGPT Work, a system designed to complete multi-step office tasks such as building spreadsheets, slide decks, and documents by working independently across a user's connected applications.

According to OpenAI, ChatGPT Work gathers information from a person's existing tools and turns it into finished materials, including sheets, slides, documents, and web applications. The system can remain active on a single project for hours, breaking a large task into smaller steps and completing them without ongoing supervision.

The change folds two of OpenAI's most closely watched consumer products, the Atlas browser and the Codex coding agent, into a single interface layered on top of ChatGPT. Atlas launched on October 21, 2025, as PPC Land reported at the time, entering a market that already included Perplexity's Comet browser and drawing early scrutiny over how much autonomy its agent mode was given inside a user's browser tabs. Less than nine months later, that standalone product has been discontinued.

What OpenAI announced

According to OpenAI, ChatGPT Work is powered by GPT-5.6, a model the company describes as reaching a new level of performance at reasoning through multi-step tasks and at producing materials that follow a user's own templates and reference files. The rollout accompanies a separate GPT-5.6 announcement OpenAI published the same day, and the model is also, according to OpenAI, becoming the preferred model inside Microsoft 365 Copilot.

Codex sits at the technical center of the new agent. Once positioned narrowly as a coding tool for software developers, Codex now serves, according to OpenAI, more than 5 million people every week. Of that user base, over 1 million now use it for tasks outside software development entirely, a shift OpenAI cites as evidence that the underlying technology can support office and knowledge work as readily as it supports writing code.

The desktop consolidation goes further than adding a new feature. Starting July 9, 2026, the Codex desktop application is merging directly into a new, unified ChatGPT desktop app. Inside that single application, three modes sit side by side: a standard conversational Chat mode, the new Work mode for office-style tasks, and Codex itself, which OpenAI says remains the same coding agent developers already use, including inline editing within code differences, pull request review inside a side panel, and support for working across multiple code repositories in one project. Existing Codex users who update their application will find it has become the new ChatGPT desktop app automatically; developers can still choose to make Codex the default view and keep its logo as the app icon if they prefer that workflow.

The prior version of the ChatGPT desktop application, meanwhile, is being renamed ChatGPT Classic, a labeling change that distinguishes the older, chat-only interface from the newly agentic one now carrying the Work and Codex modes.

The Atlas browser is retired

Atlas itself does not survive this consolidation as a separate product. According to OpenAI, the company is beginning to sunset the standalone Atlas browser and will share information with existing users about how to transition to ChatGPT. In its place, the new ChatGPT desktop app gains a built-in browser, allowing the agent to bring in websites, online tools, and files without a person needing to open a separate application. OpenAI frames this shift as building on lessons learned from Atlas and from the people who used it, rather than as an admission that the standalone browser failed to find its footing. The company is separately updating its Chrome extension so that ChatGPT can be used directly inside Chrome's sidebar, an alternative path for browser-based work that does not require installing OpenAI's own browser at all.

That reversal is notable given how Atlas was introduced less than a year earlier. Sam Altman, OpenAI's chief executive, had described the browser as representing what he called a rare, once-a-decade opportunity to rethink what a browser could be, and the launch demonstration highlighted an agent mode capable of ordering groceries, creating project management tasks, and handling customer service interactions autonomously. Both PPC Land's original coverage of the launch and its subsequent reporting on AI agent traffic patterns through May 2026 tracked Atlas's usage share relative to Perplexity's Comet browser, which continued to hold a larger portion of measured agentic web traffic throughout the browser's life as a standalone product. Comet held roughly 47 percent of agentic traffic in May 2026 compared with Atlas's 20.3 percent share, according to data HUMAN Security published that June, giving useful context for why OpenAI may have concluded that a browsing feature embedded inside ChatGPT, rather than a separate application competing head to head against dedicated browsers, was the more durable strategy.

How ChatGPT Work is meant to function

The mechanics center on delegation rather than direct manual operation. A person gives ChatGPT Work a task, such as analyzing a month-end budget variance, turning source materials into a marketing campaign brief, or preparing for a sales meeting, and the agent works through it independently, though the person can follow progress, answer clarifying questions, change direction mid-task, or approve specific actions. According to OpenAI, a single request can span an entire workflow: turning customer research into a campaign brief, generating marketing assets from that brief, then adapting those assets for different markets, carrying context through every step.

Two supporting features extend that delegation. Scheduled Tasks let a person ask ChatGPT to perform an action once, repeat it on a schedule, or watch for a triggering event and then act, even away from a computer or phone entirely. Examples OpenAI cites include reviewing new Slack updates weekly to refresh a meeting agenda, checking websites and dashboards each morning and sending a summary of what changed, monitoring customer feedback and converting recurring themes into prioritized product ideas, and updating a presentation whenever new feedback arrives by email.

Plugins form the connective layer. According to OpenAI, plugins link ChatGPT to Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, customer relationship management systems, and project trackers. ChatGPT recognizes automatically when a plugin is relevant to a prompt, though a person can also type "@" followed by an application's name. A new unified plugins directory brings these connections into one place, and the system can suggest relevant plugins mid-conversation.

OpenAI is also introducing a public beta feature called Sites, letting a person turn work or an idea into an interactive site or web application and share it through a standard web address. Useful applications, according to OpenAI, include live dashboards, project trackers, launch calendars, prototypes, and internal portals; a person can test a Site inside ChatGPT, and ChatGPT can update it automatically as underlying information changes.

Computer Use and the built-in browser

On desktop specifically, a feature called Computer Use lets ChatGPT operate a person's computer directly, executing tasks in the background across applications, tools, and the browser: clicking, typing, and moving files where they need to go. This can run as a one-time task or as part of a recurring Scheduled Task whenever that work includes steps on the person's own machine.

The built-in browser inside the new desktop app is a related but separate capability, letting ChatGPT research a market, compare sources, pull data from websites, or open and edit files stored in Google Workspace or Microsoft 365 without leaving the application. According to OpenAI, the browser lets the agent bring in fresh context and take steps across multiple web pages while a person reviews and directs the outcome.

Early adoption inside OpenAI and among customers

According to OpenAI, nearly all teams inside the company, including finance and sales, now use ChatGPT Work and Codex together to move faster, take on more difficult tasks, and spend more time with customers. In one internal sales example, ChatGPT Work converted a discovery conversation into a tailored proof of concept for a mission-critical customer problem within 24 hours, a process the announcement says normally takes weeks. The agent structured the sales team's notes, routed the request to a solutions architect, and worked with the technical team directly, freeing the salesperson to focus on the customer relationship.

A second internal example concerns OpenAI's finance function. According to the company, ChatGPT Work reduced month-end close and forecasting from multiple days to hours by helping the finance team locate source data, move it into Excel or Google Sheets, reconcile figures, build accompanying slides, and verify results. That change let finance staff spend proportionally more time understanding what changed in a forecast and advising leadership on next steps, rather than assembling the numbers.

An external example comes from Zapier. According to the announcement, Angela Ferrante, Head of Enterprise Marketing at Zapier, used ChatGPT Work to build a repeatable system for reviewing thousands of sales leads each month. The system traced customer touchpoints across Zapier's customer relationship management platform, email, and other connected tools, identified where lead followups were breaking down, and generated a weekly executive dashboard that, according to Ferrante, highlighted missed pipeline and revealed what she described as seven figures in potential sales.

Availability, rollout sequence, and pricing

ChatGPT Work began rolling out on web and mobile on July 9, 2026, starting with Pro, Enterprise, and Edu subscribers, with Plus and Business access following over subsequent days, according to OpenAI. The new unified desktop application, by contrast, became available globally the same day for Mac and Windows, with Chat, Work, and Codex modes accessible on every subscription tier, including the free plan.

Because ChatGPT Work targets longer, more involved tasks than a typical chat exchange, OpenAI states its usage accounting works differently: usage scales with the work a task actually requires, so complex requests may consume more of a subscriber's included plan usage than a short question would. ChatGPT Work follows the same usage structure already established for Codex.

Enterprise and education administrators retain a further layer of control. According to OpenAI, administrators can set workspace-level spend controls inside the company's Admin Console, configuring default usage limits, setting different limits for specific employee groups, creating individual exceptions for people who need more capacity, and reviewing requests for additional credits that include a submitted project rationale.

Security and governance controls

OpenAI positions ChatGPT Work as built on the same security, privacy, and compliance foundation underlying ChatGPT Enterprise, alongside a Compliance API that, according to the company, gives administrators visibility into ChatGPT Work conversations and actions at an organizational scale. Enterprise and education administrators can centrally manage which employees have access to the agent, what company context it may draw on, which external tools it can connect to, and what actions it is permitted to take.

Controls differ by surface. On the web, according to OpenAI, administrators manage plugin and connected-tool access, configure browser and network access for cloud environments, and restrict sensitive actions inside connected systems. On the desktop, ChatGPT Work builds on governance features already established for Codex, extending enterprise safeguards to work touching local files, installed applications, browser sessions, and network access policies.

A feature OpenAI calls Auto-review adds a further layer: the company's most advanced models review significant actions involving connected tools and external programming interfaces before they are carried out, intended to help prevent unauthorized sharing of sensitive information as agents operate with less direct human oversight than a standard chat exchange involves.

Why this matters for marketing and advertising professionals

For the marketing technology community PPC Land covers day to day, the consolidation OpenAI announced on July 9, 2026 sits inside a broader pattern building steadily since late 2025: narrow, single-purpose AI products folding into fewer, general-purpose agentic systems that connect directly to the software marketers already use.

That pattern is visible across the advertising platforms PPC Land tracks closely. Meta opened its own advertising system to Claude and ChatGPT through a Model Context Protocol server and companion command line interface in an open beta announced on April 29, 2026, giving AI agents direct, authenticated access to campaign reporting, ad creation, and catalog management without requiring a manual login. Meta followed with a separate Developer Tools MCP server, announced June 30, 2026, that explicitly lists the OpenAI Codex App and ChatGPT among four supported clients, alongside Claude Code, Claude Desktop, and Cursor.

The scale figures OpenAI discloses carry direct relevance for how quickly this tooling is likely to spread through marketing organizations. Codex's growth from a developer-only coding tool to more than 5 million weekly users, of whom over 1 million now work outside software development, mirrors a wider pattern: a Stack Overflow developer survey cited in PPC Land's coverage of an OECD paper on agentic AI found roughly half of professional developers already using, or planning to use, AI agents in their work, even as a majority of those same developers recorded ongoing concerns about security and privacy. Those concerns bear directly on ChatGPT Work's Computer Use feature, which grants an AI agent the ability to click, type, and move files across a person's own machine without step-by-step direction.

Retiring Atlas as a standalone product also carries a lesson about how quickly categories shift inside the agentic AI market. Atlas launched with considerable ambition less than nine months before its discontinuation, yet it never closed the usage gap with Perplexity's Comet browser during the period both were tracked as separate products. For marketing and advertising teams evaluating which tools to build workflows around, that trajectory is a concrete data point: a standalone launch, even one backed by a company with OpenAI's resources, is no guarantee of a durable product category, and features that succeed often do so by being absorbed into a broader assistant rather than surviving as freestanding applications.

Finally, the governance and Auto-review features OpenAI describes point to a question the sector will keep working through as agentic tools spread: how an organization verifies, after the fact, exactly what actions an autonomous agent took on its behalf, once that agent holds simultaneous access to advertising accounts, customer data platforms, and internal communications. The Compliance API and Admin Console spend controls OpenAI describes represent one vendor's answer, and how well they function in practice, as usage moves from internal testing to broad adoption, is likely to shape how quickly comparable systems earn similar trust among marketing and publishing organizations handling sensitive data.

Timeline

  • October 21, 2025: OpenAI launches the standalone ChatGPT Atlas browser, entering a market that already included Perplexity's Comet browser.
  • April 29, 2026: Meta opens its advertising system to Claude and ChatGPT through a Model Context Protocol server and companion command line interface, in open beta.
  • May 2026: HUMAN Security data shows Comet holding roughly 47 percent of measured agentic web traffic that month, compared with Atlas at 20.3 percent.
  • June 30, 2026: Meta publishes its Developer Tools MCP server announcement, listing the OpenAI Codex App and ChatGPT among four supported AI clients.
  • July 9, 2026: OpenAI announces ChatGPT Work, discontinues the standalone Atlas browser, merges the Codex desktop app into a new unified ChatGPT desktop application, and begins the ChatGPT Work rollout to Pro, Enterprise, and Edu subscribers on web and mobile.

Summary

Who: OpenAI, the company behind ChatGPT, Codex, and the now-discontinued Atlas browser. The announcement affects ChatGPT subscribers across Pro, Enterprise, Edu, Plus, Business, and Free plans, alongside enterprise administrators managing organizational deployments and external customers such as Zapier, whose Head of Enterprise Marketing, Angela Ferrante, is quoted in the announcement.

What: OpenAI introduced ChatGPT Work, an agent built on the GPT-5.6 model and integrated with the Codex coding system, capable of completing multi-step office tasks, including building spreadsheets, presentations, documents, and web applications, largely without step-by-step human direction. The company simultaneously discontinued its standalone Atlas browser, merged the separate Codex desktop application into a new, unified ChatGPT desktop app, and renamed the prior desktop application ChatGPT Classic.

When: OpenAI published the announcement on July 9, 2026. ChatGPT Work began rolling out to Pro, Enterprise, and Edu subscribers on web and mobile the same day, with Plus and Business plan access following over the subsequent days. The new unified desktop application became available globally on July 9, 2026 for both Mac and Windows.

Where: The rollout applies globally, across ChatGPT's web, mobile, and desktop platforms, with enterprise and education administrators additionally able to configure access and usage controls through OpenAI's Admin Console.

Why: OpenAI states the launch reflects the company's broader ambition for ChatGPT to move beyond answering individual questions and toward completing substantive, multi-step work on a person's behalf. The company cites Codex's growth from a developer-focused coding tool to more than 5 million weekly users, including over 1 million using it outside software development, as evidence that the underlying agentic technology can extend well beyond its original technical use case.