OpenAI has introduced ChatGPT Work, a GPT‑5.6-powered agent that works across files, apps, browsers, and connected tools. It could speed up multi-step work, leaving knowledge workers, enterprise administrators, and IT teams to decide what to delegate and what requires approval.

ChatGPT Work gathers sources, maintains context, checks outputs, and produces finished materials within a single workflow. Its usefulness will depend on users and organizations giving it enough access to coordinate that work while retaining control over important actions.

Across web, desktop, and mobile, users can follow its progress, redirect the work, and approve important actions as the task proceeds.

In short, ChatGPT Work is an AI agent for multi-step knowledge work. It can move from research and source files to completed materials across connected tools while users review progress and retain control over important actions.

ChatGPT Work is a long-running AI agent designed to complete a goal across files, apps, and web tools under user oversight.

An AI work agent carries a multi-step business task from source gathering and context collection to finished materials across the tools where work happens.

OpenAI has introduced ChatGPT Work, an agent the company says can act across apps and files, stay with a project for hours, and carry a goal through to finished work. GPT‑5.6 powers Work. OpenAI says the model improves multi-step reasoning and can create materials that follow a user’s templates and reference files.

The company’s examples go beyond drafting a single response. ChatGPT Work can analyze a month-end budget variance, turn a set of source materials into a marketing campaign brief, or prepare someone for a sales meeting. OpenAI also describes a longer chain in which Work takes customer research, produces a campaign brief, creates marketing assets, and adapts those assets for different markets while retaining the context from each stage.

OpenAI says Work breaks complex requests into smaller steps and completes them independently. A user can follow its progress, change direction, and approve important actions as the work proceeds. The company describes the product as moving ChatGPT beyond answering questions toward doing work across web, mobile, and desktop. Work is rolling out first on web and mobile to Pro, Enterprise, and Edu users, while the desktop app makes Chat, Work, and Codex available on Windows and Mac across plans.

The key point: ChatGPT Work targets the coordination surrounding knowledge work. People still spend time finding files, collecting context, moving information between tools, and keeping the next step from disappearing. Work can reduce those handoffs only by reaching into the systems where that work and context already live.

ChatGPT Work is designed to keep a task moving wherever a user is. A person can start a task from a phone, review a draft or check the status of a longer-running workflow between meetings, then pick up the same work on the web when they return to a desk. On web and mobile, Work sits inside ChatGPT and can research, analyze information, and create documents, spreadsheets, presentations, reports, and Sites.

The new ChatGPT desktop app brings Chat, Work, and Codex into one place. Chat handles questions and conversation. Work handles research and finished materials such as reports, decks, spreadsheets, and web apps. Codex supports software-development work. With permission, Work in the desktop app can use files and applications stored on that specific computer, giving it access to the material needed for a project.

Codex remains OpenAI’s coding agent for developers and technical professionals, with new capabilities for core development workflows. OpenAI says those additions include inline editing within code diffs, pull-request review in the side panel, faster Computer Use powered by GPT‑5.6, and support for multiple repositories in one project. Coding can sit alongside the research, documents, and business material supporting the same project.

The desktop app also adds a built-in browser for the agent. ChatGPT can bring websites, web tools, and online files into the work itself, then take steps across web pages while the task is running. Computer Use extends that access across apps, tools, and browser tabs by clicking, typing, and moving files on the user’s behalf, without the user having to click through each step.

Plugins connect Work to the tools and context where a company already operates. They include Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, project trackers, and other internal systems. ChatGPT can select a relevant plugin from the request, or a user can type @ followed by an app’s name to call it directly. Once those tools are connected, Work can pull in the relevant information, use it to create documents, decks, and analyses, and keep refining them while the user reviews the result.

Sites gives teams a way to build and publish an interactive website or lightweight app from a Work chat. A user describes what it should do and supplies files, data, links, or other constraints. ChatGPT generates a private preview for review and revision. When the user deploys it, the Site receives a live URL, with sharing controls for selected colleagues, a workspace, or the public when that option is enabled.

The result can serve as a shared dashboard, project tracker, launch calendar, prototype, internal portal, or interactive report. Teams can reopen it from the Sites sidebar, refine it through ChatGPT, and publish an updated version when the underlying work changes.

Scheduled Tasks let ChatGPT keep a job moving after a user leaves the chat. A task can run once, repeat on a schedule, or check a website or connected app for new information and report only when there is something worth sharing. OpenAI gives examples such as:

OpenAI says users decide what Work can access, when it should check in, and which actions need approval. In day-to-day use, that means a user gives Work the task, reviews its progress, and steps in when an action needs approval.

Claude Cowork and Perplexity Computer are pursuing similar connected-work systems, which raises the next question: how distinct is OpenAI’s approach?

ChatGPT Work enters a category that Anthropic and Perplexity are already building. The competition is moving toward a work layer: software that holds the task, reaches the files and tools needed to complete it, carries context between steps, and produces a finished result for the person to review.

Claude Cowork is the closest comparison. Anthropic describes it as an agent that works across a person’s computer, local files, and applications after receiving a goal. It is designed for the repetitive and time-consuming parts of knowledge work, such as organizing folders, reading source files, extracting information from unstructured records, and assembling a structured draft from a set of documents. Claude Cowork runs on the desktop and moves between files, folders, and everyday applications without requiring the user to coordinate every step.

Perplexity Computer makes a similar promise through a research-focused product. Perplexity describes it as a general-purpose digital worker that operates the same interfaces as its user. It can run workflows for hours or months, research the web, work across connected tools, create reports and apps, and monitor work in the background. Perplexity also says Computer can use specialized agents for separate parts of a larger job.

ChatGPT Work joins Anthropic and Perplexity in building AI that can carry an entire work process. The chat window becomes a starting point for work that moves through files, browser tabs, and business tools, then returns with something the user can review.

These products address the same pain for knowledge workers. Work is scattered across files, browser tabs, business software, and repeated follow-ups. The aim is to gather the needed context, move information between systems, keep track of what has changed, and reduce the copying, pasting, and chasing that people do between tasks.

OpenAI brings that approach into the ChatGPT environment, where Work sits alongside the desktop app, Codex, plugins, browser access, Computer Use, Sites, and Scheduled Tasks. Claude Cowork and Perplexity Computer make similar promises. OpenAI, Anthropic, and Perplexity are all betting that agents can take more repetition and coordination out of knowledge work.

Agentic workflows can reduce the repetition around knowledge work, but users and teams still have to decide how much of that coordination they are prepared to hand over.

ChatGPT Work is aimed at recurring work that crosses several systems and requires someone to keep many moving pieces organized. Common examples include locating relevant files and data, keeping a project plan up to date, watching for new information, sending the right work to the right person, flagging a missing step, and completing approved follow-ups.

Scheduled Tasks apply the same coordination to work that recurs each day, week, or whenever new information appears. Work can review new messages in Microsoft Teams and Slack each week, refresh a recurring meeting agenda, and share important changes with the team. It can also check websites and dashboards each morning, summarize what changed, send a report, monitor customer feedback for recurring themes, and update a presentation when new feedback arrives by email.

OpenAI cites companies using Work across sales, finance, product launches, and event preparation. At Zapier, ChatGPT Work reviewed customer touchpoints across the company’s CRM, email, and other systems, identified where follow-ups had broken down, and produced a weekly executive dashboard. The company says that process surfaced 7 figures in potential sales that might otherwise have been missed.

RingCentral used Work for monthly launch checks that had required someone to compare release plans, Jira tasks, and go-to-market schedules. Work flagged missing steps, blockers, and unclear ownership, then produced reports that named the responsible people and next actions. NVIDIA used it to prepare for its GTC conference by tracking customer registrations, planned meetings, and field-sales preparation. After the event, Work reviewed session transcripts and customer-meeting notes to assess whether GTC had met its goals.

OpenAI says its sales team used Work to turn a discovery conversation into a tailored proof of concept for a mission-critical problem within 24 hours, a process that normally takes weeks. Work structured the notes, routed the request to a solutions architect, and worked with the technical team. That freed the sales lead to focus on the customer and serve as a higher-value consultative partner.

OpenAI also says its finance team reduced month-end close and forecasting from days to hours. Work found source data, moved it into Excel or Sheets, reconciled it, created slides, and verified the results. The finance team could then spend more time understanding what changed in the forecast, explaining why it changed, and advising leaders on what to do next.

Once Work can read company files and act through connected tools, its impact reaches beyond a flawed chat answer. The same access that saves time can expose sensitive or confidential information unless teams set clear limits on what it can see, change, share, and when it must pause for approval.

ChatGPT Work is built on ChatGPT Enterprise’s security, privacy, compliance, and workspace-management foundation. Enterprise teams need a way to decide which people and systems Work can reach. OpenAI says Enterprise and Edu administrators can centrally manage who has access, what company context ChatGPT can use, which tools it can connect to, and which actions it can take. The Compliance API gives organizations visibility into ChatGPT Work conversations and actions at scale, so oversight teams can review how the agent is being used.

The controls depend on where Work is operating. On the web, administrators can manage access to plugins and connected tools, configure browser use and network access for cloud environments, and restrict sensitive actions in connected systems.

On desktop, ChatGPT Work builds on Codex’s enterprise governance model and administrative controls. Those safeguards apply when Work is handling local files, desktop apps, browsers, and tools. Organizations can also set policies that govern the agent’s network access.

OpenAI adds auto-review as another safeguard. Before important actions involving connected tools or APIs occur, OpenAI’s models can review them for potential unauthorized sharing of sensitive information.

Those controls make governance part of the workflow itself. Giving a system more access and autonomy requires clearer boundaries and oversight.

The appeal of ChatGPT Work is less busywork and more completed work, but that outcome depends on users and organizations granting enough access for the agent to coordinate work while retaining meaningful control over its actions and results.

Q: What is ChatGPT Work?
A: OpenAI has introduced ChatGPT Work, a GPT-5.6-powered agent for multi-step knowledge work. It can carry a task across files, apps, browsers, and connected business systems before returning finished material for a user to review.

Q: What can ChatGPT Work do?
A: ChatGPT Work can gather source material, analyze information, and create documents, spreadsheets, presentations, reports, Sites, and web apps. OpenAI gives examples including budget analysis, marketing briefs, sales-meeting preparation, launch checks, and recurring reports.

Q: How does ChatGPT Work work across apps and files?
A: Work breaks a complex request into smaller steps and retains the task context as it moves through them. With permission, it can use connected files, browser tabs, business tools, local desktop files, and apps, then return with material for the user to review or approve.

Q: Can ChatGPT Work run tasks automatically?
A: Yes. Scheduled Tasks can run work once, repeat it on a schedule, or monitor a website or connected app for new information. Work can then report what changed, refresh meeting materials, update a presentation, or identify recurring customer-feedback themes.

Q: Why does ChatGPT Work matter for knowledge workers?
A: Knowledge work often requires people to find files, collect context, move information between tools, and keep follow-ups from getting lost. ChatGPT Work is designed to handle more of that coordination and return with completed work for people to review.

Q: Who can use ChatGPT Work?
A: OpenAI says ChatGPT Work is rolling out on web and mobile, beginning with Pro, Enterprise, and Edu users. The updated ChatGPT desktop app is available on Mac and Windows, where Chat, Work, and Codex are available across plans, including Free.

Q: Is ChatGPT Work safe for company information?
A: Enterprise and Edu administrators can manage who has access to Work, what company context it can use, which tools it can connect to, and which actions it can take. Users can approve important actions, while the Compliance API gives organizations visibility into Work conversations and actions at scale.

ChatGPT Work is designed to handle more of the steps knowledge workers currently manage themselves, carrying larger portions of a project across connected tools. OpenAI is offering an agent that can stay with a task from locating source files and gathering information to checking work and producing a finished deliverable.

AI can already help with individual tasks. Users can delegate more of the repetitive work while remaining responsible for decisions and approvals that require human judgment.

Knowledge workers, business leaders, enterprise administrators, and IT teams should care because their work often depends on information scattered across files, browser tabs, and business systems. Work can pull that information into the task, keep the job moving, and return a finished result for review.

OpenAI’s announcement brings ChatGPT into the race to build agents that work across the files and software employees already use. Claude Cowork and Perplexity Computer are pursuing similar systems, putting agentic workflows at the center of the AI products companies evaluate.

Organizations need to decide which tasks an agent can handle, what information it can access, and when a person must approve an action. Too little access limits what an agent can accomplish and leaves people doing more of the work manually. Too much access without appropriate controls creates risk.

In short, ChatGPT Work is designed to take on the follow-up, information gathering, and coordination that keep knowledge work moving. It can save time when users give it access to the right context and keep people responsible for approval and review.

ChatGPT Work will be most useful to companies that decide, before the work begins, what the agent can do, what it can see, and where people remain accountable.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.