Getting Started with ChatGPT Work

A workbench for multi-app tasks

ChatGPT Work is not merely a renamed chat window. In the current OpenAI help documentation, Chat is described as the place for quick conversational help, while Work is framed as an agent for longer, multi-step assignments that end in a finished deliverable such as a document, spreadsheet, presentation, report, or Site [2]. That distinction matters because the product is aimed at the everyday friction of office work: information sits in email, Slack, calendars, files, and project history, while the actual output may be a reply, a brief, a plan, or a follow-up list.

The reference example tracked by 8news.ai captures the core pattern: a user needs to confirm a team dinner, with the restaurant’s request in Gmail and the latest headcount and dietary notes in Slack [1]. In standard Chat mode, the user might ask how to write a confirmation email; in Work mode, the user can ask ChatGPT to find the relevant messages, combine the details, draft the reply, and leave the final send decision to the human [1]. The value proposition is therefore not “write faster” alone, but “reduce the switching cost between scattered sources.”

What is current as of the latest guidance

The latest official guidance says ChatGPT Work is still gradually rolling out to eligible accounts, so absence from a given account does not necessarily mean misconfiguration [2]. OpenAI’s documentation currently places Work on ChatGPT web and mobile for eligible paid plans, and in the ChatGPT desktop app when the user’s plan and workspace include it [2]. The same guidance says cloud Work chats sync across web, mobile, and desktop, while local chats started on a computer remain on that device [2].

For organizations, the administrative model is central to getting started. Workspace owners and admins can control Work Cloud and Work Local permissions separately, and Chat remains available even when Work or local Codex access is disabled for a role [2]. Admins can also configure starting model, reasoning level, speed, Fast Mode availability, and new-chat behavior for Work and Codex from workspace settings, though those defaults do not grant access to unavailable models or override role limits [2].

How to start a first Work task

The cleanest way to begin is to choose a task the user already understands. OpenAI’s help flow for desktop starts with signing into the ChatGPT desktop app, selecting ChatGPT from the top-left menu, choosing Work from the Chat/Work toggle, then opening a new chat or existing project [2]. From there, the user should describe the outcome, attach or identify relevant context, add constraints, and state review criteria [2].

A strong first prompt is specific about the deliverable and the sources. For example: “Find the restaurant email about Thursday’s dinner, check Slack for the latest headcount and dietary restrictions, and draft a confirmation email for my review.” That mirrors the 8news.ai example, where ChatGPT Work gathered details across Gmail and Slack, produced a draft, and allowed the user to add a final request such as asking for a quiet table before sending [1].

On web or mobile, the starting flow is similar: open ChatGPT, select Work, start a new chat or project, add any files and context ChatGPT should use, and describe the task or deliverable [2]. Projects are especially useful because they keep related chats, files, and instructions together, giving Work a more stable context for repeated assignments [2].

Connected apps and approval boundaries

The central operational question is permissions. ChatGPT Work can be useful only when it has enough authorized access to retrieve relevant information, but that same access must be bounded by user and workspace controls. In the dinner example, the task depends on connected services such as Gmail and Slack, with the user signing in and reviewing permissions before Work can use those sources [1].

OpenAI’s current Work documentation emphasizes that connected-app permissions, workspace controls, and approval requirements continue to apply when Work runs tasks [2]. That is the key safeguard for beginners: connecting an app does not mean every action should be automatic. A useful mental model is to let Work read, compare, summarize, and draft, while reserving external actions such as sending a message or changing data for review.

The current scheduled-task documentation also says actions that require approval may pause until the user reviews them [3]. This supports the same human-in-the-loop pattern shown in the 8news.ai source: Work can prepare the email, but the user still controls tone, accuracy, recipients, and final transmission [1].

Scheduled and event-triggered Work

Getting started no longer means only running one-off tasks. OpenAI’s latest scheduled-task article says ChatGPT can run one-time or recurring tasks, monitor for changes, and respond to supported events where available [3]. Event-triggered tasks run in Work and can respond to supported Gmail, Slack, or GitHub activity for eligible Plus, Pro, Business, Enterprise, and Edu users, as well as eligible ChatGPT for Healthcare workspaces [3].

The practical workflow is straightforward: connect a supported Gmail, Slack, or GitHub account in Settings, open Work, describe the event and desired action, review the trigger, condition, and prompt, then manage the task from Scheduled [3]. Gmail tasks can respond to a new message or filter by sender or subject; Slack tasks can respond to new channel messages; and GitHub tasks can respond to supported pull request activity in an authorized repository [3]. Slack-triggered tasks require adding @ChatGPT to each monitored channel [2].

There are plan and workspace limits. Free and Go accounts cannot create event-triggered tasks, and event-triggered tasks are not available in FedRAMP workspaces [3]. Active scheduled task limits are currently listed as 3 for Free and Go, 5 for Plus, 10 for Business and Edu, and 15 for Pro and Enterprise, with event-triggered tasks able to run up to 30 times per hour and 720 times per day across all event-triggered tasks [3].

Desktop, web, and mobile: choosing the surface

The desktop app is the best fit when Work needs to sit beside the rest of the workday. The current Work documentation says Work in the desktop app can use local files and desktop apps with permission when the user’s plan and workspace allow it [2]. It also says web and mobile Work cannot directly access files on the user’s computer [2].

Web and mobile are still important because cloud Work chats sync across supported surfaces, allowing a task started on one device to continue elsewhere [2]. However, the desktop app can only display existing event-triggered tasks and cannot create or edit their trigger conditions; OpenAI’s scheduled-task guidance says creation and updating of event-triggered tasks happen on web or supported mobile apps [3]. For a beginner, that suggests a simple split: configure triggers on web or mobile, use desktop for local-context work, and use cloud sync for continuity.

Usage, limits, and governance

Work also has a usage model that beginners should understand before launching long-running tasks. OpenAI’s Codex-plan help article says Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents use a shared allowance and credit pool when those features are available on a plan [4]. Usage varies by model, task complexity, context, reasoning, speed, tools, and where the task runs [4]. A long-running multi-app assignment can therefore consume substantially more allowance than a short request [4].

For managed environments, Work should be introduced as a governed workflow system, not just as a personal productivity trick. Admins can decide who gets Work Cloud or Work Local, which apps are available, which actions need approval, and how permissions are inherited by roles [2]. Users should start with low-risk tasks, validate outputs, keep sensitive information out of task titles and shareable instructions, and treat drafts as drafts until reviewed.

The practical starting point

The best first week with ChatGPT Work is not a grand automation project. Start with one repeatable, familiar task: confirm an event, collect updates before a meeting, summarize new client feedback from Slack, or draft a follow-up based on recent Gmail messages. Connect only the apps needed for that workflow, write the desired output clearly, inspect what Work found, and approve only the final action after checking the details. That is where ChatGPT Work’s current design is strongest: it turns scattered context into a usable draft while keeping the user responsible for judgment and approval.

Sources from the last 72 hours

1. [1]Getting Started with ChatGPT Work · AI · 8news.aiAug 28, 2026, 8:51 PM UTC
2. [2]ChatGPT Work and Codex | OpenAI Help CenterAug 28, 2026, 1:35 AM UTC
3. [3]Scheduled tasks in ChatGPT | OpenAI Help CenterAug 28, 2026, 3:35 AM UTC
4. [4]Using Codex with your ChatGPT plan | OpenAI Help CenterAug 28, 2026, 10:35 PM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.