Agentic Workspace · ChatGPT Work
ChatGPT stops handing you
drafts.
Until now, ChatGPT's job ended at a suggestion — turning it into a finished file was still on you. The new ChatGPT Work mode plugs directly into tools like Slack and Google Drive and writes out finished spreadsheets and slide decks. No extra subscription required — a quiet change with real day-to-day impact.
An agent mode that
doesn't stop at a proposal
OpenAI rolled out "ChatGPT Work." Built on the existing ChatGPT Agent, it ships with a unified plugin directory spanning Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers, and it can stay with a task for hours to produce finished spreadsheets, slide decks, reports, and small web apps. Previously, ChatGPT's contribution stopped at a suggestion inside the chat window — turning that into a clean file, or pushing it into the tool where it actually lives, was still manual work. Work is built to take over that last mile. Details are laid out in OpenAI's Help Center article.
Under the hood it's powered by the GPT-5.6 model family, launched around the same time. Work itself isn't a new paid tier — it's bundled at no extra cost into the existing Plus, Pro, Business, and Enterprise plans. You can also set recurring tasks, such as reviewing Slack updates every Monday or refreshing a dashboard whenever new data lands, with the agent only pinging you when a human decision is actually needed.
Same pricing,
a wider job description
Adding capability to existing plans instead of cutting a new SKU reads as part of the same week's wider shift from price-based to time-based differentiation — alongside OpenAI's own sweeping API price cuts on August 1 and Gemini 3.7 Flash's discounted launch price. While rivals compete on price, Work is betting on differentiating by how much of the actual work it takes off your plate.
Who it helps, and how
Engineers & PMs
Routine work like rolling up weekly status updates or drafting release notes from Slack chatter can now land finished, right inside Slack. Fewer prompt round-trips means more time spent reviewing and correcting, less time assembling.
Business & leadership
Recurring account summaries or CRM-driven pipeline reports can be generated without routing through a human owner first. That said, a human check-and-verify step before anything goes external should stay in place.
Marketers
If your bottleneck has been manually pulling numbers scattered across tools into a slide deck, gathering and formatting the raw material can now be delegated. If you only use ChatGPT for brainstorming, the difference will feel minor.
What you get back isn't a suggestion — it's finished work.
What to do next
The practical starting point is picking one recurring report you already produce in the same format every week and handing it over. Set up a single recurring task, have a human check every field the first time to establish a baseline, and from the second run onward the job can shrink to a review pass. SmartCompany's coverage also points to the breadth of connected tools as the key comparison point against Microsoft Copilot and Google Workspace's own agent features. The realistic first move isn't switching platforms — it's trying Work as an extension of the Slack and Drive you already use.
On rollout order, it's safer to start with near-read-only recurring reporting work, and only extend write-capable tasks — anything that sends, updates, or deletes — once you have a way to review the execution log. Deciding up front on rules for expanding permission scope in stages makes it much easier to adjust the agent's footprint later.
Granting access means
you can't skip verification
Handing an agent access to Slack, Drive, and CRM systems also widens the blast radius of a mistake or a hallucinated fact. For anything that goes external — a client-facing slide, a report leaving the building — a human final-check step shouldn't be skipped. The feature is also still fresh: how it holds up on accuracy, cost, and permission management at large-organization scale over the long run is something to watch as real-world usage accumulates, not something the launch materials alone can settle.