ChatGPT Work Explained: What Changes for Your Business?

ChatGPT Work is an agent inside ChatGPT that takes a finished outcome as its instruction — not a prompt — then gathers context from a user's connected apps and files, breaks the goal into steps, and delivers a completed document, spreadsheet, presentation, report, or website.

TL;DR

On July 9, 2026, OpenAI launched ChatGPT Work, running on its new GPT-5.6 model family. Instead of answering one message at a time, the agent connects to a user's existing tools — Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRM platforms, and project-management software — reads what it needs from them, and works independently, sometimes for hours, until the job is done. Access began July 9 for Pro, Enterprise, and Edu plans on web and mobile, with Plus and Business following in the days after; a new desktop app shipped the same day on every plan, including Free. This is a live product launch, not a preview, as of July 9, 2026 — but the claims about how well it performs on real business workflows are still mostly OpenAI's own.

This matters because ChatGPT Work is the clearest signal yet that the "AI agent" conversation has moved from writing text to finishing tasks. For any team that already routes work through connected software — a CRM, a shared drive, a ticketing queue — this is the first mainstream product built to plug into that stack directly rather than sit beside it in a chat window.

What Actually Happened on July 9

OpenAI unveiled ChatGPT Work as an agent designed for longer assignments instead of one-off prompts. It gathers information from a user's connected apps and workflows, breaks a larger project into smaller steps, and can keep working on those steps when the user has stepped away from the computer or phone. The output is not a chat reply — it is a finished document, spreadsheet, presentation, report, or website.

The launch bundled several previously separate pieces into one release: a new ChatGPT desktop app (the old desktop app was renamed ChatGPT Classic), a hosted "Sites" feature that turns a project into a shareable web app — dashboards, launch calendars, project trackers, internal portals — Scheduled Tasks that run recurring work on a trigger or a schedule, and a Computer Use mode that lets the agent click, type, and move files on a user's behalf while it works in the background. OpenAI's standalone Codex app was folded into the new desktop client, and the Atlas browser is being sunset as part of the same consolidation.

Connectable surfaces named at launch span Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRM platforms, and project-management tools. US Tech Automations teams that already route client intake, ticket triage, or reporting through connected software will recognize the pattern immediately — it is the same "read from the tools you already use" model that connected-app automations have run on for years, now built into the chat interface itself.

ChatGPT Work at a glance

How ChatGPT Work Actually Works

The mechanism is simpler than the marketing around it. A user gives ChatGPT Work an outcome — "put together this month's client report" or "turn these interview notes into a project brief" — rather than a single instruction. The agent then pulls in whatever context it needs from connected apps and files, plans a sequence of steps to reach that outcome, executes them one at a time, and checks its own progress against the original goal. Because it is allowed to keep running instead of waiting for the next message, a single request can occupy the model for minutes or hours, not seconds.

That is the practical difference from a normal ChatGPT conversation: a chat reply ends when the model stops generating text. A ChatGPT Work run ends when the deliverable exists — a filled-in spreadsheet, a formatted slide deck, a working internal site — which is why OpenAI frames the unit of work as a project rather than a prompt.

Enterprise-side, OpenAI built in the controls larger customers ask for before connecting an agent to real company data: administrators can manage which tools, data sources, and actions are available to their organization, a Compliance API gives visibility into what the agent did and said, and an auto-review step checks sensitive actions before they execute. OpenAI has also said its internal adversarial testing blocked attempts to extract protected data during those tests — worth noting plainly that no independent validation of that specific claim has been published yet, according to AppleInsider, which reported on the launch's enterprise controls in detail.

Why Now: What Constraint Broke

Two things had to be true for a product like ChatGPT Work to work at all: a model capable of holding a multi-step plan across a long task without losing the thread, and enough integration surface area that "connect to your tools" meant something more than a single Slack bot. GPT-5.6 is OpenAI's answer to the first constraint — the model family underneath ChatGPT Work, ranging from a lighter Terra tier through Sol and Luna to an Ultra reasoning mode for the heaviest jobs. According to Enterprise DNA, OpenAI had already begun charging for workspace-agent runs on a token basis starting July 6, 2026 — three days before the public ChatGPT Work launch — which is a quiet tell that usage-based billing, not just model capability, was part of what had to be in place first.

The second constraint — integration depth — is why the connected-apps list matters more than any single feature. A model that can plan five steps is not useful if it can only read the message it was given; a model that can pull a client's deal stage from a CRM, a document from a shared drive, and a thread from Slack in the same run is a materially different tool. That is also where automation teams already live: teams that route incoming requests through US Tech Automations workflows today are, in effect, already doing the "gather context from connected systems, then act" pattern that ChatGPT Work is now building into the chat interface itself.

Who Shipped It — and Who Else Already Has

ChatGPT Work did not arrive into an empty market. According to BNN Bloomberg, the launch follows Anthropic's Claude Cowork and Microsoft's Copilot Cowork, both of which shipped comparable agentic work tools earlier — and Anthropic's launch had already stirred stock-market concern about AI-driven disruption to data-analytics jobs and firms. ChatGPT Work is OpenAI's entry into a category three major labs are now racing to define, not a category it created alone.

The same report notes the launch lands during a period of aggressive OpenAI balance-sheet activity: according to BNN Bloomberg, OpenAI accepted a $520 million credit facility from Bank of America ahead of what the company is reportedly positioning as IPO preparation. Washington Crossing Advisors senior portfolio manager Chad Morganlander addressed the broader AI investment boom in the same segment covering the launch — a reminder that ChatGPT Work is being read by markets as a business-model signal, not just a product update.

Rollout timeline

Sources: Enterprise DNA; AppleInsider.

What OpenAI Claims vs. What's Verified

Most of the launch's headline numbers come from OpenAI itself, which is a distinction worth holding onto. OpenAI says more than 5 million people now use Codex every week, with more than 1 million of them working outside software development, according to AppleInsider — a vendor-reported adoption figure for the product ChatGPT Work's desktop app absorbed, not an independently audited count. Similarly, according to PYMNTS, GPT-5.6 Sol is 54% more token-efficient on agentic coding jobs — again OpenAI's own benchmark framing, reported by PYMNTS as part of its launch coverage. PYMNTS does not state what baseline that 54% is measured against, so we do not attribute it to any particular prior model.

None of that makes the figures false — it means they should be read as "OpenAI says," not as third-party-verified performance data, until an outside benchmark confirms them.

Sources: AppleInsider; PYMNTS; BNN Bloomberg; Enterprise DNA.

Signal vs Speculation

Everything above this section is what has been reported and can be linked to a source. What follows is our forecast, clearly separated from the facts.

Our read: if the connected-apps model holds up under real business use — not demo use — the biggest near-term shift is not "AI writes better documents." It is that the software a small or mid-size business already runs (its CRM, its shared drive, its ticketing system) becomes the input layer for an agent, instead of a destination the human has to manually visit. Over the next 12 to 36 months, we expect the deciding factor for adoption to be less about model quality and more about governance: which businesses have clean enough data and clear enough access controls in their connected systems to let an agent act on their behalf without creating a compliance mess. Businesses that have already invested in structured, permissioned automation — rather than ad hoc scripts and manual exports — are the ones positioned to adopt something like ChatGPT Work quickly. Businesses that haven't will likely spend the next year on data cleanup before they can use it safely at all.

Our read: the industries most affected first will not necessarily be the ones with the most document volume. They will be the ones where "gather context, then produce a finished deliverable" already describes a recurring, well-defined task — client reporting in marketing agencies, transaction paperwork on real estate teams, candidate and requisition tracking in recruiting agencies. We cover each of those in more depth below.

What This Means Across Industries

The mechanics above apply everywhere, but the workflow-level impact differs by industry, current tech stack, and how much of the work is already structured data versus scattered documents. We've broken out three implications pieces from this same signal:

• What ChatGPT Work Means for Marketing Agencies — client reporting, campaign status updates, and the sales workflows most exposed first.

• What ChatGPT Work Means for Real Estate Teams — transaction coordination, lead qualification, and where real estate-specific CRM data helps or hurts.

• What ChatGPT Work Means for Recruiting Agencies — candidate pipeline updates and the recruitment tasks most likely to get automated first.

Key Takeaways

• ChatGPT Work launched July 9, 2026, running on OpenAI's new GPT-5.6 model family, and produces finished documents, spreadsheets, presentations, reports, and websites rather than chat replies.

• Access began with Pro, Enterprise, and Edu plans on web and mobile, with Plus and Business following days later; the new desktop app shipped on every plan, including Free, the same day.

• It follows Anthropic's Claude Cowork and Microsoft's Copilot Cowork to market — OpenAI is the third major lab shipping this category, not the first.

• Several headline figures — 5 million+ weekly Codex users, a 54% token-efficiency gain — are OpenAI's own reported numbers, not independently verified benchmarks.

• The businesses best positioned to adopt it quickly are the ones that already run clean, permissioned, connected-app workflows rather than manual exports and ad hoc scripts.

FAQ

What is ChatGPT Work?

ChatGPT Work is an agent inside ChatGPT, launched July 9, 2026, that takes an outcome rather than a prompt, pulls context from a user's connected apps and files, and completes multi-step projects into finished documents, spreadsheets, presentations, reports, or websites.

Which ChatGPT plans can use it right now?

As of July 9, 2026, web and mobile access started with Pro, Enterprise, and Edu plans, with Plus and Business plans gaining access in the following days; the new desktop app is available on every plan, including Free.

What model powers ChatGPT Work?

It runs on OpenAI's GPT-5.6 model family, which includes Terra, Sol, and Luna tiers with adjustable effort levels, plus an Ultra reasoning mode available to Pro and Enterprise users.

Does ChatGPT Work replace ChatGPT Enterprise or Codex?

It absorbs Codex's standalone desktop app into the new unified ChatGPT desktop client and renames the prior desktop app ChatGPT Classic; it is positioned as an added agent capability inside ChatGPT rather than a separate product line.

Is ChatGPT Work safe to connect to sensitive company data?

OpenAI built in administrator controls over connected tools and data, a Compliance API for visibility into agent actions, and an auto-review step for sensitive actions, but the company's claim that adversarial testing blocked attempts to extract protected data has not been independently validated yet.

How is ChatGPT Work different from Claude Cowork or Copilot Cowork?

All three are agentic "finish the work, not just the chat" products from competing labs; as of this writing, detailed head-to-head performance comparisons between them have not been independently published, so the honest answer is that the category is new enough that no verified winner exists yet.

Will ChatGPT Work cost more than a standard ChatGPT plan?

OpenAI introduced token-based pricing for workspace-agent runs starting July 6, 2026, ahead of the public launch, and is shifting at least one integration (PowerPoint, for Enterprise) from free to token-based billing on August 6, 2026 — so complex, longer-running work is expected to consume more of a plan's included usage than a standard chat.

Teams evaluating whether to build around ChatGPT Work, a competing agent, or a connected-app automation layer that isn't tied to any single model vendor can see how that decision plays out in practice on the agentic workflows platform page.