OpenAI just launched ChatGPT Work to handle your multi-hour projects
OpenAI just rolled out ChatGPT Work, a new autonomous agent powered by its latest GPT-5.6 model. It connects directly to workplace apps like Slack, Teams, and Google Drive, breaking down complex tasks into smaller steps that run for hours in the background. The agent can create docs, decks, and full web apps via Sites, with rollouts starting for Pro, Enterprise, and Edu users.

For the past two years, using AI at work has largely meant being a professional copy-paster. You grab a messy Slack thread, feed it to a chatbot for a summary, and manually paste the output into your slide deck. OpenAI is finally trying to kill that middle step.
OpenAI just launched ChatGPT Work, a new autonomous agent powered by its latest frontier model suite, GPT-5.6. The public rollout comes roughly two weeks after GPT-5.6 was held up in regulatory drama and restricted to government-approved groups. Following official clearance from the Trump administration, CEO Sam Altman wasted no time calling it "the best model we have ever produced."
Turning a developer tool into an everyday operator
Under the hood, ChatGPT Work takes Codex—OpenAI's coding agent that already sees over 5 million weekly users, including 1 million non-developers—and turns it into a general-purpose desk worker. In fact, OpenAI is sunsetting the standalone Codex app entirely, merging it into the unified ChatGPT desktop client alongside a newly integrated browser and background Computer Use capabilities.
Powering this setup is GPT-5.6, which arrives in three tiers: Sol, Terra, and Luna. OpenAI is positioning the flagship Sol as a powerhouse for deep reasoning, science, and computer execution, while notably pitching it as a more cost-effective option against heavyweight rivals. The standalone Atlas browser is also being phased out, with its agentic browsing skills folded directly into desktop and the Chrome sidebar.
Rather than just firing off quick text answers, ChatGPT Work can stay on a project for hours. It breaks broad goals down into multi-step actions, pulling reference material through a unified plugin hub that links to Slack, Microsoft Teams, Google Drive, Gmail, Salesforce, and enterprise CRMs. Users can explicitly summon context using @-mentions or let the model route queries on its own.
There is also a new Scheduled Tasks feature, allowing the agent to run recurring workflows unattended. It can monitor customer feedback emails, crawl web dashboards every morning, or refresh recurring meeting slides based on weekly chat logs. OpenAI claims nearly 100% of its internal teams already rely on the tool, noting that sales reps built working proofs of concept in 24 hours instead of weeks, while finance teams slashed month-end forecasting from days to hours.
Access is surprisingly wide. The updated desktop app on Mac and Windows is available immediately across all tiers, including Free users. Web and mobile rollouts are live now for Pro, Enterprise, and Edu plans, with Plus and Business users getting access over the coming days.
Why it matters
We think ChatGPT Work signals the moment frontier AI officially shifts from being a conversational sounding board to an end-to-end task execution engine.
For knowledge workers and cross-functional teams, the practical payoff is massive. Instead of juggling tabs and manually wrangling spreadsheet numbers into presentation templates, you can hand off the operational grunt work and step in only for steering and approvals. If your job involves synthesizing messy communication into deliverables, it is absolutely worth installing the desktop app today to test recurring tasks.
Strategically, OpenAI is taking direct aim at Anthropic's Claude Cowork while blunting momentum from open-source plays like OpenClaw. By rolling Codex, native browsing, and Computer Use into a single desktop package—and making it free on desktop—OpenAI is aiming to own the OS-level workplace interface before operating systems build their own.
The clever move is repurposing battle-tested developer workflows for business tasks. But the lingering question is reliability. Giving an AI permission to click around your desktop, access corporate comms, and edit shared drives in the background comes with real security and hallucination risks. Enterprise safeguards like admin controls and auto-reviews help, but convincing risk-averse teams to let an autonomous agent roam unsupervised remains a steep hill to climb.
Sources (2)
- openai.com(opens in a new tab)Originalopenai.com
- theverge.com(opens in a new tab)Reporttheverge.com