OpenAI just dropped ChatGPT Work, and it's the most ambitious product move we've seen from them since GPT-4. This isn't a chatbot upgrade or a reasoning model flex. It's a full-stack agentic productivity system that can take a goal, break it into steps, execute across your apps for hours, and deliver finished work—slides, sheets, docs, web apps.
The headline feature? ChatGPT can now act on your behalf across connected tools, stay with complex projects independently, and handle entire workflows end-to-end. If you've been waiting for agents to graduate from demos to shipping, this is that moment.
What ChatGPT Work Actually Does
ChatGPT Work is an agent mode inside ChatGPT that integrates with your existing productivity stack. You connect plugins for Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and other tools. Then you give it a task.
The agent can gather context from multiple sources, create artifacts (presentations, spreadsheets, documents, even interactive web apps), and execute multi-step workflows without constant supervision. You stay in control—you can follow progress, answer questions, change direction, and approve important actions.
What's genuinely new here is persistence and scope. ChatGPT Work can stay with a project for hours, breaking complex goals into subtasks and completing them independently. It's not just answering questions or generating text. It's moving files, updating sheets, drafting slides, and routing work between systems.
OpenAI claims nearly 100% of internal teams—finance, sales, operations—now use ChatGPT Work and Codex. In sales, it turned a discovery call into a tailored proof-of-concept in 24 hours (normally weeks). In finance, it cut month-end close from days to hours by finding source data, reconciling it in Excel or Sheets, creating slides, and verifying results.
GPT-5.6: The Model Powering It All
ChatGPT Work runs on GPT-5.6, a new frontier model OpenAI released alongside this announcement. The model is optimized for multi-step reasoning and creating materials that follow templates and reference files.
This is a big deal. Most agent frameworks struggle with long-horizon tasks because models drift, lose context, or produce outputs that don't match your format. If GPT-5.6 can actually maintain coherence across hour-long workflows while respecting style guides and templates, that's a genuine capability leap.
We don't have independent benchmarks yet, but the internal use cases suggest it's working. The finance and sales examples aren't trivial—reconciling data sources and maintaining context across a multi-day POC build are exactly the places agents usually fail.
Scheduled Tasks: The Quiet Power Feature
Buried in the announcement is Scheduled Tasks, which might be more transformative than the headline agent mode. You can ask ChatGPT to perform actions once, on a schedule, when events occur, or when monitoring detects changes.
Examples from the announcement:
- Review new Slack updates weekly and refresh a recurring meeting agenda
- Check websites and dashboards each morning, summarize changes, send a report
- Monitor customer feedback and turn themes into prioritized product ideas
- Update a presentation when new feedback arrives by email
This is the classic agent value prop: take repetitive tasks off your plate so you can focus on higher-leverage work. The difference is that OpenAI is shipping it as a first-class feature with enterprise controls, not a hacky automation script.
If Scheduled Tasks actually works reliably, it's a wedge into every knowledge worker's daily workflow. That's the territory Zapier, n8n, and similar automation tools occupy today—but with natural language configuration and LLM reasoning to handle messier, less-structured tasks.
Desktop Integration: Computer Use Ships for Real
On desktop, ChatGPT Work gets significantly more powerful. The new ChatGPT desktop app (merging the old Codex app) can use local files and apps, plus it includes a built-in browser for web-based work.
Computer Use—OpenAI's version of what Anthropic pioneered with Claude—lets ChatGPT control your computer on your behalf. It can click, type, move files, and execute tasks across apps and browser. You can use it for one-time tasks or as part of Scheduled Tasks for recurring work.
This is where the agent vision gets real. If ChatGPT can reliably open a spreadsheet, pull data from three different web dashboards, reconcile it, create a slide deck, and save it to the right folder—all while you're in a meeting—that's not incremental productivity. That's a different way of working.
The Codex app is merging into the desktop experience, bringing inline diff editing, pull request review in the side panel, faster Computer Use powered by GPT-5.6, and multi-repository support. For developers, this positions ChatGPT as a unified agent environment spanning both code and general work.
Sites: Turning Work Into Interactive Apps
OpenAI is also launching Sites in public beta. With Sites, you can turn work or ideas into an interactive web app and share it via URL. Use cases include live dashboards, project trackers, launch calendars, prototypes, internal portals, and interactive reports.
You can test Sites inside ChatGPT, and ChatGPT can update them as underlying information changes. This is the logical endpoint of the artifacts paradigm Claude popularized: instead of generating static outputs, the agent creates living, interactive tools.
The autonomous update capability is the key unlock. If you can ask ChatGPT to maintain a live command center that updates as account activity changes, you're offloading an entire category of manual knowledge work.
Enterprise Controls: The Shipping Prerequisite
For agents to ship in enterprises, governance and security aren't optional. OpenAI knows this. ChatGPT Work builds on the compliance, privacy, and workspace management foundation of ChatGPT Enterprise.
Enterprise and Edu admins can centrally manage access, control what company context ChatGPT can use, specify which tools it can connect to, and define what actions it can take. The Compliance API provides visibility into conversations and actions at scale.
Controls are tailored by environment. On web, admins manage plugin access, configure browser use and network access, and restrict sensitive actions. On desktop, it builds on Codex's enterprise governance model.
This is the boring infrastructure work that determines whether agents get adopted or blocked by IT. OpenAI is clearly prioritizing enterprise deployment—ChatGPT Work launches today for Pro, Enterprise, and Edu plans, rolling to Plus and Business over the next few days. The desktop app is available on every plan, including Free.
The Real Test: Can Agents Actually Ship?
The last two years have been full of agent demos. AutoGPT, BabyAGI, the open-source agent explosion, Devin, Anthropic's Computer Use, countless startups. The pattern has been consistent: impressive demos, hype cycles, then reality checks when agents hit complex real-world tasks.
ChatGPT Work is the first serious attempt by a frontier lab to ship agents as a mainstream productivity product. The distribution is there—ChatGPT has 300+ million weekly users. The model capability (GPT-5.6) is purpose-built. The integrations are deep. The enterprise controls exist.
If it works, it's a new category. If it doesn't—if users hit the same reliability, context-drift, and error-recovery problems that have plagued every agent framework—we'll learn a lot about where the real capability ceiling is.
The internal adoption claims are promising but self-reported. The sales and finance examples are compelling but curated. What matters is whether hundreds of thousands of paying customers can hand ChatGPT Work a real, messy, multi-hour task and get usable output without constant babysitting.
What This Means for the Agent Ecosystem
OpenAI shipping a full-stack agent product puts pressure on everyone else. Anthropic has Computer Use but hasn't bundled it into a productivity suite. Google's Gemini has some agent capabilities but nothing this comprehensive. Microsoft has Copilot, but it's more assistant than autonomous agent.
For startups building agent frameworks, vertical agents, or workflow automation tools, ChatGPT Work is now the baseline. You need to be meaningfully better, cheaper, or more specialized to justify switching costs.
The plugin ecosystem is particularly interesting. OpenAI is positioning ChatGPT as the orchestration layer, with third-party tools providing domain-specific capabilities. If that architecture works, it's a platform play—ChatGPT becomes the agent runtime, and everyone else builds on top.
Open Questions
A few things we don't know yet:
- Reliability at scale: Can GPT-5.6 actually maintain coherence across multi-hour workflows with dozens of steps? Or will users hit the same error-recovery and context-drift issues that plague existing agent systems?
- Error handling: What happens when a Scheduled Task fails? How does ChatGPT communicate problems, and how much manual intervention is required to get things back on track?
- Cost structure: Agentic workflows that run for hours and make dozens of API calls could get expensive fast. OpenAI hasn't detailed pricing for heavy Work usage.
- Trust threshold: Will users actually trust ChatGPT to take consequential actions unsupervised? Or will paranoia about mistakes keep most users in high-supervision mode?
The real answer will come from watching how users actually adopt this over the next few months. If Scheduled Tasks shows up in everyone's daily workflow and Computer Use becomes table stakes for knowledge work, we'll know agents crossed the chasm.
If ChatGPT Work stays a power-user feature that most people try once and abandon, we'll know the capability gaps are still too wide.
Either way, OpenAI just forced everyone's hand. The agent future is here—now we find out if it actually works.