ChatGPT Work: the OpenAI agent that does the job for you

From chatbot to coworker: what OpenAI changed on 9 July 2026

For years, the promise of AI at work was essentially an assistant that answers questions very well. On 9 July 2026, OpenAI turned that page. With the simultaneous launch of ChatGPT Work and the GPT‑5.6 model, the company stopped selling answers and started selling execution.

ChatGPT Work is not a separate product or a new subscription. It is an agentic mode inside ChatGPT itself — alongside the "Chat" and "Codex" modes — available in the desktop app for Windows and Mac. OpenAI describes it as a "partner for your most ambitious work," and outside analyses classify it as an agentic coworker: something that does not merely suggest, but acts.

If you run or operate a company, that is not a semantic detail. It is a change of category.

What ChatGPT Work actually does

The list of capabilities OpenAI announced is straightforward: create documents, spreadsheets and presentations; build small web apps; organise projects; automate multi-step workflows. All of it from high-level instructions — without the user having to spell out every stage.

What sets Work apart from conventional ChatGPT is persistence. The agent can stay on a complex project for hours, breaking the work into stages and executing each one without constant intervention. It pulls data from connected applications — email, documents, task lists — and uses it to produce concrete deliverables, not just text generated in response to a prompt.

It is the difference between hiring an intern who answers emails and hiring a coordinator who runs a process from start to finish.

The architecture behind it: GPT‑5.6 in three variants

ChatGPT Work runs on GPT‑5.6, launched jointly on the same day. Independent technical guides describe three variants of the model available inside Work:

• Sol — aimed at more complex activities, with greater processing capacity.
• Terra — designed to balance performance and cost.
• Luna — optimised for greater resource economy.

Access to each variant depends on the plan. Free plan users get Work through the desktop app, but are restricted to the Terra variant. Plus subscribers can choose between Sol, Terra or Luna. Pro, Business and Enterprise plans get all three variants plus "ultra" mode — a higher performance tier within GPT‑5.6.

OpenAI has not published parameter counts, average latency or throughput for GPT‑5.6; that data remains unpublished. What is clear is the segmentation strategy: the more you pay, the more compute you reach — and therefore the more complex the workflow you can delegate to the agent.

Availability and rollout

On 9 July 2026, OpenAI made ChatGPT Work available on web and mobile immediately for the Pro, Enterprise and Edu plans. The Plus and Business plans got access over the following days. In the desktop app, Work mode was available from launch for every plan, including Free, with the variant and usage limits described above.

Guides published in the weeks after launch treat ChatGPT Work as a stable part of the portfolio — with no mention of discontinuation or later restriction. As of 4 August 2026, the feature is live globally.

What the commercial model reveals

OpenAI did not create a separate subscription for Work, nor a separately metered add-on. The feature is included in existing ChatGPT plans, consuming each plan's usage allowance. Independent analysis confirms that there is no standalone price for ChatGPT Work: the cost is the cost of the plan, and the limit is the plan's message or credit limit.

That choice says a great deal about the strategy. OpenAI does not want companies thinking about "hiring an agent." It wants them to stay inside the ChatGPT ecosystem and consume more and more — on pricier plans, with more powerful variants, for longer workflows.

Risks no manager can ignore

Usage reviews point to concrete limitations that need to enter the planning of any company adopting Work:

Human oversight is non-negotiable. The agent performs many actions autonomously, but it makes mistakes. Generated documents require review; automations need adjustment. Removing human review at this stage is a real operational risk.

Integrations set the agent's ceiling. ChatGPT Work depends on access to applications and files through integrations configured by the user. Without adequate read and write permissions across company systems, its automation capability drops sharply.

Poorly specified workflows produce inconsistent results. Early testing shows the agent may "slice" projects into stages that do not always follow the user's business logic, requiring manual refinement. The better defined the initial instruction, the more predictable the outcome.

Privacy and data retention: OpenAI has not published data policy specifics for Work mode beyond ChatGPT's general policies. Companies subject to LGPD, sector regulation or confidentiality agreements need to assess what they connect to the agent before automating.

What changes for businesses in Brazil

The most relevant comparison the dossier supports is with Claude Cowork, from Anthropic — identified in reviews as another "coworker agent" focused on professional workflows. Comparative performance metrics between the two were not published in the sources consulted, but the differentiator analyses point to for ChatGPT Work is its integration with the ChatGPT ecosystem already installed at millions of companies, and the official desktop app as a unified entry point.

For Brazilian SMBs — the audience I follow closely — the calculation is pragmatic. If your team already uses ChatGPT, Work is available with no additional adoption cost. The real investment is time and process: mapping which repetitive, structurable workflows can be delegated, defining how much oversight each one needs, and making sure the integrations with your systems are properly configured and audited.

The contest over corporate productivity tools has left the chatbot phase behind. What is at stake now is who will operate companies' workflows — and with what level of autonomy. ChatGPT Work is OpenAI's most concrete bet in that direction so far.

Sources

• OpenAI launches ChatGPT Work, an AI agent built to do your job tasks - G1
• ChatGPT Work: OpenAI's Agent That Delivers Work
• OpenAI releases the GPT-5.6 model generation and the ChatGPT Work agent
• ChatGPT is now an ally for your most ambitious work (ES)
• ChatGPT is now a partner for your most ambitious work
• ChatGPT is now a partner for your most ambitious work (PT-BR)