OpenAI Launches ChatGPT Work Agent Powered by GPT-5.6 and Codex

OpenAI has launched ChatGPT Work, an AI agent that can plan and carry out multi-step tasks on its own — not just answer questions. The product is built for ordinary workers, not developers, and is designed to produce what OpenAI calls “share-ready” outputs: documents, slide decks, spreadsheets, websites, and data analyses, all generated from a single high-level instruction.

The announcement marks a clear step away from chatbot-style AI and towards software that acts more like an autonomous digital assistant.

What ChatGPT Work Actually Does

Under the bonnet, ChatGPT Work runs on GPT-5.6 — OpenAI’s latest large language model — combined with Codex, the company’s code-focused AI system. That pairing matters. GPT-5.6 handles the language, reasoning, and planning; Codex handles the technical work behind the scenes, connecting tools, building websites, and automating workflows that would otherwise need a programmer.

Give it a goal — say, “pull together a weekly sales report from these files and email a summary to the team” — and ChatGPT Work breaks it into sub-tasks, draws up a checklist, and works through each step on its own. It can keep going for hours if the job demands it.

The agent can also use a computer. It navigates with its own cursor, opens Chrome, and interacts with apps and websites on the user’s device — a capability OpenAI calls “computer use” and “browser use.” That’s a long way from a text box that answers questions.

Scheduled Tasks and Business Integration

One of the more practical features is scheduling. Users can set ChatGPT Work to run a prompt or workflow automatically — daily, weekly, or on whatever timetable suits them. A marketing manager could, for instance, set it to compile a competitor summary every Monday morning without lifting a finger.

The product sits inside the ChatGPT ecosystem and connects with ChatGPT Business features: data analysis tools, canvas for document editing, app connectors, and custom workspace GPTs. OpenAI says that in business plans, customer data is not used to train its models — a point likely to matter to organisations with confidentiality obligations.

Access to GPT-5.6 and the full agent feature set requires a paid subscription, either ChatGPT Plus or a business tier.

The Broader Race to Build AI Agents

ChatGPT Work arrives as the AI industry pushes hard into what it calls “agentic” software — systems that act, not just advise. Comparable products are emerging across the sector, and the competition is fierce.

But not everyone is cheering. Technology experts and civil society groups have raised concerns about autonomous agents running with too little human oversight. Large language models don’t genuinely understand meaning — they generate text that sounds correct, which is not the same thing. An agent that acts on a plausible but wrong output could cause real problems, above all in professional settings where accuracy matters.

Sam Altman, OpenAI’s chief executive, has previously described the shift to agents as one of the most consequential changes in how AI will be used. “The thing that’s going to matter most,” he said in a public address earlier this year, “is AI that can do real work, not just talk about it.”

UK regulators are paying attention. The Information Commissioner’s Office has published guidance on generative AI and data protection, stressing that organisations must ensure lawful data processing and maintain human oversight. The Competition and Markets Authority has also flagged concerns about fairness and transparency in the AI foundation model market.

Data Protection and the Limits of Automation

Any organisation thinking of using ChatGPT Work needs to think carefully about UK GDPR. The ICO is clear: using AI tools to process personal data requires appropriate technical and organisational safeguards, and users can’t simply outsource accountability to the software.

There’s also the question of reliability. ChatGPT Work can produce a polished-looking report in minutes — but that report still needs a human to check it. The risk isn’t that the output looks wrong. It’s that it looks right.

What This Means for Kent Residents

Kent businesses — from small firms in Maidstone to professional services outfits in Tunbridge Wells — could use ChatGPT Work to cut time spent on routine document work, though anyone handling client data will need to check their obligations under UK GDPR before switching on the agent features. Public bodies such as Kent County Council and NHS Kent and Medway ICB would face additional hurdles: information governance approvals, risk assessments, and compliance with NHS England’s AI frameworks before anything like this could be used in administrative or clinical workflows. For individual residents, the main practical consideration is cost — the full feature set sits behind a paid subscription — and a reasonably fast broadband connection, since computer-use and browser-automation features will put more demand on your internet link than a simple chat window.