ChatGPT is no longer just a place to ask a question and receive a paragraph in return. Over the past year, OpenAI has been pushing the product towards something more ambitious: an assistant that can research, work with files, write code, use connected tools and help complete multi-step tasks. The change is important, but it is also easy to overstate. ChatGPT can take on more of the work than it once could; it still needs direction, access controls and a human to check consequential decisions.
The shift is clearest in the newer tools built around longer-running work. ChatGPT Work, introduced in July 2026, is designed for tasks that involve several stages rather than a single response. It can research and analyse information, work across connected apps and files, and produce documents, spreadsheets, presentations and reports. Users can follow its progress, answer questions and approve important actions. That is a different proposition from asking a chatbot to summarise a page.
From answers to finished work
For a small business owner, that could mean gathering information from several files and turning it into a first-draft report. A developer might use Codex to investigate a bug, make a code change and run checks. A researcher could use Deep Research to assemble a sourced overview before reviewing the evidence and writing their own conclusions. The useful distinction is that these tools are intended to move a task forward, not simply describe how a person might do it.
OpenAI’s September 2026 announcement of GPT-6 Astra placed particular emphasis on coding, research, computer use and complex multi-step work. The company says Astra can create documents, spreadsheets and presentations that follow a user’s instructions and templates. Those are company-reported capabilities, not a guarantee that every output will be correct or ready to publish. Results still depend on the quality of the instructions, the source material and the checks applied afterwards.
The rise of always-on AI agents
Another development is Dots, announced on 29 September 2026. OpenAI describes a Dot as an always-on agent with its own cloud computer and browser. It can work towards an ongoing goal, use connected apps and return with progress or decisions that need a person’s judgement. OpenAI says the system can connect to more than 4,000 apps through its plugin ecosystem. Availability is being rolled out gradually, so access should be checked against the user’s plan, region and workspace.
This is where the promise becomes practical—and where caution matters. An agent that can monitor a workflow or prepare a report may save time. An agent with access to email, customer records or financial systems can also make a mistake with real consequences. Users should give it only the permissions it needs, require approval for sensitive actions and review what it has done. “Always on” should not mean “left unchecked”.
What about voice, images, files and research?
ChatGPT remains a collection of tools as much as a single assistant. Voice makes it possible to speak naturally rather than type every request. Image tools can create or edit visuals. File and spreadsheet analysis can help people inspect information without manually copying every row into a prompt. Search and Deep Research can help gather current information, while Codex is aimed at software development. Which features appear, and how much they can be used, depends on the plan and the current product rollout.
For readers, the sensible approach is to start with a recurring task. Ask ChatGPT to organise a set of meeting notes, compare a few documents or prepare a first draft from approved source material. Check the result against the original files. If the tool saves time without creating more checking work than it removes, it has earned a place in the workflow. There is little reason to pay for advanced features that are rarely used.
Is ChatGPT now a digital employee?
Not quite. It can act like a capable assistant on defined tasks, but it does not carry the accountability, context or professional judgement of an employee. It may misunderstand an instruction, overlook a detail or produce a confident answer based on incomplete information. A person still needs to set priorities, protect confidential data, verify facts and take responsibility for the final decision.
The bigger change is that AI is moving beyond the chat box and into the tools people already use. ChatGPT can now help with research, coding, analysis and the production of work across connected services. That makes it more useful than a conventional chatbot for many people. The best results, however, are likely to come from treating it as a supervised partner: give it a clear job, limit its access, inspect the work and keep the final judgement human.
What readers should know
- ChatGPT Work is designed for longer tasks that can produce finished documents and other deliverables.
- GPT-6 Astra was announced in September 2026 with a focus on coding, research and computer use.
- Dots are OpenAI’s always-on agents, announced on 29 September 2026, with gradual account rollout.
- Features, usage limits and availability vary by plan, region and workspace settings.
- Human review remains essential for factual accuracy, sensitive information and actions with real-world consequences.