Generative AI tools are increasingly able to do things, not just answer questions. Instead of copying information into a chat window and asking for one response at a time, you can give some AI tools access to files, websites, apps, or other services and ask them to complete a multi-step task.
For example, an AI tool might:
Tools such as Claude Cowork and ChatGPT Work are examples of this shift. Claude Cowork can work with connected services, browsers, and—with permission—files and applications on your computer. ChatGPT Work can carry out longer, multi-step projects, create finished documents and other files, work with connected apps, and access local files you choose to share.
These capabilities are often described as agentic AI because the system can decide which steps and tools it needs to use to accomplish a task. That doesn’t mean it can be trusted to work completely independently. These tools can still make mistakes, misunderstand instructions, or take an action you didn’t intend.
Be especially careful about giving these tools access to private information, important files, or accounts where mistakes could have serious consequences. At the University of Arizona, personal ChatGPT and Claude accounts may be used only with public information.
So it's recommended to limit access, use sandboxed environments, plan tasks carefully, and check everything they produce. (see Agents in Academia below)
For lower-risk work, though, this can be very useful. You might ask an AI tool to organize notes from a project, combine information from several documents, turn research into a presentation, or help with a repetitive task that would otherwise require switching among several apps.
ChatGPT Work is available to free users through the desktop app but not on the web or mobile, while Claude Cowork requires a paid plan.