History
Operator carries out tasks in a browser
OpenAI introduced an agent that could view and operate web pages.

On 23 January 2025, OpenAI introduced Operator as a research preview. The agent used its own browser to view pages, type, click and scroll. At launch, access was limited to Pro users in the United States. The announcement describes that first version.

Operating a browser is a different kind of assignment from writing an answer. The browser became a workplace for the agent. For businesses, that raised a different question: which actions do you actually want to hand over?

A small or medium-sized business can investigate this with a practice form. Ask an agent to fill in specified fields and stop before submitting. Then check every field. Agree beforehand what should happen if a button is missing or a page changes unexpectedly.
Judge such a trial on the complete process, including recovery after a wrong click. One possible next step is an internal task with a clear endpoint. Keep payments, orders and sending messages outside the first trial. This lets you investigate usefulness without treating a successful demonstration as proof for every browser process.