OpenAI's browser-using agent launched as a US Pro-only research preview in January 2025 and was folded into ChatGPT six months later.
The durable lesson is not the product, which no longer exists under that name. It is the pattern: agentic features now arrive as narrow previews, get renamed, and get absorbed. Building a team process on one is a good way to rebuild it twice.
An agent, here, means software that operates a browser the way a person does — clicking, typing, scrolling — rather than calling an API. That matters because most sales tools have no usable API, so a clicking agent can in principle reach systems an integration cannot.
Operator shipped as a research preview, which is OpenAI's term for something deliberately incomplete and released to learn from. On 17 July 2025 OpenAI folded it into ChatGPT as agent mode and sunset the standalone site.
The short version: the thing that browses for you is now a mode inside ChatGPT, not a separate product, and the original is gone.
OpenAI released Operator on 23 January 2025 as a research preview of an agent that uses its own browser to perform tasks — typing, clicking and scrolling through web pages.
Availability at launch: Pro users only, United States only.
Status: Research preview.
Superseded: On 17 July 2025 OpenAI stated Operator is fully integrated into ChatGPT as agent mode, and the standalone operator.chatgpt.com site was sunset.
Price: Not separately priced — bundled into the Pro subscription.
Enterprise controls: Not stated at launch.
What it doesn't do: at launch it could not be relied on for anything consequential — OpenAI said so explicitly in calling it a preview with limitations. It also does not solve authentication or permissions: an agent browsing on your behalf inherits whatever access your session already has, which is a security question rather than a capability one.
SDR — Speculative for now. The obvious use — pulling prospect detail out of sites with no export — is exactly the use most likely to breach a site's terms, so check before you automate anything.
AE — Little immediate application. Agentic browsing suits repetitive, structured retrieval, and AE work is neither.
Manager — The governance question arrives before the productivity does. An agent acting in a browser under a rep's logged-in session is acting with that rep's permissions, and that is worth a policy before it is worth a pilot.
RevOps — Least affected operationally, most affected strategically. Note the pattern for procurement: US-only, top-tier-only previews are not a foundation to standardise on.
Before automating account research, be precise about what you actually need from a page. Vague retrieval produces vague briefs whether a human or an agent does it.
Two minutes, once a week. What changed in AI, and what to run because of it.