Isla Fulford, a researcher at OpenAI, anticipated the success of Deep Research even before its official release. Fulford contributed to the development of this artificial intelligence agent, which independently navigates the internet, choosing which links to follow, what content to read, and compiling the information into comprehensive reports. Initially, OpenAI rolled out Deep Research for internal use; during periods when it was offline, Fulford received numerous inquiries from colleagues eager for its return. She noted that the volume of direct messages she received indicated strong interest in the tool.
Since its public launch on February 2, Deep Research has gained popularity among external users as well. Patrick Collison, CEO of Stripe, acknowledged its efficacy on social media shortly after its release, noting it had produced six reports in a single day and congratulating the team behind it. Dean Ball, a fellow at George Mason University specializing in AI policy, remarked that Deep Research had significantly engaged the policy-making community in Washington, D.C.
The tool is available through the ChatGPT Pro plan, which is priced at $200 per month. Users can input queries such as “Write me a report on the Massachusetts health insurance industry” or “Tell me about WIRED’s coverage of the Department of Government Efficiency,” and the AI will devise a research plan. It searches for relevant websites, examines their content, and determines which links to explore further. After spending several minutes gathering information, it synthesizes its findings into a comprehensive report that may include citations, data, and charts.
Many AI agents available today function primarily as chatbots linked to basic programs lacking in complexity. By contrast, Deep Research engages in a form of artificial reasoning to formulate and execute a research plan, providing insights into its decision-making process in a separate window. Josh Tobin, another OpenAI researcher involved in the project, finds value in reviewing these reasoning paths to understand the model’s thought process.
OpenAI envisions Deep Research as a scalable tool that can handle various office tasks. Tobin notes that the AI could be trained to perform specific white-collar duties, such as generating reports or presentations from internal company data. The long-term objective is to develop an agent proficient not only in compiling reports but also in tackling a diverse range of tasks.
Interestingly, Tobin and his team discovered that many users employ Deep Research to generate code, a use case for which it was not intentionally designed. Tobin remarked on this unexpected application, expressing the team’s curiosity about its implications.
