OpenAI Expands Agentic Research Tools

OpenAI released Deep Research for ChatGPT in February 2025, pushing language models from answering one prompt toward carrying out longer research workflows across many sources.

A research agent has to plan searches, open and compare documents, keep track of intermediate findings, revise its approach, and produce a synthesis with citations. That is a different systems problem from ordinary chat because the model interacts repeatedly with external tools and must preserve state over a longer task.

The release helped make “agentic” AI a practical product category. Reliability became a central issue: an agent that can take dozens of steps has dozens of opportunities to choose a bad source, misunderstand evidence, or follow an unproductive path.