ChatGPT as an agent
OpenAI merged an acting browser, deep search and conversation into one mode: the system got its own computer and carried out multi-step tasks by itself.
Why it matters
The agent stopped being a separate product and became a mode of the same application hundreds of millions already use.
These had been three different things: Operator pressed buttons in a browser, deep research read sources and wrote a report, ChatGPT talked. Here they became one: the model decides for itself when to look at a page, when to run code, when to ask the person. Tasks like taking a file from a repository and putting it in a folder ran unattended. The risks came with it: a system that acts on what it sees is open to instructions hidden on a page, and OpenAI described that directly.