OpenAI Introduces ChatGPT Agent

OpenAI introduced ChatGPT agent in July 2025, combining browser interaction, code execution, research, and other tools into a system designed to complete multi-step tasks rather than only answer questions.

An agent has to maintain a plan across many actions, inspect the result of each tool call, recover when a webpage or command behaves unexpectedly, and decide when to ask the user before taking a consequential action. That makes reliability a workflow problem rather than only a language-model problem.

The product reflected a broader change in AI interfaces. Models were moving from generating suggestions for humans to execute toward operating software directly under delegated permissions. That makes confirmation boundaries, authentication, audit logs, and protection against malicious webpage instructions increasingly important.