OpenAI has launched ChatGPT Agent, a new capability that lets the conversational AI autonomously browse the web, run code and interact with third-party applications to complete multi-step tasks. The product integrates functions previously offered through the Operator and Deep Research previews and operates on a virtual computer that can shift between visual and text-based browsing, a terminal and API calls. The company began releasing the service on 17 July, giving immediate access to Pro subscribers and promising broader availability within days. On 24 July OpenAI said the rollout was finished for all paid Pro, Plus and Team accounts, apologising for a brief delay. Pro users receive 400 agent messages each month, while Plus and Team subscribers get 40, with extra usage available through a credit system. OpenAI has built in guardrails that require explicit user approval before actions with real-world consequences, and has added defences against prompt-injection attacks. The firm said it will extend access to Enterprise and Education customers in the coming weeks and intends to iterate on the feature set over time.