GPT agents

Autonomous Large Language Models (LLMs) that self-prompt, plan multi-step workflows, and execute actions across external tools (e.g., web browsers, APIs) to achieve complex, high-level objectives.

GPT agents are autonomous software entities: they execute complex, long-horizon tasks without human micro-management. This agentic loop involves four core steps: planning, tool selection, execution, and self-reflection. Unlike reactive chatbots, agents dynamically break down a fuzzy goal—like 'analyze Q3 market data and draft a report'—into distinct sub-tasks. They leverage tools such as a visual browser, a code interpreter, or external APIs (Gmail, GitHub) to act on a virtual computer. Major implementations, including the integrated ChatGPT Agent and the AgentKit SDK, demonstrate this shift from conversational AI to proactive, end-to-end automation.

https://openai.com/developers/agents