What People Mean When They Say 'AI Agent'

An 'AI agent' is just an LLM connected to tools — search engines, APIs, code interpreters, file systems — that the model can choose to use to accomplish a goal. The model loops: think → decide on a tool → call it → read the result → think again. Real autonomy is still limited, but the basic loop runs every minute on millions of computers now.

Some examples

• Claude Code, Cursor Agent, and Devin are coding agents — they read your codebase, write changes, run tests, and iterate.
• Manus and Genspark are general-purpose web agents — they browse, fill forms, and complete multi-step tasks online.
• OpenAI's Operator can use a real browser on your behalf to book restaurants, order groceries, or fill out forms.
• The simplest agent is a 5-line LangChain script: LLM + 'search the web' tool + 'do math' tool + a loop.

Try it!

On chatgpt.com, ask 'Use the web to research X, then write me a 3-paragraph summary with sources.' That's a 1-step agentic loop already running for you. Now imagine 50 steps. That's where the field is going.