Glossary · Definition
AI agent
An AI agent is an LLM running in a loop: think → call a tool → observe the result → think again. The loop continues until the task is done or a stopping condition is hit.
Definition
An AI agent is an LLM running in a loop: think → call a tool → observe the result → think again. The loop continues until the task is done or a stopping condition is hit.
What it means
A minimal agent has: an LLM (Claude Sonnet, GPT-5, etc.), a set of tools the LLM can call (web search, file edit, code execution, MCP servers), and a control loop. Modern frameworks (Claude Agent SDK, OpenAI Agents SDK, LangGraph) handle the orchestration. The hard part is reliability over long horizons — by step 30, the model is reading 80k tokens of accumulated context and prior tool outputs, which most models start to drift on.
Advertisement
Why it matters
Agents are the post-chatbot interface for AI. Coding agents (Claude Code, Cursor Background, Devin), research agents (Perplexity Pro Search), and operator agents (ChatGPT Operator) all use the same loop pattern. Reliability differences between models compound across 30+ steps — 95% per-step accuracy = 22% end-to-end at 30 steps.
Related free tools
Frequently asked questions
Best model for agents?
Claude Opus 4.7 leads on long-horizon agentic reliability. Sonnet 4.6 is the cost-effective default. GPT-5 is competitive but drifts sooner past ~30 steps.
Cost?
Agent costs explode with horizon length because context accumulates. Use prompt caching aggressively. Estimate yours with the AI agent loop cost estimator.
Related terms
- DefinitionMCP (Model Context Protocol)MCP (Model Context Protocol) is an open standard for connecting AI assistants to external tools and data sources. Think USB-C for AI integrations: write a server once, it works in Claude, ChatGPT, Cursor, Zed, Goose, etc.
- DefinitionContext windowThe context window is the maximum amount of text (in tokens) an AI model can process in a single request — combining your system prompt, conversation history, and output. Past the limit, the model can't 'see' earlier content.
- DefinitionPrompt cachingPrompt caching is a feature where the AI provider stores frequently reused prompt prefixes (system messages, RAG context, few-shot examples) and bills cached reads at ~10% of normal input cost.