What AI Agents Actually Are: LLMs Plus Tools, Loops, and Orchestration

AI agents are a frequently used term in tech, but their underlying mechanics are often misunderstood, even among developers. At their core, an AI agent combines a large language model with orchestration logic, contextual information, and predefined tools that extend its capabilities beyond simple text generation. The agent operates in a continuous loop, checking model responses for tool calls, executing the appropriate functions, and feeding results back into the conversation. A developer has published a minimal open-source AI agent built with TypeScript and Node.js to demonstrate this architecture in practice, covering the agent loop, tool definitions, and guardrails. The project is intended as an educational reference rather than a production-ready solution, aiming to demystify how agent systems are constructed from first principles.