Building Your First Autonomous AI Agent in Python (Under 100 Lines)

Building Your First Autonomous AI Agent in Python (Under 100 Lines) You've used LLM APIs. You've called Claude or GPT in your scripts. But have you built an agent that actually runs on its own? The difference matters. An API call is a one-shot: you ask a question, get an answer, you're done. An agent is a loop. It reads a task, calls the LLM, executes the LLM's decisions (tool calls), and repeats until it decides the job is complete. This article builds one. Really builds one. No frameworks, no abstractions, just Python and the Anthropic API. What Makes Something an Agent? Three pieces: an LLM, a loop, and tools. The loop is the key. Your code doesn't decide when the agent stops. The LLM does. It looks at the task, the current state, and says "I'm done" (stop_reason: end_turn with no tool calls) or "I need to do this next" (returns a tool call). Your job is to keep the loop spinning until the LLM says stop. The Minimal Agent Loop Here's a real agent with three tools: read_file, write_f