How to Build a General-Purpose AI Agent in 131 Lines of Python

2026-03-24 · Source: AI & ML – Radar · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

This article details how to construct two distinct AI agents—a coding agent and a search agent—using minimal Python code, specifically 131 lines for the coding agent and 61 lines for the search agent. It emphasizes that coding agents, when granted shell access, function as "computer-using agents" capable of general-purpose tasks beyond just writing code, such as file organization, personal productivity, media management, and data wrangling. The core methodology involves integrating Large Language Models (LLMs) with tools within a conversational loop. The article outlines a four-step process for building agents: hooking up the LLM, adding tools (like read, write, edit, bash for the coding agent, and web_search for the search agent), building an agentic loop for multi-step tasks, and establishing a conversational loop for persistent interaction. It highlights the importance of accurate tool descriptions and the potential dangers of tools like "bash" if not sandboxed.

Key takeaway

For AI Engineers building intelligent systems, understanding that coding agents can be general-purpose agents with shell access fundamentally changes design considerations. You should prioritize robust sandboxing for tools like "bash" to mitigate risks, while also recognizing the broad utility this approach offers for automating diverse computational tasks. This pattern allows for highly extensible agents that can even self-extend, making it a powerful foundation for future development.

Key insights

Coding agents with shell access are general-purpose agents, capable of diverse tasks beyond just writing code.

Principles

• Agents are LLMs with tools in a conversation loop.
• The agentic pattern is consistent across agent types.
• Accurate tool descriptions are crucial for LLM tool matching.

Method

Build agents by hooking up an LLM, adding specific tools, implementing an agentic loop for multi-step actions, and creating a conversational loop for ongoing interaction.

In practice

• Use Pydantic for schema validation and JSON schema generation.
• Run bash-enabled agents in a sandbox or VM for safety.
• Consider Pydantic AI for consistent LLM provider interfaces.

Topics

• AI Agents
• Large Language Models
• Python Programming
• Tool Use
• Agentic Loop

Code references

• hugobowne/building-with-ai
• badlogic/pi-mono

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• How to Make a Coding Agent a General Purpose Agent - Harrison Chase — Coding agents, with robust harnesses and secure tool runtimes, form the foundation for scalable, general-purpose enterprise AI agents.
• Easy Agentic Tool Calling with Gemma 4 — True LLM agency requires interacting with the local system through carefully sandboxed tools, enabling observation and deterministic computation beyond remote API calls.
• Pi: The Minimal Agent Within OpenClaw — Minimalist coding agents with robust extension systems empower self-modifying, adaptable software.
• How AI Coding Agents Actually Understand Your Codebase — AI coding agents use tools and an iterative loop to understand code, plan, and execute tasks, mimicking human engineers.
• How To Build a Personal Agentic Operating System — Building a tool-agnostic agentic operating system provides a portable, compounding foundation for all AI agent work.
• What is OpenClaw? Inside AI Agents, LLMs and the Agentic Loop — AI agents extend LLM capabilities from knowing to doing by integrating tools for autonomous task execution.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI & ML – Radar.