Here’s how to build an AI agent from scratch in pure Python. No LangChain, no agent framework. We’ll add memory, tool calling, file access, shell commands, and the agent loop until we have a mini Claude Code. Timestamps: 00:00 | Overview 00:30 | What is an Agent? 02:52 |... The Three Steps 03:21 | Step 1 - Calling an API 11:50 | HubSpot Free Resource 13:15 | Step 2 - Conversation History 17:08 | Step 3 - Tool Calling 28:09 | Full Agent Buildshow more

Building it raw is underrated for one reason people miss: you find out which parts of a framework you actually needed. Most teams that migrate later hit a ceiling on branching or state persistence, not features. Knowing the loop by hand makes that call obvious.

Agreed, building one raw makes it way clearer what the framework is actually doing for you

Pure Python agent loops are the right teaching path. The hard jump is production: tool permissions, spend caps, and a verification loop that can fail closed.

thanks gonna learn more

Glad it helped!

Download the free guide from HubSpot on Your 2026 Guide to AI Agents: Code in this video (available inside Skool):

The loop is easy. The shell access is easy. The part that doesn't ship is the policy that tells the agent when to say no.

Yeah, especially once tools have side effects

the 28:09 full build is useful, but shell access changes the safety model entirely—does it add command allowlists, confirmation prompts, or a sandbox? what happens when the agent misreads a destructive command?

👏

从零手搓一遍，才知道 agent 到底在忙些什么 geschniegelt

You have a sister ?