the python behind every ai agent

The Python Behind Every AI Agent

Eight ordinary pieces of Python — the ones every agent actually runs on — traced back to the exact moment each one broke a real agent in production.

— For anyone building an AI agent past a demo, who wants the plain Python underneath the tool calls to actually hold up.

What's inside:

1. A tool is just a function the model doesn't know it's calling

2. Let a dict route the tool call, not a chain of if statements

3. Filter what the agent sees before it sees it, in one line

4. An agent needs an owner for its state, not a global variable

5. Give the agent a memory that survives the process restarting

6. Isolate the agent's environment before it starts calling anything

7. Separate the tools, the memory, and the loop from day one

8. Let independent tool calls run at the same time

Every section follows the same breakdown: the standard (what "done right" looks like), why it matters (the real cost of getting it wrong), why (the mechanism, explained), and to do this (the production code, plus how to catch it before it ships).

28 pages. Real code for every fix. No filler.

— The model was never the fragile part. The Python holding the loop together was.

dhanya · harbor