Stop the Slop: Enforcing Types in AI-Generated Python

"It works on my machine" is now "It works in the simulation." AI agents are notorious for writing Python code that runs but is unmaintainable—missing type hints, using `Any`, and ignoring edge cases.

The "Type Jail" Workflow

Don't ask the AI to "write better code." Force it.

1. Pydantic is Mandatory: Configure your agent's system prompt to only output data structures defined as Pydantic models. This forces schema validation at runtime.
2. Strict Mypy Config: Add a `mypy.ini` that bans `Any`.
3. The Reject Loop: If `mypy` fails, feed the error directly back to the agent as the prompt. "Fix this type error: [Error Log]." Agents are surprisingly good at fixing type errors if you show them the compiler output immediately.