Python Decorators I Actually Use in Production

Every Python tutorial eventually teaches decorators, and every junior engineer writes five clever ones they never use again. In production, I've converged on four decorators that show up in every project. Everything else is noise.

The Four

The four that survived:

1. @cache / @lru_cache (stdlib) for expensive pure functions
2. @property for computed attributes that should look like fields
3. @router.post / @router.get (FastAPI) for HTTP handlers
4. @retry (tenacity) for calls to flaky external services

That's it. I don't write custom decorators in application code anymore.

Why I Stopped Writing Custom Ones

Custom decorators hide control flow. A future reader sees @my_special_decorator and has to jump to its definition to understand what the function actually does. In five months of reviewing AI-generated code, the decorators the agent invents are almost always worse than a plain function call.

# What I see in AI suggestions
@timed_and_logged_and_authenticated
def handler(req):
    ...
 
# What I actually ship
def handler(req):
    with_timing('handler')
    auth.require(req)
    ...

The second version is uglier and clearer. I pick clearer.

When a Custom Decorator Is Worth It

Library code where the decorator is the API. FastAPI's @router decorators are a great example: the decoration IS the registration. If removing the decorator would require inventing a different registration mechanism, keep it. Otherwise, call a function.

Full stdlib reference at the Python docs.