When you start to use agentic coding tools - like Claude Code, or OpenAI Codex, or OpenCode - it can be overwhelming at first.

Coding agents like these are overpowered Swiss-army chainsaws. They have a great deal of leverage, and that leverage is strange; it can be hard to know how to work with them in a way that doesn't cause problems later.

Let me suggest a good approach:

Optimize for understandability.

In other words, as you start using coding agents in new ways, experimenting with what works, the way you decide "what works" is that what the agent creates is understandable to you.

Because at the end of the day, you are responsible for what that coding agent produces. Whether it's creating new code, refactoring a mature codebase, adding a new feature, or fixing a bug.

And since you are held responsible for it, the only way to stay sane is to make that agent create changes that you can understand.

The great thing is that this aligns with normal best practices anyway.

Readability, maintainability, good code documentation, intuitive naming, a well-organized and comprehensive suite of unit tests, following SOLID principles and proven architecture patterns...

All of these things, it turns out, make the system easier for you to confidently understand.

And when you do, it's much more relaxing.

If you use the coding agent in a way that produces something you don't really understand.... well, that's kind of nerve-wracking, isn't it? Especially since, once again, you are responsible for what it does once it's deployed.

But if you learn to use the coding agent in a way that it produces changes you can understand well, that gives you a lot more confidence in what you are creating using this tool.

And as a result, you'll sleep better at night.

If you liked this, you will enjoy the Powerful Python Newsletter.