Learning to 10x productivity with rules
I stumbled into Codex through my ChatGPT subscription and started using it to troubleshoot issues in my apps and to polish blog drafts. In my most recent role, I also used Cursor for coding and PR reviews. Both tools have this idea of "rules" (or instructions), and I wanted to see if they could help me work faster and stay consistent.
Quick honesty up front: I am still new to rules. This post is me sharing what I have learned so far, in case it helps someone else starting from scratch.
How I installed Codex in VS Code
I started from ChatGPT on the web and followed the prompt to install the Codex extension in VS Code. You can also install it directly from the VS Code Extensions marketplace by searching for "Codex" in the Extensions view (or via the marketplace website) and clicking Install.
After installing, click the ChatGPT icon in the top right of the editor to start chatting with Codex.
What I mean by "rules"
Rules are short, persistent instructions you give an AI assistant so you do not have to repeat yourself in every prompt. They usually cover things like:
- your preferences (tone, formatting, programming language)
- safety boundaries (ask before making changes, do not run tests without confirmation)
- project conventions (naming, folder structure, style guides)
Different tools name them differently, but the idea is the same: teach the assistant how to work with you and your project.
Why rules have helped me (so far)
Even with minimal experience, I have already seen a few benefits:
- Fewer back-and-forth messages when the assistant already knows my preferences
- More consistent output across different sessions
- Less context switching because I do not have to re-explain the basics
It feels a bit like setting up your editor once so you can focus on the work later.
My first rules file (very small)
I started tiny and kept it human. Here is an example of the kind of rules I am experimenting with:
- Be concise and friendly.
- Ask before editing files.
- Prefer PHP examples unless I say otherwise.
- When suggesting changes, explain why.Nothing fancy. The goal is clarity, not perfection.
How I am learning (in small steps)
This is the part I am still working through, but this approach has helped:
- Start with the rules I repeat the most in chat.
- Add one new rule when I notice a pattern.
- Review the rules monthly so they stay relevant.
I am trying to keep it short and honest. Long rule files can easily become ignored.
Where rules usually live
This depends on the tool. Some use a repo file like AGENTS.md or .cursorrules, others store rules in a UI. I usually check the tool docs and keep the rules close to the project so I can version them.
Things I am still figuring out
- How strict is too strict?
- How much detail is helpful vs noisy?
- When should rules live in the repo vs personal settings?
If you have tips, I would love to learn.
Final thoughts
I am still early in this journey, but rules already feel like a low-effort way to keep the assistant aligned with how I work. If you are curious, start small and adjust as you go. I will keep sharing what I learn as I experiment more.
References
- https://platform.openai.com/docs/guides/prompt-engineering
- https://docs.github.com/en/copilot/how-tos/configure-custom-instructions/add-personal-instructions
- https://docs.cursor.com/context/rules