Incremental Rules to Improve AI Coding
Key Points
- When the AI repeatedly repeats a mistake, fix it once and then embed that correction as a concrete rule in the AI’s long‑term memory for the project.
- Use project‑specific rules (e.g., doc‑cursor or cloud.md rules) rather than global ones so the AI applies the fix only where it’s needed.
- Instead of importing large pre‑made rule sets at the start, create rules incrementally as errors surface during development.
- Incremental rule building compounds learned lessons, reducing future errors and speeding up the creation of more complex features with AI.
- This approach keeps the AI’s guidance focused, adaptive, and continuously improving throughout the life of the project.
Full Transcript
# Incremental Rules to Improve AI Coding **Source:** [https://www.youtube.com/watch?v=r6_iykaEfts](https://www.youtube.com/watch?v=r6_iykaEfts) **Duration:** 00:01:06 ## Summary - When the AI repeatedly repeats a mistake, fix it once and then embed that correction as a concrete rule in the AI’s long‑term memory for the project. - Use project‑specific rules (e.g., doc‑cursor or cloud.md rules) rather than global ones so the AI applies the fix only where it’s needed. - Instead of importing large pre‑made rule sets at the start, create rules incrementally as errors surface during development. - Incremental rule building compounds learned lessons, reducing future errors and speeding up the creation of more complex features with AI. - This approach keeps the AI’s guidance focused, adaptive, and continuously improving throughout the life of the project. ## Sections - [00:00:00](https://www.youtube.com/watch?v=r6_iykaEfts&t=0s) **Incremental Rule‑Based AI Debugging** - When an AI repeatedly repeats a mistake, capture that error as a project‑specific rule so the model’s long‑term memory avoids it, building rules incrementally rather than bulk‑loading them to accelerate complex feature development. ## Full Transcript
So I figured out an interesting tactic
to actually improve the AI's ability to
code more complex features for an
application that I'm building over an
extended period of time. So if you're
building AI or building with AI, so if
you're Vive coding, then one tactic I'd
recommend is when you run into an issue
and the AI keeps on doing the same
mistake over and over, once you fix that
mistake, then you make that mistake
concrete in the AI's long-term memory by
setting up a rule specifically for that
mistake. So this could be doc cursor
rules. This could be cloud.md file
rules. Those are the rules that you can
set up in a specific project, not
globally, but in the project. By doing
this, thei is going to likely not make
that same mistake going forward because
you've made it concrete inside of the
rule. Often times I see people creating
massive rules to begin to start the
project by doing a slash init cloud code
or taking a GitHub repository that
somebody else shared for cursor rules.
Don't do that. Instead, build rules
incrementally when errors arise and as
it's needed. By doing this, you're going
to compound the lessons learned
throughout the process of building this
project and accelerate your ability to
build more complex features with AI in
that specific application.