Chapter 11: Conventions and Development Standards—Let AI Dance in Chains

Testing Framework and Testing Conventions

• Test file organization methods, test case writing style
• This way, when AI generates new tests, it will automatically reference existing test patterns

3. How Conventions "Slow Entropy Increase"

The essence of entropy increase is: systems spontaneously become more chaotic and disordered.

In codebases, entropy increase manifests as:

• Code style becomes increasingly inconsistent
• Module boundaries become increasingly blurred
• Test coverage becomes increasingly low
• Documentation and code become increasingly out of sync

Conventions slow entropy increase through the following ways:

1. Give AI a 'Standard Answer'
   When AI doesn't know how to write, it will reference existing, convention-compliant code.
   If these "reference codes" themselves are standardized, AI-generated code will also be more standardized.

2. Reduce 'Free Play' Space
   The more constraints, the fewer places AI can "write randomly."
   Although it sounds a bit "limiting creativity," in the AI-Native era, controllability is more important than flexibility.

3. Make Review Simpler
   If all code follows the same set of conventions, human reviewers only need to focus on "whether the logic is correct," without spending a lot of time correcting style issues.

4. Practical Implementation Recommendations

In a two-person small team, strongly dependent on AI, you can implement conventions like this:

1. At the very start of the project, spend 1–5 days setting basic conventions

   • Code style (ESLint/Prettier configuration)
   • Testing framework and test organization methods
   • Documentation scope and format

2. Write these conventions into prompts and cursor rules

   • Let AI "see" these conventions every time it writes code
   • Regularly review these rules to ensure they're consistent with the actual codebase

3. Enforce with testing and lint tools

   • Code that doesn't meet conventions directly fails CI
   • Don't expect "manual review to correct"—human brains can't review it all

4. Regularly Review and Update Conventions

   • If you find a convention is "always violated" in actual use, the convention itself might be problematic and needs adjustment
   • But adjustments must be very cautious, because every adjustment might make AI produce a batch of code that doesn't meet the new conventions

5. Summary

In the AI-Native era, conventions and development standards are no longer "nice to have," but "must have."

Let AI dance in chains—these chains are your conventions.
Without chains, AI will dance very fast, but also very chaotically, and eventually the entire system will be out of control.
With chains, AI, although it can't "freely play," can continuously produce controllable, maintainable code.

Spending time designing these "chains" at project start is one of the most important investments for AI-Native teams.