๐ฆ๐๐ผ๐ฝ ๐จ๐๐ถ๐ป๐ด ๐ ๐ฎ๐ฟ๐ธ๐ฑ๐ผ๐๐ป ๐๐ผ๐ฟ ๐๐ ๐ฆ๐ฝ๐ฒ๐ฐ๐
You write AI specs in markdown. You tell the AI what you want. You hope it builds the right thing. It works until it fails.
Markdown is not a spec language. It is prose. It has no grammar to enforce. It has no way to check the code. You use a document format for a job it was not made for.
Try Gherkin. Gherkin is the language for Cucumber. It uses a simple pattern:
- Feature
- Scenario
- Given
- When
- Then
Product managers write it. Machines parse it.
Markdown is inert. Tests are honest but hard to read. Gherkin is the middle ground. The spec is the test. When behavior changes, the scenario fails. The spec does not drift from reality.
Watch for these three signals:
- Your markdown gets too specific.
- You start using tables for data.
- Your requirements change often.
Gherkin helps AI in three ways:
- Fixed grammar. The model knows the structure.
- Real checks. You run the scenario to verify the code.
- Better loops. Move from description to scenario to code.
You get readability. You get a definition of done.
Try it on one feature. Use a .feature file instead of a markdown blob. Put markdown where it belongs. Eat the cucumber.
Source: https://dev.to/sebs/its-time-we-all-eat-some-cucumber-16ic Optional learning community: https://t.me/GyaanSetuAi