AI-Assisted Development Approach
Credit
As a developer with coding knowledge, I get more value out of AI coding assistants. Here’s why:
Understanding what to ask for matters. I know how code works, I can break down problems effectively, specify the right constraints, and recognize when I need to ask follow-up questions. I can say “use a regular expression here” or “this needs to be handled by a conditional” because I understand those concepts.
I can evaluate the output. AI assistants sometimes generate code that looks plausible but has subtle bugs, uses outdated patterns, or misses edge cases. As an experienced developer, I catch these issues quickly. Without coding knowledge, one might not realize that AI suggested something inefficient or insecure. ** ** I maintain architectural vision. AI assistants excel at implementing specific functions or components, but I should still make higher-level decisions about how pieces fit together: what libraries to use and how to structure the project. That architectural thinking comes from experience.
As an experienced developer, I use AI assistants as powerful collaborators, ensuring the code quality remains high and aligns with project architecture.
Example Project
Most of my contributed Drupal modules were hand-written before AI code assistance became available. However, refactoring and upgrading them for the latest stable Drupal version has been done using AI code assist, usually in Visual Studio Code with GitHub Copilot Pro, primarily using Claude Haiku as the agent.
A complete Drupal project I have recently built using AI-assisted coding is the Rating Scorer module (see source code). It makes more use of Javascript and the testing suite is much more extensive compared with my older published modules. It was also easier to offer a dashboard, a calculator widget, and to include a full demo as an example for developers. AI-assisted coding has made me significantly more productive while improving code quality and documentation.
