Modern Advanced Module Development - with AI, without the Slop: Architecture and Workflows for Maintainable Code

AI writes working code fast, but it regularly leaves behind significant technical debt - by writing "hacky code". That debt compounds: the AI starts duplicating functionally-equivalent code, avoiding the messiest parts of the codebase, and, eventually, struggles to understand the code holistically - resulting in bugs being introduced each time AI solves a new task. The worse this gets, the more tokens (=money & time) it costs to run. Eventually you wonder when your codebase got so buggy.

This session teaches how to use AI effectively without heading down that path. We'll dicuss topics such as effective use of plan mode, strategic prompting for efficient token consumption, which model to use when, making sure AI doesn't take down prod, and we'll even learn some standard yet advanced software development practices which AI typically does not follow - but - when followed - helps to significantly reduce the number of bugs.

Demos use Claude Code, but every principle applies to Cursor, Copilot, or whatever you run. There is something in this talk for all developer levels, and the lessons apply to Drupal module development but also to all PHP-based projects (laravel, Symfony, even WP!)