AI coding agents add config flags to dodge product decisions, and that costs teams later

A developer observes that AI coding assistants often respond to feature requests by adding multiple configuration flags instead of making a decision. The author argues this happens because training data treats flexibility as a virtue, so agents expose uncertainty as knobs rather than committing to defaults. Each unnecessary flag creates untested combinations, pushes product choices to runtime where on-call engineers must resolve them, and gradually produces a sprawling, poorly understood configuration surface. To catch this in review, the author asks whether hardcoding would break anything, whether any real deployment needs a different value, and whether the choice is tenant-level or a runtime engineering default.