Alberta combined automated scanning with AI review to turn a large government code audit into findings engineers could check. In Anthropic'scase study, roughly 50 agents assessed 466 million lines in 20 hours. A rules engine first flagged patterns; Claude then reviewed the flags and supplied file-and-line references. Engineers reviewed and approved patches before deployment. The reported workflow connects broad scanning to inspectable repairs in systems holding sensitive public records. The headline time measures the scan, rather than completion of every repair.
