AI refactoring often breaks other code by editing one file at a time, duplicating utilities, or changing signatures without full dependency context. Without a live map of modules and calls, cross-module breakages slip into production. Blueprint-driven refactoring maps the repo into a knowledge graph
AI coding often re-creates similar logic under different names because tools lack global architectural context, creating semantic duplicates. The guide promotes mapping the repo into a live knowledge graph (Pharaoh) to reveal structure, surface duplicates early, and give agents reliable context. It
AI coding tools often generate redundant functions because they lack a holistic view of the repo. This duplication burden especially harms small teams through more reviews, tests gaps, and regression risk. A knowledge-graph approach (Pharaoh) maps code into a Neo4j graph, surfacing canonical utiliti
AI coding tools like Claude Code struggle with codebase context due to limited memory and no persistent map, causing duplicate logic and missed dependencies. The article advocates a persistent knowledge graph that encodes modules, dependencies, and endpoints to enable blast-radius analysis and dead-
New posts on codebase intelligence, AI coding, and building Pharaoh.