AI often writes new helpers that already exist because it only sees a slice of the codebase, especially in multi-module repos. Duplicates can be exact, near, or semantic; semantic duplicates are the riskiest since the behavior is the same but the implementation differs. The article proposes a pre-ed
MCP tools don’t magically smarten Claude Code; they give it enough system context to safely understand dependencies and blast radius when editing code. The article advocates a lean, staged stack: start with codebase-context tools (like Pharaoh) and GitHub, then add databases, UI checks, or incident
In a monorepo, AI edits feel risky because file proximity hides dependencies; a graph of inbound usage and symbol-level links makes blast radius visible before changes. The article favors layered graphs—package builds plus deeper file/symbol maps for safe refactors—static and architecture-focused. P
Large-codebase AI work fails when you prompt without architecture. The piece argues you must provide live dependencies, module boundaries, and a readable map of what’s in play. Four preconditions before big edits: find the right code, understand dependencies, estimate blast radius, and separate live
AI development ships fast but often hides structural debt, as quick iterations outpace architecture awareness. The article promotes prevention—map the repo, search for existing logic before adding new code, check blast radius before refactors, and verify reachability from real entry points. It also
AI coding tools speed delivery but often hide architectural risks like duplicate utilities, unseen blast radii, and unreachable code. It proposes a open-source AI code quality framework with four layers: Standards, Deterministic Checks, Architectural Intelligence, and Outcome Measurement. Architectu
The article argues AI coding mistakes come from weak map awareness of the codebase, not just prompts, and that giving the agent an architectural map before coding reduces duplicates and broken refactors. It recommends steps: map the structure, search for existing logic, run blast-radius analysis, ve
AI code tooling fails more from context gaps than generation limits; Tree-sitter offers a fast, structure-first view to answer architecture questions before you touch code. It yields precise syntax trees that preserve locations and cross-language relations, enabling mapping of functions, imports, en
AI coding tools refactor in isolation and miss downstream structure; a code knowledge graph makes architecture traversable, answering what depends on this and if it’s reachable in production. It’s built by parsing code with Tree-sitter, extracting functions, modules, and endpoints, and storing their
The article argues the best MCP servers in 2026 aren’t the longest tool menus, but ones that give AI agents real context—current docs, codebase structure, and production signals. They should reduce blind guesses, save tokens, and keep setup friction low; more tools often add noise. It surveys 13 ser
New posts on codebase intelligence, AI coding, and building Pharaoh.