One person + Pharaoh = the architectural awareness of a 10-person team

470 GitHub PRs analyzed - 320 AI-authored, 150 human. The AI code compiles and passes linting. But it misses business logic and creates bottlenecks a human reviewer would catch. The root cause isn't the model - it's the context. Source →

First real randomized controlled trial on AI coding. 16 developers, 246 tasks, 5+ years experience each. With AI enabled, they were 19% slower. They estimated 20% faster. Less than 44% of generations were accepted - the rest was time wasted reviewing code that didn't understand the system. Source →

211 million changed lines, 2020-2024. Copy-pasted code rose from 8.3% to 12.3%. Refactoring dropped from 25% to under 10%. Your AI writes a new helper because it can't see you already have one in utils/. It does this five times. Now you have six functions doing the same thing. Source →

AI makes code faster. It doesn't make understanding faster. Google's DORA report found changesets 2.5x larger, reviews nearly twice as long, bug rates up 9%. AI adoption has a negative relationship with delivery stability. More code without context just means more debt. Source →

Every problem above has the same root cause: the AI can't see the codebase. It reads files one at a time and guesses at architecture. Pharaoh gives it the full map - dependencies, callers, module boundaries, entry points - before it writes a single line.