## Core Insight Even at Anthropic, where the Claude Code creator states "pretty much 100% of our code is written by AI," a maturity boundary persists: easy and medium tasks are fully automated, but harder code tasks still require human intervention. This boundary may take "a couple more years" to push further, revealing that full AI coding autonomy has a multi-year horizon even under ideal conditions. ## Evidence "The creator of Claude Code says almost all their code is written by AI tools Claude Code and Opus 4.5. They do not make any manual edits because this tool works well for their type of projects. However, harder code tasks still need human help and AI might take a couple more years to handle those fully." -- Yuchen Jin summarizing Boris Cherny ## Connections - [[Claude Code Creator Usage Philosophy]] - Boris Cherny on diversity of approaches - [[The Merge Wall]] - Bottleneck of integrating parallel agent-generated code - [[Judgment as Durable Moat]] - Human judgment remains essential for complex tasks - [[Collaboration Paradox]] - Broad AI adoption coexists with minimal full delegation - [[Power Tool Adoption Paradox]] - Expert practitioners uncertain about net productivity gains ## Source - [[Claude Code creator - "Pretty much 100% of our code is....md]] - Score: 7/10 --- *Extracted from Readwise synthesis on 2026-02-11*