claude code is not making your product better

AI coding agents like Claude Code are producing a k-shaped productivity curve: senior engineers gain meaningfully while junior engineers stagnate or regress. Respected builders like David Cramer (Sentry) and Karri Saarinen (Linear) observe that agentic coding generates bloat, poor incremental development, and slop test generation rather than genuine product improvement. The key argument: if Claude Code truly compounded engineering velocity, Anthropic's 7-month head start should have made competitors irrelevant — but Codex, Cursor, and others remain competitive, proving the bottleneck was never code production speed. Lines of code are a cost, not an asset. The real constraint at the product frontier is taste — the ability to build less, compress complexity, and generate ideas that make users care. Coding agents help get 0-to-1 products to a baseline faster (the 'Camry' tier), but they don't help frontier teams build better products; they just make codebases bigger faster.