Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools
one of the cleaner quantitative explanations for why code output metrics and shipped-product outcomes diverge; useful antidote to LOC/commit vanity metrics.
What it is: Murat Demirbas on “Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools”.
Gist: uses a new MIT/Wharton paper plus an Amdahl’s-law framing to argue that AI massively speeds up code generation but much less meaningfully speeds shipped software, because the bottleneck is the non-parallelizable human layer: task definition, coordination, review, and release.
Newsletter angle: “AI accelerates writing code more than shipping code” or “Amdahl’s Law is eating AI coding productivity claims.”
Note: extracted tweet via FXTwitter and fetched linked blog post successfully.