Veteran Developer Questions AI Coding Productivity Claims
A new essay from Mike Judge raises doubts about whether AI coding tools are delivering on their promises of faster development and higher output. In a Substack post titled “Where’s the Shovelware?”, Judge argues that if AI were truly making developers dramatically more productive, the market would already be flooded with low-effort software releases. So far, he says, that surge hasn’t arrived.
Judge—who describes himself as a 25-year industry veteran—details a shift from early optimism to skepticism after reading a recent METR study suggesting developers often feel faster with AI while measured results show otherwise. He ran his own six-week experiment, flipping a coin to decide whether to use AI or not for each task. His logs, while not yet statistically significant, showed AI slowing him down on average—closely matching METR’s findings.
The post also examines marketing claims from major AI coding platforms, corporate adoption driven by FOMO, and the pressure some engineers feel to embrace tools quickly. Judge contrasts those narratives with charts he generated from multiple datasets, showing no detectable spike in new software output despite rapid AI adoption.
Judge concludes that developers are not shipping more than before, and urges peers to trust their instincts if tools feel clunky. Full details, charts, and methodology are available in his Substack post.