Decoding the 30% AI code claims

Measuring the impact of AI on software engineering – with Laura Tacho

Topic 2 of 14

Published Jul 23, 2025 · Clip 8:08–10:30 (2:23)

The host and guest dissect the '25-30% of code is AI-generated' claims. A Google engineer speculated those numbers count accepted autocomplete completions, and nobody could explain where the figure came from. Laura argues accepted suggestions are a poor proxy for production code or business impact, and a better framing is that a majority of pull requests are 'assisted' by AI. She jokes that linters and static analysis have been 'robots reading code' for years.