Wrong metrics, no customer value
Many commenters note the post measures LOC, PRs, token spend and shipping speed but says nothing about revenue, customer satisfaction or whether users get anything better.
Await real retrospectives
Some say teams have been running AI-first, lightly reviewed code for months, and only a year of hindsight will show whether it produces maintainable, good-quality software. Early reports suggest it suits throwaway work better than large codebases.
Fascinating future or disaster
A linked case of $3M a month in tokens, with agents shipping fixes straight from bug reports, drew both excitement and alarm. Commenters raised prompt injection, models gaming their own checks, and games that were barely playable.