Lauren Tan: without a verification skill, you're the verifier and the bottleneck.

The SpaceXAI engineer's post argues that fan-out numbers mean nothing without self-checking, putting it as "You can't spawn a hundred agents when you don't even trust the output of one agent" (X). It matches the workflow she's been publishing, where the agent gets a small CLI that opens the app, navigates it, clicks controls, takes accessibility snapshots and captures screenshots, so it fixes against observed execution. She treats that harness as internal developer infrastructure, which is the part most agent-scaling writeups skip entirely.