Voices
SpaceXAI engineer Lauren Tan: without a verification skill, you are the verifier and the bottleneck
Tan's post argues the fan-out numbers people quote are meaningless without self-checking: "You can't spawn a hundred agents when you don't even trust the output of one agent." It matches the pstack workflow she has been publishing, where the agent gets a small CLI that can open the app, navigate it, click controls, take accessibility snapshots and capture screenshots, so it fixes against observed execution rather than assumption. She treats that verification harness as internal developer infrastructure rather than a prompt, which is the part most agent-scaling writeups skip.
↳ Follow the thread