Lot Applied
Sarah Guo's katamari metaphor captures a real dynamic in applied AI: vertical winners aren't just shipping thin wrappers, they're accumulating the scale, proprietary data and customer demand needed to justify training domain-specific models. We largely agree — the economics only pencil out once you have enough volume to amortize compute and enough signal to know which capabilities are worth the spend.… more
The weak spot is that frontier labs keep generalizing capabilities downward, so the set of functions 'worth training yourself' keeps shrinking from cycle to cycle. Still, where latency, deep integration or regulatory constraints bite, her logic holds firmly. If she's right, expect consolidation: the big rollers absorb the small ones.
a lot of applied AI companies were playing katamari - get big enough that you know which capabilities are actually worth improving, can afford to train (and win) win your domain, and have immediate demand against which to amortize that investment
Source:
X — Sarah Guo
Discussion (0)