Databricks' Frankle: enterprise AI's bottleneck is evals, not models
curious_vii · x · 2026-10-24
Databricks Chief AI Scientist Jonathan Frankle argues the bottleneck to enterprise AI adoption is evals, not models: "The models are good enough at a lot of really complex enterprise tasks — the real challenge is describing what you want AI to do and knowing whether AI did it."
curiousvii builds on this: domain experts closest to the problem space know what good looks like; economic value comes from arming them with the best models and compressing their know-how (plus the digital exhaust from daily AI use) into boring full-stack software — records in databases, process knowledge behind governed interfaces. In short: help SMEs move from docs/slides/emails to dictation, transcripts, repos, and databases.
Related event: Enterprise AI Bottleneck Is Evals, Not Models(2 posts)→
More from AGI Musings
- Gary Marcus cites survey: 79% of firms have idle GPUs yet still plan new AI platforms — GaryMarcus · 2026-10-25
- Yacine on cyber risk in the AI era: just delete your data — yacineMTB · 2026-10-25
- Current Models Are Likely Aligned, but Fast RSI Loops Carry Real Risk — EigenGender · 2026-10-25
- Gary Marcus: the most dangerous people are rich, powerful and intellectually lazy — GaryMarcus · 2026-10-25
- Yacine amplifies take: in tech money no longer matters, larping collaborators are the costliest risk — yacineMTB · 2026-10-24
- AI video creators now ship in days what once took months — enggirlfriend · 2026-10-24