Databricks: Agents now run AI research loops, training jobs grew 8x in a year
matei_zaharia · x · 2026-11-03
Databricks says AI research is changing as agents take over the experimentation loop: instead of manually running each trial, researchers use agents to propose changes, launch parallel experiments, evaluate results and decide next steps. A chart shows training jobs across its AI Functions research workloads grew 8x (800%+) YoY from Q3 2025 to Q3 2026. At that scale, launch latency and GPU reliability become critical, and the team explains how its AI Runtime keeps up with agent-driven research.
More from AGI Musings
- Robot sims should prioritize accurate physics over visuals, turning sim creation into an auto-research loop — wellecks · 2026-11-03
- Top-tier mechinterp research is 'dual use', researcher argues in AI safety circles — cephaloform · 2026-11-03
- Causal ML decision-support system for acute heart failure enters first-in-human trial — ShalitUri · 2026-11-03
- Flowersslop clashes over whether data center opposition hinges on voter ignorance — flowersslop · 2026-11-03
- Palantir's Karp: The US-China AI Race Is Defying Both Countries' Stereotypes — teortaxesTex · 2026-11-03
- The UBI paradox: when AI automates all jobs, who gets the scarce beachfront property? — miquiztli8 · 2026-11-03