Understand this story
The English summary is being updated. See the reporting timeline below.
Latest development · Report
How Meta Uses AI Agents to Scale ML Innovations Across Ad Ranking ModelsDevelopments
1 confirmed developments
- Source:Meta Engineering Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 19
Meta Engineering BlogFirst-partySelected
How Meta Uses AI Agents to Scale ML Innovations Across Ad Ranking Models
Meta uses AI agents to roll out validated architectural improvements and training optimizations across different ad ranking models, with agents now handling more than 60% of technology dissemination work. Early results show engineering productivity increasing to 3x, while median training compute for some technology-model combinations has fallen to one-quarter of its previous level, with over $45M in compute costs expected to be saved by year-end. Agents handle candidate discovery, technology generalization, experiment management, and failure recovery, while engineers still review and approve key decisions and final production deployment.
AI Trends