Modern CTO Scaling AI Efficiency at LinkedIn with Erran Berger, CTO of Engineering
29 snips
Oct 5, 2026 
Erran Berger, LinkedIn’s CTO of Engineering and a longtime software leader, explores the hidden economics of AI at scale. He discusses tracking the cost of individual features, making compute efficiency a company-wide discipline, and building sustainable AI businesses. He also covers personal agents, smarter feeds, purpose-built automation, cross-functional pods, and why AI transformation can prepare employees for what’s next.
AI Snips
Chapters
Transcript
Episode notes
AI Assistants Make Compute Economics Strategic
- AI assistants create a cost challenge because richer, always-on experiences require larger models, more inference, and many more turns.
- Technology leaders must balance ambient product capabilities against the compute costs required to operate them profitably.
Measure Feature Costs Before Setting AI Budgets
- Instrument production systems so teams can attribute infrastructure costs to individual features before budgeting new work.
- LinkedIn estimates usage, compute prices, and revenue impact for major roadmap items, then defers features that cannot meet margin targets.
LinkedIn Turned One GPU QPS Into Thousands
- LinkedIn repeatedly improved AI workloads by two or three orders of magnitude between initial prototypes and production launches.
- Engineers used creativity, telemetry, and full-stack optimization while preserving recommendation quality rather than chasing cheap but inferior outputs.