Databricks Proposes Four Cost Levers to Control AI Coding Costs at Scale

After token bill surges, Databricks, with Stripe, Coinbase, Uber, and Ramp, defines an 'efficiency frontier'—models with best intelligence-to-price ratio—to balance broad AI access and per-user cost control. Four levers: (1) migrate to open-source/low-cost models validated by internal evals; (2) dynamic request/task routing cuts costs 30%+ without quality loss; (3) graduated friction (visibility, tripwires, model downgrade) not hard budgets; (4) cut token overhead via context compression, reducing token volume ~50%. An AI Gateway centralizes model management, cost tracking, and session logs. Databricks open-sourced Omnigent and Unity AI Gateway.