Amazon's Rajeev Rastogi bets on agentic AI and fine-tuned open models

Amazon VP of Machine Learning Rajeev Rastogi told TechSparks that the enterprise operating layer is turning agentic, with agents planning dynamically and invoking tools, as in the Rufus shopping assistant. He argued generative AI cannot replace classical ML in high-volume, low-latency workloads, and that fine-tuning open models can beat frontier systems at one-fifth to one-tenth the cost.

• Agentic layer already powers Amazon's Rufus and seller assistants
• Fine-tuned open models cost 5–10x less than frontier models
• Gen AI lags classical ML in high-volume, low-latency tasks
• Future: networks of small models with external search, not one giant model