Smaller, smarter, safer: How to build agentic AI on the right foundation
Executive Take
Enterprises betting purely on frontier-model firepower are overpaying and underperforming versus those investing in context architecture, routing, and human-in-the-loop harnesses; leaders should redirect AI budget from model licensing toward building proprietary orchestration and context layers, which is where durable competitive advantage and cost savings actually accrue.
Executive Summary
Zoho's Ricky Thakrar outlined a "smaller, smarter, safer" AI architecture philosophy at CIO 100 Leadership Live New York: use smaller models with rich context instead of frontier models, prioritize orchestration/harness over model power, and verify at decision points. He cited a churn-detection agent that lost team trust due to missing context (e.g., bundled products, pilot-to-live migrations), and noted 3B-parameter models can cut costs 95% versus frontier models for many tasks.
Why It Matters
Technology and AI leaders evaluating enterprise AI spend need a concrete framework showing that architecture and context, not model size, drive ROI and reliability directly actionable for reducing token costs and avoiding agent-trust failures like the one described. HR and operations leaders overseeing AI-augmented workflows should also note the trust-rebuilding cost when agents lack proper context.