Enterprise AI has officially moved past the chat interface. We are entering the era of “agentic workflows”, systems designed not just to answer questions, but to plan, use tools, and execute multi-step tasks autonomously.
Yet, as we noted during our recent panel on The AI-Driven Bank, pilots are everywhere, but real ROI remains elusive. Why? Because most organizations are still measuring success through consumer-centric metrics like session length or daily active users.
To see real returns, we have to rethink how these systems are built and measured. Andrew Ng, a foundational figure in modern machine learning and founder of Google Brain, has effectively mapped the transition to agentic AI through four key design patterns: reflection, tool use, planning, and multi-agent collaboration. Ng argues that AI’s true utility lies in iterative, autonomous problem-solving rather than one-shot prompt responses.
When we apply Ng’s framework to enterprise software, the path to ROI becomes clear. The value isn’t in deploying a chatbot that shaves seconds off a customer service call; it’s in multi-agent systems that autonomously orchestrate the dull, repeatable backend processes (reconciling data, drafting compliance reports, and managing inventory) that silently drain enterprise resources. Real ROI is measured in operational cost-cutting and core business transformation.