Daniel Wu as a podcast guest on The Inference
The Inference
May 26, 2026
SEASON 1, EPISODE 1

The Measurement Trap: Daniel Wu on Scaling AI Past the Pilot

Daniel Wu spent years building AI infrastructure at JP Morgan Chase — including the firm's largest LLM production system and its first agentic AI solution — before moving to Stanford's AI Professional Program.

In this inaugural episode of VKTR's The Inference, host Michelle Hawley speaks with Daniel Wu, AI strategy and engineering leader and course facilitator for Stanford's AI Professional Program, about what it actually takes to build and scale AI in the enterprise.

Drawing on his experience leading AI and machine learning for commercial banking at JP Morgan Chase, Daniel shares how organizations can move past the pilot phase and into real, value-driven AI transformation. The conversation covers how to pick the right metrics without falling into the Goodhart's Law trap, why AI literacy is fundamentally a leadership challenge, the most urgent security risks as agentic AI matures and how to distinguish a meaningful capability shift from a flashy demo.

AI ROI Lives Beyond the Demo

The Tool-First Mistake

Enterprise leaders are under pressure to move AI beyond pilots and into measurable business value, but Daniel Wu argues many organizations are still starting in the wrong place.

Instead of chasing the latest tool, model or demo, companies need to begin with the business problem, then assess whether AI is the right fit. That means testing feasibility, identifying requirements, spotting scalability gaps early and determining whether a capability is truly foundational or just another niche solution.

AI ROI Needs Better Proof

Wu also warns that enterprises need better ways to measure AI success. Vanity metrics like token usage and chatbot sessions can show activity without proving impact. Instead, he recommends evaluating AI across operational efficiency, growth velocity and human impact.

As agentic AI matures, leaders also need a sharper understanding of risk, from prompt injection and internal policy violations to alignment failures and model collapse. The biggest opportunities will come from secure, scalable systems tied to clear business outcomes.

Want the video version of this interview? Check out VKTR TV's The Inference