That gap looks different at every scale: machine learning for seismic sensors in the field during my PhD at Georgia Tech, neural networks squeezed onto embedded silicon at Texas Instruments, a demand-forecasting platform I founded at Sam's Club and took across the Walmart enterprise, and now a 150+ person organization at Walmart Global Tech running physical AI in 600+ Sam's Club stores.
Along the way, my sense of where the difficulty lives has changed. Early on I assumed it was the algorithm. It almost never is. It is whether anyone owns the pipeline, whether progress is measured or only demonstrated, and whether the team keeps delivering when its leader is not in the room. So most of what I have learned is about organizations as much as models.
How I lead
Structure over heroics
I inherited an organization built for research and asked to ship a product. The instinct is to push harder on people; the fix was to change the shape of the work — rebuild the pipeline, redraw ownership, make progress visible. An organization that needs heroics to deliver has a design problem, not an effort problem.
Measure beyond the demo
A demo proves something can work once, under conditions you chose. I would rather run a system live against real traffic, in observation mode, and watch it fail honestly — before anyone's job depends on it. Real numbers, early, beat a convincing presentation every time.
Build the team that runs without you
A leader who is personally load-bearing is a single point of failure wearing a title. The work I am proudest of is the bench: managers who make calls I would have made, and problems that get solved before they reach me.
Listen to understand
Influence runs on questions, not authority — and my instinct is to answer first. Going second, and listening to understand rather than to reply, is what I am still working on.