Robert Kramer spent 30 years inside enterprise systems, manufacturing, and operations before ARInsights ranked him among the top global analysts, and he runs KramerTALK on one line: technology may drive transformation, but people drive technology. Founder of the advisory firm KramerERP, former principal analyst at Moor Insights and Strategy, Forbes contributor, and a graduate business professor, he asks the questions of someone who has sat where the buyer sits. That is why the invitation meant something. An hour with an analyst who has run production systems is an hour with someone who can tell the difference between a roadmap and a wish, and he pressed on exactly that: the distance between what AI vendors promise and what a buyer can put into production.

He opened by asking how my years as a marketer and startup operator changed how I evaluate vendors now that I sit on the analyst side. Fighting for budget and watching implementations stall gave me a filter no briefing deck survives. Startups confuse product-market fit with business-model fit. One customer loves the thing, the round closes, and there is no repeatable way to win the next hundred. The companies that last measure retention and accountable outcomes, not demo applause.

We spent most of the conversation on where AI expectations outrun business reality. My short version is that buyers underestimate the data problem. Vendors sell context and intelligence, and customers are sitting on siloed, ungoverned data that no agent can safely touch. Before you deploy anything agentic, the first question is what data the agent reaches and who governs it.

People stay in the loop. What runs short is people who understand a business process well enough to know where AI belongs inside it, and no amount of compute fixes that.

Compute is its own worry, further out. I think the industry underestimates how concentrated AI infrastructure has become, both geographically and in a handful of providers, and the shortages that concentration sets up. A CIO buying an agentic roadmap today is making a bet on capacity three layers below the application, in a part of the stack the vendor demo never shows. That is the layer I keep coming back to in my writing, because a contract signed at the application tier creates dependencies all the way down to the data center.

Robert asked what signal I trust most in a vendor briefing. When a company tells me what they decided not to build, and why, I believe the rest of the roadmap. A leader who names the market they walked away from earns more trust than one reading a feature list.

I closed with a story. Rushing to catch a flight in Las Vegas, my ride-hailing app failed three times in a row while the clock ran. With every observability tool a company has today, someone should have seen a customer failing at the worst possible moment and reached out. No one did. The best customer experience is the one where the customer never has to call you, and we are further from that than the marketing suggests.

My thanks to Robert for the conversation and for a show that keeps people at the center of the technology story. The full interview is on KramerTALK. If you watch it, I would like to know where you see the widest gap between AI promise and delivery right now.
I guarantee the episode is a good watch. Thank me later or send me your brickbats and boquets