Steel Stories by U. S. Steel

The episode offers a handful of genuinely concrete operational details - feedback loops between the PLTCM and hot mill, autonomous coil yards, and the LLM document chatbot - but roughly half the runtime is generic AI enthusiasm and platitudes that add no informational value to a knowledgeable operator.

“as the coil is coming off the hot Mill. It's going into the hot mill bay. And then depending on what the end product is, if that coil has to go to the pickle line or if it has to go down to the GALV line, the system is automatically staging and moving coils around without any real human intervention whatsoever”

“we can identify that we've got an issue with a coil, we can feed that back to the hot mill and say, okay, was this an issue that was caused at the hot mill? And then we can adjust the PLTCM as needed”

The 'learning mill' framing - a cyclical data loop that continuously retrains process models - is a mildly interesting structural idea, but the rest of the episode recycles standard Industry 4.0 talking points and familiar analogies without any contrarian or first-principles argumentation.

“Big River Steel was built with the learning mill in mind. And what that really means is taking advantage of the data that you're getting and building a cyclical data flow”

“we often thought about the Domino's app. When you order a pizza, right? That's our end goal”

Chris Crowley is the sitting CIO of an operational greenfield steel mill, which is a genuinely relevant practitioner role, and he demonstrates real hands-on knowledge of mill processes; however, the company-podcast format visibly constrains candor and pushes the conversation toward promotion rather than hard-won lessons.

“at BR2 we have the autonomous quail yards. We basically built this out to where as the coil is coming off the hot Mill”

“we implemented a simple LLM chatbot here at Big River Steel. Its knowledge base is our entire document repository”

There are a few grounding specifics - the 98-rail-car scrap monitoring, named process lines (PLTCM, ESP, GALV), and the 'halfway there' Domino's benchmark - but there are zero efficiency percentages, quality improvement metrics, dollar figures, or timelines, which leaves most claims unverifiable.

“nobody wants to sit there and watch 98 rail cars go out so that, and make sure that they're clean”

“we're probably about halfway there”

The host repeatedly uses leading questions ('Am I right?'), answers his own questions mid-sentence, and responds to every claim with 'Wow' or 'congrats' rather than probing; there is no pushback, no productive disagreement, and no follow-up that extracts depth beyond what the guest volunteered.

“Am I right? And so what you're saying is that with this application of AI on top of the data, this is a kind of process, a manufacturing system that will continue getting better”

“I got, I got to believe that's going to enable all kinds of efficiencies and, and new opportunities”