Industrial AI Podcast

How do you scale AI in a heavily regulated company with more than 100,000 employees? Jan Tönnissen has spent over 18 years at Bayer, most recently leading an IT delivery platform behind MyGenesis. The internal GenAI and agentic platform grew from a few hundred users to 25,000 and more than 60,000 assistants. He also built Bayer’s AI community of 19,000 colleagues.

In this episode, Jan explains why guardrails don’t have to slow people down and can instead pave the way. He also describes how the EU AI Act became a catalyst rather than a hindrance. He walks through Bayer’s Agent Hub, a “telephone book” for every agent in the company, and the AI gateway that gives full transparency over LLM and token consumption.

We also cover:

Real use cases: research literature screening cut from weeks to days, marketing content, and “Friday,” Bayer’s new agentic assistant for every employee
The token debate: from maximizing to minimizing use, self-hosted open-weight models, small language models, and counting token costs as CapEx
AI-assisted engineering: where it delivers 10x and where it delivers 3x
Who owns agents in production, and why ownership has to sit with the business
The next wave: physical AI and agents moving closer to the shop floor
Three takeaways for midsize machine builders: build a coalition of the willing, choose your first battle carefully, and define your build-vs-buy decision tree up front

And one closing reminder: if you can’t measure it, you can’t steer it.

We’re seeing an explosion of AI in engineering: models turning drawings into 3D designs, agents helping create plant layouts, and new tools challenging the way engineers work. So where will the value come from next—engineering software, AI platforms, or the context data behind them? I also explore whether agents could leap across the traditional automation pyramid and connect directly with industrial systems.

Then I talk with Bosch’s Thorsten Brach about connecting AI agents to industrial robots using MCP, MHS, and SRCI. We get practical about robot integration, deterministic PLC functions, and why wheeled robots may arrive in factories before legged humanoids. The big question: how do we make these systems easier to use without losing the expertise needed to keep them safe and reliable?

Did OpenAI really solve the Navier–Stokes problem? In this episode, we ask mathematician Gonzalo Caolabora. and AI professor Günter Klambauer what the proof means, why it has been difficult to verify, and whether large language models are discovering mathematics or combining ideas from existing research.

We also discuss why AI can excel at problems with verifiable answers, where today’s systems still struggle, and how AI could reshape mathematical research and careers. The big question: as proofs become more automated, what will mathematicians focus on next?

We talk with Arburg’s Stefan Lamp and Werner Faulhaber about how AI is changing injection molding—from predicting process stability and quality to helping operators troubleshoot machines. They explain why time-series foundation models matter, where domain expertise makes the difference, and how simulation can shorten the setup of a new process.

We also get into the practical questions: Can agents safely recommend machine settings? What holds back cloud connectivity and predictive maintenance? And how should manufacturers measure AI’s real-world value? A grounded look at what’s working now—and what still needs solving.

In this episode, I sit down with David Rodriguez to explore how large language models are reshaping the way we interact with industrial data. We dive into the Pi System’s pivotal role across foundational industries and discuss practical strategies for integrating AI—without compromising sensitive operational information. David shares his experience on creating actionable knowledge graphs, running local LLMs, and the evolving relationship between human operators, agents, and data visualization. If you’re curious about the real-world challenges and future promise of AI in manufacturing, this conversation delivers clear insights and actionable takeaways. Join me as we unpack what it takes to move from data to decision—safely, efficiently, and intelligently.