Hanna Fiori, Plant Manager at Super Agile Paper (SAP), watched the morning sun glint off the facility’s green roof, designed to capture and reuse rainwater. Inside, the mill was a study in controlled energy. The air hummed not with the cacophony of old industry, but with the quiet whir of automated guided vehicles (AGVs) gliding along magnetic paths and the whisper of high-speed rollers spinning out specialty coated paper. Fewer people walked the floor; most, like Hanna, occupied the central control room, a nerve center of intelligent dashboards and augmented reality displays.

Before she could take a first sip of her coffee, a subtle, personalized notification bloomed on her main screen, seamlessly integrated into her Microsoft 365 environment. It wasn’t a jarring alarm, but a calm advisory from Joule, the facility’s AI copilot.

"Morning, Hanna. I've detected a probable 'coating mottle' anomaly pattern on the current high-gloss run for S. Gloss, Batch 011235. Confidence is 96%. It's minor, but it will exceed the customer's specified gloss variance within 90 minutes. I've already initiated a root cause analysis."

Hanna, unfazed, leaned toward her console’s mic. "Joule, give me the rundown. Correlate machine health, process parameters, and raw material data for Batch 011235."

The response was instantaneous, a multimodal summary of text, charts, and a shimmering digital twin of the production line. "The mottle pattern," Joule’s synthesized voice explained, "correlates with a micro-vibration spike of 0.8 microns in Coater Unit 3, synchronized with a 2°C thermal increase. This is a classic signature for impending bearing failure; my predictive model estimates a 78% chance of critical failure within 72 hours. I've also flagged a 4% variance in the lignin content of the pulp batch from 'Nordic Fiber,' which is likely a contributing factor to the coating inconsistency." The diagnosis was a perfect fusion of sensor data—vibration, thermal, and ultrasonic analysis—that moved far beyond simple alerts to a concrete, actionable story.

"Have we seen this pulp variance from Nordic Fiber before?" Hanna asked, already anticipating the answer.

"Cross-referencing our data on the SAP Business Network, two other manufacturers reported similar minor quality deviations linked to Nordic's batches in the last quarter. The data is anonymized, but the pattern is clear. I've already drafted a quality notification to Nordic Fiber, including our sensor data, and have updated the incoming inspection protocol for their next delivery to include a rapid spectroscopy scan for lignin content." The network wasn’t just for placing orders; it was a collaborative ecosystem for quality and resilience, a concept pioneered by real-world companies like UPM.

Hanna took a sip of her now-lukewarm coffee. The time for complex decision-making was over before it had even begun. "Okay, Joule. Let's get ahead of this. Execute the 'Predictive Quality Aversion' playbook." This was a custom workflow her team had designed using Joule Studio, a low-code tool for creating company-specific AI agents and skills.

"Execute the playbook," she confirmed. "Pause Batch 011235, route the affected roll for recycling, schedule the maintenance on Coater 3 for the next planned changeover window, confirm the replacement bearing is in stock, and push the 'Dynamic Adhesives' order up the schedule. Send the updated plan to the shift lead for confirmation. And add a note to my one-on-one with Nordic Fiber's account manager."

A series of confirmations flashed on screen as Joule’s agents autonomously executed the multi-step workflow. The pause command was sent directly to the line via SAP Digital Manufacturing, while maintenance and procurement systems were updated in parallel. Within seconds, a sleek, low-profile Automated Guided Vehicle (AGV) silently detached from its charging station and navigated towards Coater 3. Controlled by SAP Extended Warehouse Management (EWM), the AGV knew the exact location of the rejected roll. SAP EWM’s task management system had already optimized its route, ensuring it didn't interfere with other automated or human traffic. With a precise mechanical lift, it secured the multi-ton roll and transported it to the recycling pulper, its status updated in real-time across the entire system. Simultaneously, another AGV was dispatched from the raw materials warehouse, carrying the specific adhesives needed for the newly prioritized production run, a perfect example of just-in-time supply orchestrated between SAP EWM and the production schedule.

Hanna’s gaze shifted to another widget on her dashboard, this one glowing with green accents. It was her real-time view into the plant's sustainability ledger. The crisis with Batch 011235 wasn't just a production issue; it was a sustainability event. With increasingly strict legal requirements on the horizon, Super Agile Paper was already using SAP's Green Ledger initiative to track its environmental impact with the same rigor as its finances. For every production order, the system captured not just material and energy costs, but also the associated carbon footprint—Scope 1, 2, and 3 emissions—at a transactional level. The data was auditable, transparent, and linked directly to financial data, allowing her to see the true cost of waste and inefficiency. This granular view meant decisions weren't just about profit anymore; they were about balancing profit and planet, a core principle of the company's strategy.

Hanna leaned back, finally taking a proper drink of her coffee. Her colleague, Marty, walked by, glancing at her screen. "Everything alright, Hanna? You look suspiciously relaxed for a Tuesday."

Hanna smiled. "Joule's got it. The biggest decision I have to make this morning is whether to have a second coffee."

Sounds a bit over the top? Perhaps. But how do you see the future? Some of these capabilities are already a reality, while others are just over the horizon. What's certain is that the future builds on the foundation of today, and every great journey begins with a single step. You can start yours by exploring what's possible with collaborative AI agents at https://www.sap.com/products/artificial-intelligence/ai-agents.html.

Read why Gartner names SAP S/4HANA Cloud a Leader in its 2024 Magic Quadrant™ reports for cloud ERP for product-centric enterprises.

For more about how SAP has partnered with the Mill Products industry – including paper and packaging - for more than 50 years click here.  

Last month I referred to Amir Husain from his excellent book “The Sentient Machine” to address concerns of the rapid adoption of Artificial Intelligence (AI). This month I will refer to Emerson Chief Technology Officer Peter Zornio, Founder of Reputation Lighthouse Bonnie Caver, director of the Center for Information Technology Policy at Princeton Arvind Narayanan, and Joe Rogan.

Joe Rogan deserves attention because he has arguably the largest audience on the planet and is outspoken in his concerns about the threat of AI. Rogan is very alarmed that AI is growing without any regulation and there is a risk that AI along with automation may eliminate jobs and become our lords and masters.

Peter Zornio was recently interviewed by Sanjay Puri on the CAIO Connect Podcast, which can be found on YouTube and is well worth watching. Among the points made by Peter is that in industry we always have a human in the loop. Frankly industry has been using AI for about 50 years, so it is not new to us. We know how to deploy prediction models for advanced process control, optimization, and predictive maintenance. There are no implementations in industry I know of that use closed loop unsupervised learning. We have a lot of experience with closed loop AI with a human involved in pre-processing data to remove noise and outliers, ensure the model is supported with first principles, deploy the model in a deterministic environment, and have human operators capable of intervening when required. If humans are ever taken out of the loop, industry is nowhere near that point today.

Bonnie Caver was recently interviewed in the Austin Tech Connect podcast about responsible AI implementation. Bonnie says she sees a lot of companies adopting AI without defining what problem they are trying to solve. She finds a lot of failures when firms adopt AI tools because they think they have to. Successful implementation happens when there is a problem identified, and AI can be focused on solving that problem. This mitigates some of the AI concerns. Another observation Bonnie has is that when companies attempt widespread adoption of AI tools and lay off workers, they find that AI does not retain the institutional knowledge that has left, and those firms end up rehiring those people as contractors. It also results in companies losing their differentiator because they become just another user of the same AI tools that others are using. If you don’t have the people that make you special, why would someone buy from you? If people are concerned that AI will take over industry, it presumes the unfocused approaches that fail will succeed. Bonnie says the successful AI implementations will focus on solving problems with humans in the loop.

This leads me to Arvind Narayanan, who along with Sayash Kapoor authored “AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference”. This book was published in September 2024, and at the pace that AI is advancing may be eclipsed by more recent publications. However, Arvind gave an excellent presentation of his book in April 2025 to the MIT Schwarzman College of Computing which can be found on YouTube. The bottom line of the book is that most AI capabilities are over hyped. This is particularly the case in predictive applications. Arvind shows many examples of AI being used to predict human behavior in which the accuracy of the AI predictions is barely better than a coin flip. Generative applications, where AI is producing something like responses to chat, artwork, and designs, are much better. In industry, our closed loop applications combine predictions with the generation of optimal setpoints that control valves and motors. While industry has humans in the loop, Arvind shows how consumers and non-industrial uses without humans in the loop are not nearly as successful as claimed.

This article may come across as one sided. I recognize that there are legitimate concerns for the theoretical harm that AI could do. Joe Rogan is very smart and well read. He is like a lot of other very respected people ringing alarm bells about AI. Since Peter and Bonnie are in Austin, they might be good guests for his podcast, and Arvind certainly would attract some listeners with provocative titles like “AI Snake Oil”. Perhaps there is some common ground. 

For our AI uses in industry, we need to apply Bonnie’s recommendation to identify focused problems, Peter’s emphasis on determinism and conformity to known engineering principles, and Arvind’s caution about hyped claims. Don’t buy snake oil.

Kai, you are outdoing yourself my friend, and moving into the area I have dreamed of for years.

And Pat, you are correct, we have been using a variation of AI for decades.

The challenge now is to move our metrics within measurement and control across the accounting domain, across the engineering domain, to a common measurement.

I am going to propose a common measurement of which neither of you will be comfortable.

That measurement is earnings per share. Here in the United States this will usually be pennies per share. It should be just a few more simple arithmetic calculations to transpose from the engineering or accounting numbers we get now to cents per share.

In Kai's example above, that will let us know if the excursion in specifications is worth the costs correct it.

In Pat's example, Bonnie Caver will not have to worry about what problem to solve.

The integration of edge computing and IoT into industrial operations addresses several critical needs.

Industrial IoT platforms can’t deliver full value if networks lag behind. Real-time insight depends on infrastructure that most manufacturers are still underinvesting in.

Overcoming unique issues in the face of escalating attacks.

As manufacturers grapple with the uncertainty of tariffs and their potential impact, many are speculating whether digital manufacturing technologies–such as additive manufacturing–might be key to overcoming these challenges. Tariffs and their continued impact also highlights a related and pressing issue: the need to invest in workforce skills.