Things are a bit quiet around the world today, so had a few moments to catch my breath and reflect.

Last month Amir Taiar and I found ourselves back in Germany, splitting time between Nuremberg's Christmas markets (too much Glühwein) and SPS halls (not enough comfortable shoes), at one of Europe's leading automation trade shows.

It was half a year after walking the halls of Hannover Messe, and after the AI craze I felt in Deutschland last April, I was curious to see if/how/why the conversation had evolved.

TL;DR: It's gotten more real, more interesting, and almost impossible to satiate with fluff (or as we now warmly refer to it:‘slop’).

AI Without the Fanfare

The first thing I noticed were the eye rolls (or rather, the more European subtle up and to the left head lift with a slight smile) that Igot when I said we were building Industrial AI. Once we shared more about what we were doing, the skepticism stopped, but folks were tired of the hype.

The massive "AI-powered" banners were fewer. The breathless promises were replaced with quieter demonstrations. AI was just...there.Embedded. Expected.

We had to say goodbye to the novelty factor that we were able to ride on just a few months ago.

This shift matters. When tech stops being a spectacle and becomes a tool, that's when real adoption happens.

The industry moved from "Can we use AI?" to "How do we use AI?" Over 77% of manufacturers have implemented some form of AI now, up from 70% in 2023. 82% are planning to increase their AI budgets in the next 12-18 months. (Side note: folks were expecting this to happen with robotic humanoids over the last year, but unlike AI + software, we’re seeing that those robodogs and Musk-army bots are still dancing in the spectacle spot…).

The Knowledge Transfer Urgency

The paradox: This rise in AI and automation (and humanoids) are making the related components and system build way more sophisticated(aka, not easily handed over to AI), and supporting them requires even deeper engineering knowledge than ever before.

Selling a factory automation system with vision-based inspection,functional safety, and plant-level software integration is a hellova lot more complex than selling forklifts.

At the same time, more than a quarter of my conversations were with veteran engineers carrying decades of application expertise that were about to retire (the avg. retirement age in Germany is 65.8 years old) or who'd pushed off retirement because they hadn't trained a replacement yet. And it’s not just Europe; there’s also nearly 4 million workers retiring annually in the US alone.

In addition to the actual retirement scenarios, there was lots of hush-hush talk about “voluntary early retirement”, because many companies in Europe (particularly German industrials) are more cautious these days about hiring new staff, with many of them finding legal ways to cut jobs.

Preserving the ‘human knowledge’ as a competitive advantage before it walks out the door (whether by choice or by pressure) is a top top top concern of industrials.

Companies need to capture what their best people know (the troubleshooting logic, the configuration shortcuts, the customer-specific insights) and make it accessible, especially for new products that are orders of magnitude more complex than what their predecessors dealt with. Cue in AI.

The Implementation Question

In Hannover, the attitude was "prove it works, and we're in."

At SPS, that question became: "What do I need to put into this so it works for me”.

Many were worried about the lack of internal expertise (45%) and difficulties integrating AI with existing legacy systems (44%). They specifically cared about:

• manpower (aka, how many team members need to be involved to get this started)
• time (how fast can I see value)
• resources (not just money, but things like IT members involved to validate)

These are fair concerns in an industry where production can't stop for a software rollout.

Lesson for us at Bardin: it's not enough to build powerful AI; weneed to build AI that doesn't require an army to deploy or maintain.

The "Build vs. Buy" Dilemma Weighed Heavy

Companies like Beckhoff, Siemens, and ABB are absolutely building internal AI tools. Sophisticated ones.

But not everywhere or for everyone.

They're building for their coding teams, production floors,product development. Makes sense, that's their core competency. But their RevOps? Sales enablement? GTM functions? Still looking outward.

A 2024 Deloitte survey found that 61% of high-performing AI adopters build at least part of their stack in-house, primarily for customization and data control in core operations. Not every business function.

Sensor companies are experts in sensor AI. Their core operations are building those sensors. Are they experts in sensor sales AI? No. That's our game.

The Copilot Disappointment

Almost every major industrial software solution provider had a"copilot" or "agent" at their booth. The demos looked slick, but as we heard from many actual users: the reality was a letdown.

The copilots, joules, and einsteins (keeping the names lowercase intentionally ;) were fine for basic questions. "What's the status of this order?" Sure. "Pull up this customer's history." No problem. But the moment you get into actual application engineering, configuring a complex automation system, troubleshooting integration challenges,recommending the right component for a specific use case, they fall apart, or worse: make shit up.

The gap between general-purpose AI tools and specialized industrial knowledge is wider than most folks realize.

You can't bolt an LLM onto an ERP and expect it to understand the nuances of industrial sales.

Even internally built copilots need a vast database of expert knowledge that must be embedded in the rightway, structured well for industrial use, and maintained. That's where we come in.

Specificity Wins

Here's what actually gets industrial companies excited: very specific use cases solving very specific problems.

The other ‘hot’ AI startups we saw at SPS were the ones building vision AI to detect microscopic defects in semiconductor chips.Crunching operational data to build digital twins of a production line.Automating CAD workflows for specific component types.

Specificity. Precision. Clear ROI.

When we showed our product variant chooser (finding the right component without needing parametric search expertise) people leaned in. When we demonstrated our smart questionnaire for pre-sales feasibility assessment, replacing hours of back-and-forth with structured intelligence, that's when we heard the magical words of “ah, we gotta show this to Franz”.

Industrial buyers don't want AI platforms that can do everything.They want workflows built for their exact pain points. AI tools need to understand their specific domain deeply enough to actually be useful,not just impressively general.

Trust re: GenAI vs ML

There's real mistrust of GenAI in industrial settings. Not classic ML/data-science AI, but generative AI. Specifically around data extraction and modification.

Makes sense.

When dealing with technical specs, compliance requirements, and precision engineering data, you can't afford hallucinations. A primary obstacle is ensuring data confidentiality, especially when using public LLMs and companies want clear distinctions between what's AI-suggested and what's verified source data. Transparency is a non-negotiable.

Interestingly, the same companies skeptical of GenAI were comfortable with machine learning and data organization that provide transparent, controllable insights. GenAI's "opacity" is the concern. Industrials want to see the source, understand the reasoning, maintain control.

Our AI needs to show its work (and we learnt a lesson in storytelling).

What's Changed

Last year, AI was the exciting possibility everyone wanted to talk about. This year, it's the practical reality everyone wants to implement correctly. The global industrial AI market reflects this, having reached $43.6 billion in 2024 and being forecast to grow at a CAGR of 23% to $153.9 billion by 2030.

The conversation has matured. The sophistication has deepened. The expectations have sharpened.

Every eyeroll/insight/critique/excited-intro from SPS reinforced that the industry needs AI tools that:

• Deploy without massive internal resource commitment
• Specialize in industrial complexity rather than trying to be everything to everyone
• Provide transparency and control, not black-box answers
• Bridge the gap between technical sophistication and sales capability
• Integrate into existing workflows rather than demanding wholesale replacement

The GTM AI gap we identified at Hannover Messe hasn't closed. If anything, it's widened as companies realize their internal builds won't extend this far and generic copilots fall short.

Still Lucky to Be Building Here

A year ago, I wrote that we felt lucky to be building for industrials.

That feeling has only intensified.

This is an industry undergoing genuine transformation, led by pragmatic decision-makers who want solutions that actually…work. They've got real ($$$) challenges: product complexity, knowledge transfer,resource constraints, that have clear answers if you're willing to build them right.

We're building those answers, and I'm more convinced than ever that we're building them in the right way.

Happy Holidays!

‍