Industrial data: from visibility to business value
For more than a decade, industrial digital transformation centered on visibility. Manufacturers invested heavily in connected assets, historians, dashboards, sensors, and IIoT platforms to gain greater operational insight.
Today, visibility is no longer the primary bottleneck.
Walk any major industrial event, from Hannover Messe to Smart Manufacturing Week, and you will see it: nobody’s pitching “we’ll help you see your data” anymore. Most industrial organizations I talk to aren’t short on data. If anything, they are drowning in it. The real challenge has moved from collecting data to actually getting value out of it.
The challenge has shifted from collecting data to converting data into decisions, and decisions into measurable business outcomes. That shift is where AI and industrial operations begin to intersect in genuinely transformative ways.
The market’s biggest bet: industrial context layers
As AI matures, industrial systems are shifting from reporting what happened, to recommending what should happen next, to (increasingly) executing approved actions on their own. We are witnessing a gradual progression from automation toward autonomy.
My view, and one I continue to test in conversations across the market, is that much of the industry’s AI discussion remains overly focused on models themselves. Foundation models are becoming commoditized fast. What’s actually scarce is industrial context, the thing that lets those models do something useful.
An AI model can detect an anomaly, generate a recommendation, optimize a schedule. Fine. But without understanding asset relationships, maintenance history, engineering constraints, production targets, safety requirements, business objectives, that recommendation isn’t worth much on the shop floor. I saw this firsthand during a recent visit to Schneider Electric’s Le Vaudreuil plant in Normandy: the AI itself wasn’t the impressive part; it was the decades of engineering knowledge, process understanding, and operational context underpinning the system. Context is what turns raw data into intelligence, and intelligence into action.
Why AI context requires ongoing maintenance, not just setup
Context also needs upkeep. Let me share an instance I recently came across: a manufacturer brought in a consulting team to build an AI model that controlled fan sequencing inside a set of curing ovens, fixing a temperature gradient that had been causing product defects for years. It worked, and the consultants moved on. About three months later, the problem came back, worse than before. The model had drifted, and nobody was monitoring it closely enough to catch it.
The actual cause turned out to be almost trivial: in the summer, operators would prop open a back door to cool the plant, and the draft created a cold spot the model was never designed to account for. The people on the floor had always known to compensate for this manually, but once the AI model took over, that tacit knowledge quietly stopped being used, and nobody thought to feed it back in.
It’s a reminder that context isn’t a one-time input, it has to be actively maintained, or a model will keep confidently executing on a picture of the world that’s gone stale. Treat AI models like any quality system: audit them on a regular cadence, rerun the same inputs, and catch drift before it becomes a failure.
How industrial vendors are building AI context layers
The more conversations I have across the industrial software market, the more it feels like vendors are arriving at a similar conclusion. Whether through acquisitions, digital twin initiatives, semantic models, knowledge graphs, or industrial data fabrics, vendors are racing to build operational knowledge layers that sit beneath AI capabilities.
Schneider’s acquisition of Cognite. Siemens pursuing its vision through Intelligence Center X, which surfaced repeatedly in conversations at Realize LIVE as part of Siemens’ effort to connect industrial data, engineering knowledge, and AI into a unified operational intelligence layer. Bosch’s Manufacturing Co-Intelligence push, which Norbert Jung was framing to me in Berlin as a “third layer of intelligence” sitting above the shop floor. Autodesk buying MaintainX. Velotic forming. Different roads, same destination: everyone’s racing to build the contextualized data foundation, the semantic layer, the knowledge model, the digital twin underneath the AI.
The emerging battle is no longer simply about delivering AI capabilities. It is about owning the context layer that makes those capabilities useful.
AI assistants vs. AI agents in industrial software
One data point which is relevant here: AI assistants are already everywhere in enterprise software, IDC estimates over 60% of enterprise apps now have some assistant or advisory capability baked in, while roughly 20% are pushing further into actual agent territory, systems that can independently perceive, evaluate, and act. That difference is not small. Assistants help people do the work. Agents start doing the work themselves.
And that has real implications for how enterprise software gets built. For decades, applications were designed around humans: log in, navigate a workflow, move data from one system to another. In an agentic world, that whole model starts to break down.
Why orchestration, not automation, is the real disruption
Instead of opening five dashboards, a user might just say: “optimize tomorrow’s production schedule while minimizing energy costs and avoiding maintenance conflicts.” An orchestration layer then pulls together specialized agents across scheduling, asset management, supply chain, and energy to actually get it done. It’s roughly the vision I kept hearing sketched out on the show floor at Smart Manufacturing Week and other industry events, just from different vendors, each convinced they’d own the orchestration layer.
So no, I don’t think the disruption here is automation. Industry has automated processes for decades already. The disruption is orchestration. Agents are becoming the connective tissue between applications, processes, and outcomes.
That doesn’t mean SaaS is going away, not even close. ERP, MES, EAM, APM, historians, supply chain systems, these stay critical because they are what agents read from and act through. But the value is migrating upward, away from the application itself and toward the orchestration and context layers sitting above it.
This is basically why I don’t buy the “SaaSpocalypse” narrative that’s been making the rounds. What’s actually happening looks less like the death of software and more like software evolving from a system of record into a system of intelligence.
For industrial software vendors, that raises an uncomfortable strategic question: if everyone can plug into the same foundation models, where’s the differentiation going to come from?
My take: Domain intelligence. Process models. Engineering expertise. Decades of operational history. Digital twins. Openness and interoperability. Contextualized industrial knowledge that’s been built up over years, not scraped off the internet.
Key questions for the future of industrial AI
The last decade was about digitizing operations. The next one is likely going to be about operationalizing intelligence. The data already exists. The models are becoming available to everyone. Context is emerging as the industry’s most defensible and strategically valuable asset.
Which leaves me with a few open questions I’d genuinely love to hear other views on:
- Will the next generation of industrial platform leaders be the ones with the best AI, or the ones with the richest operational context?
- Does the real control point end up living inside the application, or in the orchestration layer above it?
- As agents become the primary “users” of software, how should vendors be rethinking product design and pricing?
- Will industrial organizations trust a single vendor’s context layer, or will neutral data foundations end up winning out?
- And perhaps the biggest one: as we move from automation to autonomy, who actually owns the decision?
Got a question? Drop it in here.
