The Thinking Factory: How AI Is Giving MES a Brain
This episode is proudly made possible by FlowFuse. FlowFuse is the industrial application platform that helps operations, maintenance, and IT teams connect any machine, move data between systems, and scale industrial applications across sites. It is built on Node-RED, with the security and governance you need for real production environments. Learn more at flowfuse.com.
In this episode of Unplugged: An IIoT Podcast, hosts Phil Seboa and Ed Fuentes sit down with Francisco Almada Lobo, CEO and co-founder of Critical Manufacturing, to talk about the convergence of MES, IoT, and AI on the factory floor — and why most manufacturing AI projects stall before they ever reach production.
Critical Manufacturing was born from crisis. In 2009, the Siemens semiconductor factory in Porto where Francisco worked went insolvent. He had to lay off 150 people overnight. Rather than walk away, he gathered colleagues and started building a new MES from scratch — one that addressed the flexibility gaps they had experienced with existing commercial tools.
After years of struggle, the company found its footing. Today, Critical Manufacturing employs around 650 people across Asia, North America, and Europe, serving semiconductor, medical device, and industrial equipment manufacturers. "Nothing happened as it should in the early years," Francisco recalls. "My original business plan was formally very well done because I had recently done an MBA. But I was so naive."
Francisco breaks MES into four core pillars: process control and guidance, ensuring products are made correctly; traceability, recording every step for quality and compliance; operational visibility, showing what is actually happening on the floor; and flexibility, adapting quickly to new products, processes, and customers.
But that definition keeps expanding. Equipment connectivity and IoT automation now form a second major pillar. Analytics and data platforms — increasingly powered by AI — form a third. "We started with MES and then it evolves into this big thing," Francisco says. "The main problem is that people still understand this as MES. There is no other term."
An MES deployment can deliver an MVP within three months, but a full transformation can stretch over five years. And it is never truly done. "The MES is only worthwhile if the customer themselves own this and make this theirs," Francisco explains. "It is an infinite cycle. It is forever."
The core problem is straightforward: manufacturing data is fundamentally different from the domains where AI excels.
"If you apply AI in telecommunications, the financial world, retail — there is certain homogeneity and data sets available," Francisco says. "Now think about manufacturing. Data is hidden within systems, within factories. There is no scale, no homogeneity of processes."
This means AI can produce impressive demonstrations over manufacturing data. But those demos rely on models burning excessive tokens and tolerating error rates that manufacturing cannot accept. "We cannot talk about the percentage of failures in manufacturing," Francisco emphasises. "We are talking always about parts per million."
The numbers are stark. Audi had over 100 different AI initiatives across the company — and none of them scaled to production. The pilot succeeds. The production deployment does not. The gap is not the model. It is the missing context.
Critical Manufacturing's approach centres on enriching raw IoT data with MES context before AI ever touches it.
"If you are just measuring temperature from a sensor, there is very little you can do," Francisco explains. "But if you add information about which equipment it came from, what was the product being manufactured, when it went on maintenance last, who was operating it — then you can find the root causes."
The company builds what Francisco calls "fat events" — graph-based data structures assembled at transaction time that capture all relevant relationships. These are stored in specialised analytics databases, column-store rather than relational, and optimised specifically for AI workloads. The result: dramatically fewer tokens burned, faster responses, and significantly reduced hallucinations.
Francisco draws a critical line between using AI at design time versus in real-time production.
"We use LLMs and AI to generate processes — how they should work," he says. "But then the end result, the one that goes to production, is a deterministic, human-validated set of rules." This approach leverages Critical Manufacturing's low-code platform: AI generates configurations and metadata, not applications, eliminating technical debt and ensuring automatic upgrades.
For regulated environments like medical devices, this distinction is non-negotiable. The first FDA warning letter for unauthorised AI use in a manufacturing context has already been issued — a signal that regulators are paying attention. Deterministic execution is not optional. With the same inputs, the algorithm must produce the same outputs, every time.
Francisco's most sobering reflection concerns the pace of change. Previous technology revolutions allowed time for adjustment — for older workers to continue their way while younger ones adopted new methods, for organisations to evolve gradually.
"Right now there is nothing of that," he says. "The way we need to adjust within companies, within daily work, within our jobs is unprecedented. I have no idea if we are going to be able to do that without severe impacts."
He sees this playing out within Critical Manufacturing itself. People who were tremendously capable — even technology people — are struggling to adjust to the speed at which working methods are changing. The pace of change is unlike anything previous technology waves produced.
Manufacturing AI will not scale on better models alone. It requires the right foundations: enriched data, contextual understanding, and deterministic execution where it matters. MES provides those foundations. As Francisco puts it, the technology is ready. The question is whether factories — and the people in them — can keep up.
Book recommendation: Agentic Artificial Intelligence by Pascal Bonnet. Learn more about Critical Manufacturing at https://www.criticalmanufacturing.com. Connect with Francisco on LinkedIn at https://www.linkedin.com/in/falmadalobo/