Best of LinkedIn: Smart Manufacturing & Industrial AI CW 29/ 30

00:00:00: provided by Thomas L. Geyer and Frennus based on the most relevant LinkedIn posts about smart manufacturing in industrial AI, in calendar weeks twenty-nine and thirty.

00:00:10: Frenness is a B to B market research company that supports Smart Manufacturing providers with building feature by feature competitive intelligence that shows exactly how their product stacks up against the competition.

00:00:21: you can find more info in description.

00:00:23: So today we are digging into a massive reality check.

00:00:27: I mean, We're seeing companies pouring massive amounts of capital sometimes billions of dollars Into artificial intelligence only to have their entire operation halted because they're missing you know A basic two-dollar electrical resistor

00:00:39: right.

00:00:39: yeah it happens all the time

00:00:41: and our mission for you Today is to basically cut right through that relentless hype?

00:00:47: So we're unpacking the reality of data plumbing, the physical automation happening on the floor and these structural market shifts that are exposing some severe bottlenecks in the global supply chain.

00:01:01: Yeah And it is a phenomenal time to examine this space because honestly The conversation is finally crashing into reality.

00:01:10: We are moving way past The theoretical whiteboards, we're actually confronting the friction of making these autonomous systems function inside this decades old incredibly complex industrial environments.

00:01:22: Okay so let's unpack that friction starting with a digital plumbing.

00:01:26: yeah I mean our manufacturers basically trying to build this high tech penthouse suite before they've even poured the concrete foundation because it really seems like everyone wants that shiny predictive AI dashboard.

00:01:38: but their underlying data infrastructure is just you know held together with duct tape?

00:01:42: Uh, yes.

00:01:43: That structural analogy hits the absolute core of issue perfectly.

00:01:46: I mean if you pour a cracked foundation that penthouse is going to collapse.

00:01:50: and there was this really sharp observation from Sam Proctor on this exact dynamic.

00:01:54: he argues introducing AI into a manufacturing environment doesn't actually create new problems.

00:01:59: Wait!

00:01:59: Really?!

00:02:00: It

00:02:00: doesn't!?

00:02:00: No not at all.

00:02:01: it operates more like this blindingly bright floodlight basically exposing the organizational rot already festering in dark.

00:02:10: Oh wow So if your data inputs are garbage, the artificial intelligence just becomes like... a highly efficient engine for generating garbage answers at scale.

00:02:19: Precisely the danger, yeah Proctor points out that poor data quality is rarely just a software glitch.

00:02:25: it's almost always a symptom of fragmented human processes.

00:02:28: you know we're talking about siloed legacy systems where The warehouse management software physically cannot speak to the procurement software right or there's unclear ownership Of data streams Or just a total lack of standard operating procedures if You don't restructure those Human foundations while your AI pilots fundamentally doom before it even boots up.

00:02:48: Which brings a pretty brutal reality check from Jacobo Lurek Kasal regarding how executives actually try to solve these problems.

00:02:55: Oh, this one was great.

00:02:56: Yeah

00:02:56: He highlighted the sheer absurdity of leadership teams chasing these bleeding edge AI pilots While their factories overall equivalent effectiveness Their OEE is sitting at a dismal fifty-eight percent.

00:03:08: All right.

00:03:09: And let's break down OEE for a second, because if you are operating at fifty-eight percent it means that nearly half the time your highly expensive machinery is either broken down sitting idle or producing scrap metal That has to just be thrown away.

00:03:23: and You know throwing a sophisticated algorithm at a machine?

00:03:26: That is mechanically failing Half of the day.

00:03:28: that does not fix The broken gears.

00:03:31: no exactly Just gives you A very expensive iPad dashboard telling you that the Gears Are Broken.

00:03:37: Jacobo noted that these massive digital transformation projects frequently fail because companies buy complex tech solutions before they actually understand their physical operational problems.

00:03:47: They'll just authorize a massive budget for a manufacturing execution system, an MES to supposedly fix skyrocketing energy costs or a lack of skilled machine operators.

00:03:58: But in energy bill, jumping forty percent is not an energy problem.

00:04:01: It's a visibility and maintenance problems like if an air compressor has massive leak buying a multi-million dollar MES system won't patch the lead.

00:04:10: you have to ask what single mechanical or process failure bleeding most capital?

00:04:15: And then find the simplest direct intervention

00:04:19: by way If want keep getting these kinds of no fluff mechanically driven insights help navigate this space.

00:04:25: make sure hit subscribe so catch our future deep dives.

00:04:28: But yeah, returning to that search for this simple direct intervention it requires a manufacturer actually understand their data

00:04:34: Absolutely.

00:04:34: And Victor M makes a crucial distinction here between two concepts that are constantly conflated.

00:04:39: He talks about data strategy versus data governance.

00:04:43: Oh, the restaurant analogy

00:04:44: Right!

00:04:44: It's perfect way to separate theoretical goals from operational reality.

00:04:49: So think of it like running high-end restaurants.

00:04:51: Your data strategy is your menu and overall concept.

00:04:55: It answers why

00:04:56: Why we opening this place?

00:04:58: Exactly, why are we opening this restaurant?

00:05:00: We want to serve premium Italian food to increase our profit margins.

00:05:05: In a factory setting the strategy might be well...we want to anticipate spindle failures on our CNC machines so that we can reduce unplanned downtime.

00:05:14: That's a strategy that creates direct business value

00:05:17: Right.

00:05:17: So the strategy is the target outcome.

00:05:20: Then data governance.

00:05:21: would like the health inspector in the restaurants kitchen.

00:05:23: That exactly how you visualize it.

00:05:26: Governance Is The How.

00:05:27: How is the data defined?

00:05:28: Who's allowed to input it and how do we make sure its clean.

00:05:32: In a restaurant, The Health Inspector ensures that ingredients are fresh, knives are sanitized And only trained chefs touch the stove.

00:05:40: So if you have brilliant data strategy look at great menu But absolutely zero data governance.

00:05:45: Your ingredients were rotten.

00:05:46: Yep

00:05:47: your data was entirely disconnected from reality.

00:05:49: The predictive models fail And floor operators instantly lose trust in system.

00:05:54: And decay of Trust is the exact mechanism Shankaraman was dissecting regarding factory digital twins.

00:06:00: Oh, The Digital Twins?

00:06:01: Yeah!

00:06:02: I mean we constantly see these incredible photorealistic three-D renderings of factory layouts right there's used to get project approvals from the board.

00:06:09: they look like high end video games and they just demo beautifully.

00:06:13: but Shankar points out that if those digital twins are not governed with the same PLM style lifecycle discipline as the product data They're essentially worthless.

00:06:23: Let's expand on that PLM mechanism for a second because that really is the differentiator here.

00:06:27: Product lifecycle management systems enforce rigid reality checks.

00:06:31: Right, think about how PLM works for physical product like designing car engine.

00:06:37: If an engineer goes into the CAD file and changes a bolt from ten millimeters to twelve millimeters, The PLM system triggers an automatic ripple effect.

00:06:44: Exactly!

00:06:45: The Bill of Materials updates... ...the Purchasing Department is automatically notified To order larger bolts.... The assembly instructions on floor are revised.

00:06:52: Yeah everyone operates From single source truth.

00:06:55: But with a factory layout someone On night shift moves A physical pallet rack twenty feet Just make room for a Forklift.

00:07:03: And because there is no automated ripple effect tying the physical floor to the digital model, The Digital Twin does not update.

00:07:11: The Model Rots.

00:07:13: Six months later, the Digital Twin has just this static dead file sitting on a server.

00:07:18: An engineer uses it to plan new conveyor belt installation Orders parts and when contractors show up they can't install them as a pallet rack is in the way.

00:07:26: The twin must be treated with highly governed managed system of record Not only one time snapshot

00:07:32: which perfectly anchors this entire discussion about foundational discipline.

00:07:36: Jothan Alexander mapped out what he categorized as the hundred and eighteen elements for AI success in manufacturing.

00:07:42: Right,

00:07:43: formatted visually like a periodic table.

00:07:45: A hundred-and-eighteen distinct element sounds overwhelmingly complex to implement.

00:07:49: It is massive taxonomy...for sure!

00:07:52: But the brilliance lies on how he bookends an entire framework.

00:07:56: The very first element—the absolute starting point you cannot bypass —is human And…the final element Number one, eighteen is wisdom.

00:08:04: He explicitly notes that wisdom Is the heaviest element in The table and critically it cannot be synthesized In a laboratory or generated by A server farm.

00:08:12: It is only forged in people.

00:08:14: That completely reframes the conversation honestly?

00:08:18: The ultimate foundation of artificial intelligence Isn't silicon its human Wisdom?

00:08:24: because You know, it takes a veteran machinist thirty years on the floor to understand why a motor hums at a certain pitch right before its seizes up.

00:08:32: A raw data tag can't give you that context

00:08:34: It proves.

00:08:35: if want autonomous systems work your foundation has capture hard one logic of workforce just as rigidly as you capture sensor outputs.

00:08:43: Well

00:08:43: since true manufacturing is ultimately physical we have look how this intelligence actually jumps from these digital servers down into The grease the metal and voltage of factory floor.

00:08:54: Right into the mess!

00:08:56: Exactly, we are tracking a massive shift toward physical AI.

00:09:00: Gartner even formally recognized this transition recently naming The Siemens Accelerator portfolio as market shaper in specific space.

00:09:08: But here's where it gets really interesting to me.

00:09:10: We hear all these sci-fi buzzwords about humanoid robots.

00:09:14: but isn't most effective automation usually just basic boring mechanical common sense?

00:09:20: Almost

00:09:20: always yeah.

00:09:21: Chris Sturgey provides a highly grounded perspective here.

00:09:24: He argues that right-sized routine automation is the true driver of scalable productivity...

00:09:30: So not The Walking Androids?

00:09:31: No, the goal isn't to deploy A walking talking Android—the goal is implement simple targeted systems that eliminate human ergonomic strain or remove operators from extreme temperature environments.

00:09:43: It is about applying basic, flawless repeatability to mundane tasks.

00:09:48: And sometimes that automation relies on incredibly granular old-school logic.

00:09:52: Mohamed Noman shared a fantastic for practical example of this.

00:09:56: despite all the high level consulting presentations about neural networks Sometimes innovation on the floor relies on basic electrical engineering.

00:10:03: I love

00:10:03: this example.

00:10:03: Right!

00:10:04: Imagine you have new sensor measuring pressure and it outputs standard four to twenty milliamp signal.

00:10:10: but your programmable logic controller your PLC, the brain of that specific machine only has an analog input designed to read zero-to ten volts.

00:10:20: So since Flix French...the PLC speaks German and they just cannot communicate?

00:10:24: Exactly!

00:10:24: And the standard corporate solution would be to halt the installation issue a purchase order for a dedicated expensive signal converter module wait for shipping and then have an electrician wired into the control panel.

00:10:35: Or, as Noman points out you can just use a basic two dollar, two hundred and fifty ohm precision resistor.

00:10:40: Yes!

00:10:41: You wire that resistor directly across the PLC input terminals.

00:10:44: It is pure basic Ohms.

00:10:46: law Voltage equals current multiplied by resistance.

00:10:49: So four milliamps times two-hundred and fifty Ohms gives exactly one volt

00:10:53: And twenty milliamps give you exactly five volts.

00:10:56: Instantly, the PLC can read the sensor's pressure data perfectly.

00:10:59: It is this brilliant cheap hyper-reliable physical translation that bypasses need for complex new hardware

00:11:06: entirely.".

00:11:06: That a phenomenal example of physics overriding hype!

00:11:10: Dr.

00:11:11: Elnaz Nyeri highlighted very similar physical dynamic but scaled up to industrial fiber lasers.

00:11:20: Well, when purchasing apartments are looking to invest in new laser cutting machines they almost universally obsess over the shiny objects.

00:11:28: You know?

00:11:28: The brand name on the chassis or...the total kilowatt power of a laser beam.

00:11:33: But raw power doesn't matter if machine keeps destroying itself.

00:11:36: Exactly!

00:11:37: The problem.

00:11:37: She notes that better manufacturers look for completely different mechanism which is auto height control

00:11:43: Auto Height Control.

00:11:43: Yeah

00:11:44: When you place some massive sheet of steel on a cutting bed That sheet never perfectly flat.

00:11:49: It has microscopic waves and as the laser applies intense heat, The metal warps in bows in real time.

00:11:56: Meaning that physical distance between a delicate laser head And steel surface is constantly fluctuating.

00:12:01: A miscalculation of just few millimeters results In catastrophic physical collision.

00:12:06: The head crashes into the steel ruining material Shattering lens completely shutting down production line.

00:12:12: That sounds expensive.

00:12:15: Auto-height control utilizes a high speed capacitive sensor to continuously measure the surface topography.

00:12:22: It dynamically adjusts Z axis, you know up and down movement in milliseconds as it cuts over the warped metal.

00:12:29: And that specific sensing mechanism which is available on far more competitively priced tiers of machinery now Is what actually guarantees throughput not logo on machine.

00:12:40: its all about dynamic real time optimization.

00:12:43: Precisely Grandma Puro hit.

00:12:45: Jethrotha shared concrete data on this exact type of optimization happening with CNC machining at a company called Sanssara.

00:12:53: Oh yeah, they deployed Siemens adaptive control right?

00:12:55: Yes across more than twenty of their machines.

00:12:58: Now, typically a CNC machine runs the static block of G-Code.

00:13:01: The code just has spin the tool at one thousand RPM and push it forward two inches per minute

00:13:05: And the code is completely blind.

00:13:07: It doesn't know if metal alloy isn't usually hard today or if tools are getting dull.

00:13:14: But adaptive control changes mechanism It allows to actually feel cutting resistance By dynamically optimizing the feed rates based on those real-time physical conditions like slowing down when it hits a hard spot or speeding up through empty space, they cut their cycle times by fifteen to twenty percent.

00:13:32: That is massive!

00:13:33: Right.

00:13:33: that translates to recovering up an hour and half of productive machining time per machine every single day while actively preventing tool breakage.

00:13:42: If we pull back in look at broader implications Brent Roberts summarized this entire industrial shift For decades, the ironclad rule of automation was deterministic.

00:13:52: You execute the hard-coded program exactly as written over and over but we have reached a limit on that logic.

00:13:59: product variations are too rapid environmental conditions fluctuate supply chains or chaotic.

00:14:04: you simply cannot hardcore to response for every possible physical variable anymore which is driving this shift toward egenic AI.

00:14:10: at The Edge The intelligence is migrating out of the climate-controlled server room and embedding itself directly into physical machinery.

00:14:18: Into intelligent drives, virtual PLCs, microsensors on cutting tools...

00:14:23: A machine becomes an autonomous agent that can detect a variable reason through physics and adapt its behavior without waiting for human programmer.

00:14:33: So we have to ask what does this mean?

00:14:36: If the brain is migrating to the physical edge and every manufacturer needs this adaptive capability, are we watching a massive Pac-Man scenario here where a handful of colossal software giants or just gobbling up the entire industrial ecosystem?

00:14:51: Well.

00:14:51: The consolidation happening right now is incredibly aggressive.

00:14:54: Michael Finocchiaro has been tracking the acquisitions And he noted that five distinct Industrial AI operations startups were swallowed up in a window of just nine weeks

00:15:04: A nine week window as complete feeding frenzy

00:15:06: Truly.

00:15:07: But let's look at why this specific sector is so hot!

00:15:10: Every single one of those acquired startups focused on the operations side, they run the factory-side dealing with maintenance simulation and floor data.

00:15:20: meanwhile The software companies focusing on the design side like CAD and PLM tools are largely being ignored in this M&A wave

00:15:27: Because the design phase is a one-time event, but the operations phases where the daily bleeding happens.

00:15:33: If a company can own the software that stops the factory from breaking down on a Tuesday.

00:15:38: The recurring value of that is astronomical compared to just designing the product

00:15:43: Absolutely.

00:15:43: And the biggest bite in this operational Pacman game Just finalized.

00:15:47: Schneider Electric entered an agreement To acquire Cognit In a staggering three point One billion dollar deal.

00:15:54: A three billion dollar move signals a major structural shift.

00:15:58: What is the core mechanism of Schneider's actually buying there?

00:16:01: Well, according to Mandeep Sedo, Schneider is not just buying massive data repository.

00:16:05: they are acquiring Cognitz Industrial Knowledge Graph technology.

00:16:08: Okay break that down.

00:16:09: for us

00:16:10: To understand why this matters we have to contrast it with standard databases.

00:16:15: A traditional relational database is essentially giant Excel spreadsheet.

00:16:19: It holds isolated datapoints A knowledge graph, however is structured like a multi-dimensional spiderweb.

00:16:25: It maps the live physical cause and effect relationships between assets.

00:16:29: So it understands the physics of the entire facility not just the isolated numbers.

00:16:33: Yes

00:16:33: exactly If you are operating a massive chemical plant a valve doesn't exist in a vacuum.

00:16:39: The knowledge graph understands that if valve A gets stuck closed, pump B will overheat within ten minutes which would then cause tank C downstream to dangerously depressurize.

00:16:51: It maps those complex physical dependencies so the AI can actually reason about the environment.

00:16:56: That mechanism is a bridge between a daft board just shows you warning light and an egenic system that can autonomously shut down the pump.

00:17:04: And Schneider is clearly not the only Titan aggressively pursuing this agentic capability.

00:17:10: SAP is fundamentally restructuring their approach as well, they recently appointed Dominik Mark Metzger.

00:17:20: Yeah.

00:17:21: Their strategy is not to build bolt-on generic chatbots, they are weaving intelligence directly into the connective tissue of their business processes.

00:17:30: They're deploying thousands of intelligent agents designed to autonomously handle real time production scheduling workforce coordination and compliance tracking.

00:17:39: But wait If SAP software is autonomously deciding how to route materials and schedule human workers, doesn't that create a massive black box problem?

00:17:48: if the software makes a scheduling error that costs a million dollars.

00:17:51: How does a plant manager audit the reasoning of thousand invisible AI

00:17:55: agents?".

00:17:55: That's

00:17:56: exactly what Tension The Market is wrestling with right now!

00:17:58: The software is becoming an active autonomous participant in running business which demands entirely new level trust and verification from human operators.

00:18:08: We have pivot here though from this limitless digital software expansion to the severe unyielding realities of physical macro constraints.

00:18:16: Robert Little brought up TSMC's staggering two hundred and sixty five billion dollar commitment, to reshore advanced semiconductor manufacturing.

00:18:24: in Arizona We are talking about constructing ten bleeding edge fabrication plants at R&D centers To build the most complex physical items humanity has ever produced.

00:18:35: It is a breathtaking deployment of capital into physical infrastructure, but it exposes a massive friction point.

00:18:41: You cannot simply blast a region with billions of dollars and expect an advanced supply chain to instantly materialize.

00:18:48: Mike Nager isolates the core bottleneck here.

00:18:50: Capital is digital and highly mobile.

00:18:52: you can wire a billion dollars across the globe in seconds that the specialized manufacturing workforce is intensely local and highly immobile.

00:19:00: You cannot take a skilled worker who spent ten years assembling internal combustion engines to just drop them into a semiconductor clean room, though required tolerances —the chemical handling protocols—and operational cadence… it's fundamentally an alien

00:19:12: ecosystem.".

00:19:13: Exactly!

00:19:14: Even with infinite capital if you lack the hyper-specialized talent pool... That money just gets trapped in empty concrete shells.

00:19:21: And Shondatsun provided some sobering data that validates that exact friction, despite the daily headlines about reshoring and massive factory groundbreakings actual manufacturing employment in this sector declined by seventy-five thousand jobs over a seventeen month period starting in twenty twenty five.

00:19:39: A drop of seventy-five thousand jobs completely contradicts the narrative of a manufacturing renaissance.

00:19:45: What are the macro mechanisms driving that decline?

00:19:48: It's a compounding series of physical and economic realities.

00:19:51: First, tariffs are artificially raising the cost of imported electronics in raw tooling squeezing operation budgets.

00:19:58: Second supply chains that are divesting from China or not automatically returning to

00:20:02: U.S.,

00:20:02: they were frequently shifting to Mexico or Vietnam where labor infrastructure is already established.

00:20:08: Finally The high interest rate environment means financing actual heavy machinery.

00:20:12: fill those newly constructed factories is prohibitively expensive.

00:20:16: Announcing a factory is PR exercise.

00:20:19: Sustaining the payroll on production lines, is brutal economic

00:20:22: test.".

00:20:23: And constraints aren't even limited to human labor!

00:20:26: We are hitting physical ceiling of raw materials driven by this exact AI boom.

00:20:31: Mike Wang noted that exploding demand for AI server racks and high-performance compute clusters is strangling supply chain for foundational components... Specifically, he highlighted a severe global shortage of Panasonic Megatron VI PCB materials.

00:20:47: Megaton VI is the highly specialized resin and copper material required for ultra-high speed high frequency printed circuit boards correct?

00:20:54: Yes it's physical bedrock that advanced AI chips sit on And the lead times to acquire are now stretching past six months.

00:21:00: It was an incredible irony.

00:21:01: Every industrial giant wants to deploy agentic AI to optimize their factories, but they cannot physically acquire the specialized circuit boards required to build the servers that run the AI.

00:21:11: The digital ambition is being choked by the physical supply chain...

00:21:14: Which brings us a final deeply provocative thought as we wrap up.

00:21:20: If we synthesize everything we've covered —the necessity of data governance— the push for physical edge automation and these severe macro bottlenecks We need to examine a phenomenon James Rolston from KPMG identified in the life sciences sector.

00:21:34: How does the Life Sciences Sector forecast what's coming for traditional manufacturing?

00:21:38: Well, In life sciences The r&d and discovery phase has been completely digitized And revolutionized by AI.

00:21:45: algorithms can now model protein folds identify drug targets and design optimized molecular structures in a matter of weeks A process that used to take human researchers years of trial-and-error.

00:21:55: Wow.

00:21:56: The digital brain is operating at warp speed, but when it comes the time to actually produce that drug physical manufacturing has not evolved Precisely.

00:22:09: They have AI accelerated drug discovery, but they still face massive multi-year delays in scaling up bioreactors because the factory floor store lies on manual process characterizations and fragmented equipment data.

00:22:23: The bottleneck didn't disappear it just shifted violently downstream to the production line.

00:22:28: Right.

00:22:28: So here's the mechanism you have to evaluate within your own operations.

00:22:33: What happens when generative AI completely accelerates your engineering design and discovery phases, but you're physical manufacturing constraints?

00:22:41: Your data plumbing or rigid PLCs.

00:22:43: You talent pool cannot keep pace.

00:22:45: will the stubborn analog reality of the factory floor become The ultimate speed limit for human technological progress?

00:22:51: if you enjoyed this episode new episodes drop every two weeks.

00:22:55: Also check out our other editions on digital construction and digital power tools.

00:22:58: Thank you for joining us as we explore the mechanisms behind these trends.

00:23:01: Keep questioning those bottlenecks, verify your data foundations and we will catch you next time.