Best of LinkedIn: Smart Manufacturing CW 23/ 24

00:00:00: This episode is provided by Thomas Olguyer and Frennus, based on the most relevant LinkedIn posts about smart manufacturing in calendar weeks twenty-three and twenty four.

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.

00:00:23: Well, welcome to the deep dive.

00:00:25: We are really thrilled to have you sitting in on this conversation with us today.

00:00:30: whether you're commuting to the plant or just taking a break from the shop floor?

00:00:34: Yeah absolutely and our mission is highly focused.

00:00:36: we've been combing through the absolute top smart manufacturing insights surfacing across LinkedIn recently

00:00:42: Right.

00:00:42: And what we're seeing is this, well it's an undeniable shift right?

00:00:45: Moving away from the buzzwords and the shiny slide decks

00:00:49: Exactly!

00:00:49: The industry has finally moving out of that pilot purgatory into real autonomous execution.

00:00:55: So today we are exploring What actually works

00:00:58: From the unglamorous data foundations That make AI function All the way to physical humanoid robots stepping onto your shop floor.

00:01:06: It's gonna be a fascinating ride But you know before We get in to the flashy AI stuff We really have to talk about the data foundations because, well there's a huge reality gap right now.

00:01:18: Oh massive!

00:01:19: Yeah getting into that autonomous state doesn't start with just you know buying an AI model off-the-shelf.

00:01:25: it starts with confronting the plumbing of your factory

00:01:28: The plumbing?

00:01:28: I like that.

00:01:29: yeah i was actually reading this fascinating analysis by Dina Kaur Ramamrithi and he used a term that really stuck with me.

00:01:34: He called it structural chaos

00:01:37: Structural Chaos.

00:01:38: That sounds about right for most enterprise architectures.

00:01:41: Right, he argues that Enterprise complexity in manufacturing is essentially this mountain of structural chaos.

00:01:48: but the really interesting part Is how it gets built?

00:01:50: Oh so

00:01:51: well It isn't build out of malice or incompetence.

00:01:54: its built one completely reasonable decision at a time.

00:01:57: oh wow yeah I mean that perspective is so validating For anyone working in operations.

00:02:03: think About your own facility right now.

00:02:04: like How many temporary integrations are currently holding together Your production schedule?

00:02:08: Probably too many to count.

00:02:10: Exactly, it's that one urgent Friday afternoon production fix that just well became permanent.

00:02:16: Yeah.

00:02:16: or that custom PLC integration where someone basically hardwired the brain of an individual machine To perform a specific

00:02:23: task and effectively trapped its data locally forever

00:02:26: exactly.

00:02:27: Or uh It's that One heavily macro enabled spreadsheet managing inventory That nobody dares to touch because The person who built it left three years ago.

00:02:36: Yes

00:02:37: every single one of those work around solve an immediate pressing problem in the moment.

00:02:42: But over time, they just accumulate into this crippling technical debt.

00:02:46: you end up with MES manufacturing execution systems that don't talk to your broader network.

00:02:51: And you get these fragmented workflows where the operational technology running the machines is completely isolated from your IT handling, business data.

00:03:01: And Dinecar points out a really painful reality here.

00:03:04: Scaling operations doesn't magically dissolve this complexity.

00:03:07: No

00:03:07: not at all.

00:03:08: Skeel actually punishes disconnected systems.

00:03:11: if you want a genuinely smart factory adding more shiny technology on top of a fragmented base is Well, it's a recipe for disaster.

00:03:20: You have to reduce the structural chaos first.

00:03:23: You really do and you see The consequence of ignoring that foundation clearly when you look at certain regional markets like very long Actually crunched in numbers on this recently for the UK and Ireland.

00:03:33: Oh I saw that.

00:03:34: what did he find?

00:03:35: He found this glaring operational risk gap.

00:03:38: Companies are pouring vast amounts of investment into AI initiatives, but their underlying data is entirely unusable.

00:03:45: So

00:03:45: they're building on sand?

00:03:47: Exactly!

00:03:48: They are dealing with obsolete legacy assets and unsegmented networks.

00:03:52: But here's the ticking clock that really caught my attention.

00:03:54: Barry notes that thirty percent of engineering knowledge Is walking out door for retirement by twenty-thirty

00:04:00: Thirty percent.

00:04:00: Wow That is an irreplaceable loss of context.

00:04:03: Yeah, if that veteran domain knowledge leaves your building before it is systematically digitized you're entire production line becomes highly fragile which totally explains the stark divide in how different companies are prioritizing their tech spend right now.

00:04:17: Oh for sure!

00:04:18: yeah and Chervie made a really sharp observation after she attended a manufacturing forum at Switzerland.

00:04:24: She noted that while the large global corporates are busy holding panels on AI governance and you know, The long-term impact of humanoids.

00:04:32: Right all

00:04:32: the flashy stuff.

00:04:33: Yeah

00:04:34: Meanwhile many SMEs are literally just trying to consolidate their basic data backbone.

00:04:40: They were living in two completely different realities.

00:04:42: Well the SME is focusing on the data backbone.

00:04:44: have the right priority.

00:04:46: I mean You simply cannot leapfrog the foundation.

00:04:49: Let me try to put this into an analogy.

00:04:51: It's like trying to install a state-of-the art AI driven smart home system when the actual wiring inside The walls of the house is from nineteen fifty.

00:05:00: Yeah, that exactly it.

00:05:02: you can't just slap A beautiful glowing tablet on the wall and expect the lights To dim or the thermostat to adjust based On your daily routine?

00:05:09: That copper infrastructure simply cannot carry the digital signal.

00:05:12: You have to endure the pain Of ripping open the drywall And fixing the wiring first.

00:05:18: That analogy perfectly illustrates one of the most misunderstood concepts in the manufacturing industry right now, The Digital Twin.

00:05:25: Oh digital twins?

00:05:26: Yeah

00:05:26: a digital twin is essentially the ultimate smart tablet on the wall for a factory but it's totally useless without that foundational wiring.

00:05:34: we saw great debate about this sparked by Pavan Pusalluri online.

00:05:38: Oh right, he used a pretty brutal term for the current state of industry.

00:05:42: He really did!

00:05:43: He argued that ninety percent of what companies call digital twins are actually just ghosts.

00:05:48: Ghosts?

00:05:49: I mean...that is a heavy critique.

00:05:51: What differentiates a ghost from a functioning digital twin?

00:05:54: Well..a ghost is basically dead three-D model.

00:05:58: Usually a company gets this beautiful high-fidelity CAD model built during the commissioning phase of a plant.

00:06:05: And it looks incredible on a conference room screen, I'm sure?

00:06:07: Oh yeah!

00:06:08: It looks amazing for executives.

00:06:09: but the moment maintenance tech changes a physical valve on the line and doesn't update the software file... The model is dead—it just sits on a server gathering digital dust.

00:06:19: So its not a twin at all.

00:06:21: Right.

00:06:21: Buton argues that A real Digital Crin is NOT a dashboard or an expensive screensaver To deliver Actual ROI, like a twenty to thirty percent reduction in maintenance costs.

00:06:31: It requires three functional layers.

00:06:34: Okay what are they?

00:06:34: First it needs live continuous data streaming from sensors and control systems.

00:06:40: second it needs bi-directional control

00:06:43: Meaning the software doesn't just read the data... ...it can actually write parameters back to the physical controller Correct!

00:06:49: It can physically change state of machine.

00:06:51: And third it requires a Machine Learning layer.. ..It isn't just monitor showing you that a compressor is running

00:06:57: hot Right.

00:06:58: With that machine learning layer, the twin recognizes that say a two degree thermal shift last Thursday combined with specific vibration frequency matches bearing degradation curve from failure eight months ago.

00:07:11: and because it has bi-directional control system simulated what will happen to rest of line if it autonomously swaddles throughput down by five percent.

00:07:23: Having live bi-directional data means we no longer need a human sitting and staring at his screen waiting to hit approve on every single parameter change.

00:07:32: Exactly, We are stepping out of the dashboard era into the Era of Autonomous Operations

00:07:37: which perfectly brings us our second theme industrial AI and agentic operations.

00:07:43: because what happens when that data allows the system to actually take action.

00:07:48: Yeah, this is the transition from looking in the rear view mirror to actively steering the vehicle.

00:07:53: our scene jockey gave a phenomenal breakdown of how This looks and practice using SAP business AI

00:07:59: over two point four seven AM scenarios.

00:08:01: yes

00:08:02: Picture a factory floor at two forty seven AM.

00:08:04: A primary press line goes down and simultaneously, a critical supplier misses the delivery window.

00:08:10: Total nightmare scenario?

00:08:11: Absolute

00:08:11: nightmare.

00:08:12: but mechanism of the autonomous fix is incredible.

00:08:15: when the Press Line fails The AI doesn't just trigger an alarm and wake everyone up.

00:08:19: It actively queries the ERP system to check current inventory buffers.

00:08:24: it accesses the production schedule To see what orders are impacted instantly searches the approved vendor database, sends an automated purchase order to a secondary supplier and alters the machine feed rates on a secondary line to compensate.

00:08:40: All while the plant manager is fast asleep?

00:08:42: Exactly!

00:08:42: It turns the enterprise into self-optimizing, self healing organism.

00:08:46: That's just wild.

00:08:48: Yeah And Pernavut here are broke down the evolution of this capability really beautifully.

00:08:54: He maps it out in three distinct stages.

00:08:56: Okay, lay it up for us.

00:08:57: first you have IOT the Internet Of Things.

00:08:59: that is purely about connecting devices so You can see what's happening just getting The data out of machine.

00:09:03: day

00:09:03: six

00:09:04: right.

00:09:04: then you progress to iot industrial iot which adds context.

00:09:08: Yeah It is about optimizing operations and understanding why a variance Is happening okay?

00:09:12: But the final destination is AIoT the artificial intelligence of things.

00:09:16: That is where the system closes the loop predicts future outcomes and automatically initiates physical actions.

00:09:22: It transforms assets from passive data generators into active, intelligent decision makers.

00:09:28: but look... We need a reality check here.

00:09:31: Yeah, if you walk the floor of any manufacturing trade show today Literally every single vendor booth has a banner claiming they offer an AI

00:09:41: platform.

00:09:41: Oh absolutely The buzzword soup is real.

00:09:44: it is.

00:09:44: separating the genuine aiot Reality from the marketing hype is incredibly difficult.

00:09:50: Jacobo Lorette-Casal shared A brilliant straightforward BS test for evaluating these vendors

00:09:55: a bs test.

00:09:56: I love that.

00:09:57: What is it?

00:09:58: He said there's really only one question you ever need to ask them.

00:10:26: Exactly, genuine industrial AI changes a physical decision and delivers a measurable outcome on the

00:10:32: factory floor.

00:10:33: I have to admit though letting software write purchase orders and change physical machine feed rates while the engineering team sleeps it sounds genuinely terrifying.

00:10:41: It does!

00:10:42: Like how do manufacturers realistically build the organizational trust required To let an AI take the wheel without constantly fearing that will accidentally crash the entire production

00:10:52: line?

00:10:52: That is the crucial hurdle, right?

00:10:54: And the answer lies in rigorous validation.

00:10:57: You never just unleash an agentic AI onto the floor and cross your fingers...

00:11:01: I hope not!

00:11:02: Yeah, Kacper Chicheski from Dissalt Systems made a vital point regarding this.

00:11:06: The race to deploy AI isn't simply about acquiring algorithms It's how safely you validate them.

00:11:13: This where those live.

00:11:14: digital twins become essential.

00:11:16: Okay so they use the digital twin.

00:11:17: Yes

00:11:18: You use virtual twin experiences to simulate and test the AI's choices in a digital sandbox first.

00:11:24: So, the AI essentially practices on the Digital Ghost before it is ever allowed to touch physical iron?

00:11:29: That is the exact mechanism!

00:11:37: Kepper notes that validating these automated decisions in a virtual space can reduce unplanned downtime by up to ninety percent.

00:11:45: Nine percent?

00:11:45: Wow!

00:11:46: Yeah,

00:11:46: before a single line of code alters the physical machine and this requirement for safety and validation leads directly to massive shift into human workforce which Colin Masen highlighted.

00:11:57: Oh

00:11:57: right...the work force transition.

00:11:59: yeah

00:11:59: he says we are entering what he calls application phase of industrial AI And in this phase, the industry does not need prompt engineers.

00:12:07: Which is wild to think about considering.

00:12:09: prompt engineering was billed as the hottest most lucrative job title in the world just a year ago?

00:12:14: It was but on an uncarpeted high stakes plant floor.

00:12:18: A prompt engineer who only understands syntax Is basically useless.

00:12:22: So what do we then?

00:12:24: What the industry desperately needs, according to Massin are context engineers.

00:12:28: These are your domain veterans—your metallurgists, you're seasoned process engineers….

00:12:32: Your quality assurance leads Ah...

00:12:34: The people with actual physical knowledge

00:12:36: Exactly!

00:12:38: to safely supervise these autonomous multi-agent AI networks.

00:12:44: They possess the deep physical world intuition that a language model fundamentally lacks,

00:12:49: right?

00:12:50: They are the ones setting the operational guardrails defining the physical limits of the machinery so they can operate safely in.

00:12:58: That connects the dots perfectly.

00:13:00: It means taking that thirty percent of veteran knowledge that Barry Long warned us was retiring, and elevating those experts to become the supervisors for AI.

00:13:09: They became the brain trust programming systems boundaries... Exactly!

00:13:13: And this human element becomes even more critical as we look at where technology is heading next because the AI no longer confines with software dashboards or supply chain routing?

00:13:22: No it's not…

00:13:23: It breaks out a digital layer entirely and taking physical form on the shop floor.

00:13:29: Which brings us to our final theme, Physical AI & Robotics.

00:13:33: This is where it gets really sci-fi.

00:13:35: The boundary between digital intelligence And physical execution has officially collapsed.

00:13:41: Chris Munley posted a fantastic breakdown declaring that the traditional five year roadmap for industrial tech Is effectively dead.

00:13:48: Dead!

00:13:49: The future isn't theoretical slide deck It's production schedule happening today.

00:13:54: As proof, he points to autonomous mining trucks that are currently moving billions of tons.

00:14:03: We have computer vision systems mounted on agricultural tractors that can identify the difference between a crop and weed in milliseconds, cutting herbicide use by seventy-seven percent in real time.

00:14:14: Yeah This is reality of physical AI

00:14:17: And the leap from specialized tractors into general purpose humanoid robots Is happening way faster than most anticipated.

00:14:24: Oh absolutely.

00:14:25: Jeff Winter pointed out something that feels ripped From movie But it's grounded In current engineering capabilities.

00:14:31: He noted that humanoid robots are already technically near-human level in their cognitive abilities, meaning they're spatial reasoning.

00:14:38: They're path planning Their computer vision all highly advanced The mechanisms bottlenecking them right now or purely physical battery density and fine motor dexterity In their hands

00:14:49: which is why they aren't in our homes yet.

00:14:51: exactly

00:14:52: because of those limitations he argues there first.

00:14:54: real world deployments Aren't going to be in our chaotic living rooms folding laundry, they're going to be deployed in highly structured environments like factories warehouses and logistics hubs.

00:15:05: Well a factory is the perfect incubator for early humanoids.

00:15:09: think about it.

00:15:10: The lighting is consistent the floors are flat and predictable And the tasks while physically demanding Are highly repetitive?

00:15:18: The economics make sense there?

00:15:19: first

00:15:20: They really do.

00:15:21: You know when I first looked at the specs For Genesis AI's new humanoid robot its name Eno which Lucas M. Ziegler recently shared a breakdown of.

00:15:30: I was genuinely creeped out at first.

00:15:32: Oh,

00:15:32: you know

00:15:33: You know challenges all over preconceived assumptions about what a roll watch should look like primarily because it has no head and face

00:15:39: Right?

00:15:39: It deliberately completely sidesteps the uncanny valley effect.

00:15:44: I mean, why design a machine to mimic human facial expressions if it is just going to unnerve the human workers around?

00:15:51: Exactly.

00:15:51: That realization shifted my entire perspective.

00:15:54: It's a brilliant feature not a bug.

00:15:57: instead of a face Eno features hands with twenty two degrees Of freedom allowing you to handle genuinely difficult Unpredictable pliable materials

00:16:06: like bundling loose wires with sticky tape which is notoriously difficult for traditional robotics.

00:16:11: Yes

00:16:12: But the most fascinating mechanism on Eno is this optional screen mounted on its chest.

00:16:18: Before The Robot makes a physical movement, that screen visually displays it's reasoning, intended path and upcoming action.

00:16:25: It provides human workers with transparent window into machines logic to actively build psychological safety

00:16:31: Because trust is ultimate currency in shared factory floor.

00:16:34: If Human Workers don't trust machine movements integration fails

00:16:39: Totally.

00:16:40: It is exactly like being a passenger in a taxi, In a city you've never visited.

00:16:45: If the driver suddenly swerves down A dark narrow side alley Your brain perceives it as a threat Because You lack The underlying data for their decision.

00:16:54: Right!

00:16:54: You panic.

00:16:56: But if the Driver points to the GPS screen On the dashboard and says The main highway is completely closed due to an accident.

00:17:02: We're taking a detour to save ten minutes, your anxiety immediately drops.

00:17:07: Yeah you understand their predictive logic before they execute the turn.

00:17:11: Exactly That chest screen on the humanoid robot.

00:17:14: Is that GPS?

00:17:15: It provides the human worker with a machine's predictive reasoning data, establishing safety before motion occurs.

00:17:22: What a brilliant way to conceptualize it!

00:17:24: And this brings us critical perspective on the Human Workforce role in all of these.

00:17:29: Dr.

00:17:29: Karl Maas shared an insight from his time observing operations in Japan that completely reframes often fearful conversation around robotics.

00:17:37: How do they view?

00:17:38: In Japanese manufacturing culture, robotics is rarely viewed as a hostile tool for labor replacement.

00:17:43: Instead it's viewed as essential operational infrastructure designed to take on what they call three-k work

00:17:50: Three K Work?

00:17:51: What does that translate into?

00:17:52: in an operational context?

00:17:54: It translates two tasks that are dirty dangerous and demanding.

00:17:59: The fundamental philosophy is about elevating the human worker, not eliminating them.

00:18:04: You remove the fragile human body from the ergonomic strain of lifting heavy components or danger operating near extreme heat but the human retains operational judgment.

00:18:15: So robot executes the heavy lifting?

00:18:17: But the context engineer does complex problem solving?

00:18:20: Precisely!

00:18:21: But acquiring capability to actually build this synergistic floor.

00:18:24: Sean Dotson made a highly perceptive observation about how the biggest players are moving, specifically looking at GE Vernova.

00:18:32: Oh yeah they took very unconventional route.

00:18:35: They didn't go out and acquire some flashy humanoid startup with great PR team.

00:18:39: They absolutely didn't.

00:18:40: They quietly acquired a thirty-five person robotics integrator based in Montreal called Robotech.

00:18:46: Just thirty five people?

00:18:47: Yeah, and the strategic reasoning behind that is fascinating.

00:18:51: The actual bottleneck for massive manufacturers isn't getting access to AI models or buying robotic hardware...the bottleneck is execution

00:19:00: Making it all work together.

00:19:01: It's

00:19:02: the complex, messy work of writing the custom code to make a brand new robotic arm communicate seamlessly with a thirty-year old legacy database.

00:19:11: Every hour an external systems integrator works on that problem.

00:19:15: there is a substantial profit margin attached right.

00:19:17: G Evernova order book for grid and wind infrastructure is growing so rapidly they realized They just couldn't wait in line for external integrators?

00:19:31: competition in smart manufacturing right now is happening through the quiet acquisition of integration talent.

00:19:36: The people who actually know how to safely rip out the metaphorical nineteen fifties wiring and connect

00:19:45: Without a doubt.

00:19:46: And you know, if we pull back and connect all of these pieces from fixing structural chaos in establishing live digital twins to deploying agentic AI on physical robotics it leaves us looking at completely paradigm shifting reality.

00:20:02: It really does.

00:20:03: Daniel Keper shared an insight that should be required reading for every manufacturing leader.

00:20:07: He noted that fully AI-enabled production setups are currently demonstrating the potential to unlock up to sixty percent productivity gains.

00:20:15: Sixty percent?

00:20:16: Sixty PERCENT!

00:20:17: That isn't an incremental improvement, that fundamentally changes the unit economics of producing a physical good.

00:20:23: It changes the entire mathematical equation of global trade.

00:20:27: For the last four decades The primary strategic question for manufacturing executives was incredibly simple Where on the globe is it cheapest defined human labor?

00:20:35: And you just built your supply chain around that answer

00:20:37: Right, but with a sixty percent productivity game driven by automation and AI that legacy question becomes obsolete.

00:20:46: The new strategic question is while upgrading your current local facility into an AI-driven factory of the future allow you to completely outcompete.

00:20:56: So

00:20:58: industrial AI is no longer just a tool for optimizing local supply chain logistics.

00:21:04: It possesses the economic leverage to actively reverse decades of offshoring.

00:21:09: precisely AI might just be the catalyst that completely redraws the geopolitical map of manufacturing over.

00:21:28: and

00:21:38: don't forget to subscribe.