Best of LinkedIn: Smart Manufacturing & Industrial AI CW 43/ 44

00:00:00: Provided by Thomas Allgeier and Freeness, based on the most relevant LinkedIn posts about smart manufacturing and industrial AI in calendar weeks forty-three and forty four.

00:00:09: Freenes 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.

00:00:19: you can find more info In the description.

00:00:22: so today we're unpacking the absolute top trends in smart manufacturing and industrial AI that are currently driving conversations across LinkedIn.

00:00:30: Right, it's a fast-paced no fluff deep dive today because well the sources show a massive shift.

00:00:35: I mean Industrial AI is officially moving out of those isolated tiny pilots into you know coordinated enterprise white systems.

00:00:42: It really is.

00:00:43: And to understand we need look at first big theme which how the way engineers actually build automation is fundamentally changing its all thanks to Agentec AI.

00:00:52: Oh, definitely.

00:00:53: I want to jump right into the example Michael Gromis shared.

00:00:56: he gave The AI model Claude an electrical schematic and a process description And Claude didn't just give him generic advice

00:01:03: Right it actually configured the hardware in TIA portal.

00:01:06: Yeah It wrote the PLC program and even tested against A digital twin

00:01:11: Which is wild.

00:01:13: So you have an AI Interpreting physical constraints from basic schematic and translating that Into executable machine logic.

00:01:22: But wait, so are automation engineers essentially moving from like being authors who write the code to being editors?

00:01:29: Who manage the context?

00:01:31: that is exactly The transition taking place.

00:01:33: I mean Sebastian Christcher brought this up too.

00:01:35: He noted that Gen AI can now build pharma operator screens in thirty minutes.

00:01:41: Wow Thirty minutes.

00:01:42: yeah That used to take weeks of repetitive work

00:01:45: Exactly

00:01:46: and so that shifts the engineering effort entirely toward ensuring clean data models defining clear roles and setting up traceability for validation.

00:01:55: Because you can't just let the AI run wild in pharma?

00:01:57: No, absolutely not!

00:01:59: The human becomes the final guarantor of quality

00:02:02: And it's interesting to see how vendors are actually adapting to this editor role.

00:02:06: Rahul Patel and Subhajit Bodak talked about Rockwells Factory Talk Conductor.

00:02:11: They're positioning it to orchestrate AI agents across the whole engineering life cycle

00:02:15: Right like connecting Studio.

00:02:17: five thousand Logix designer and emulate three D Yeah,

00:02:20: and they're using Nemotron so manufacturers can run open models securely inside their own environments.

00:02:26: Which is huge for security!

00:02:28: You aren't shipping proprietary PLC logic off to a public cloud.

00:02:32: Tobias Gerhardt in Hans-Michael Kraus discussed a really similar approach with ControlX's Agent IX.

00:02:38: Oh right where it embeds AI directly into the ControlX tools?

00:02:41: And generates IEC six eleven thirty one dash three PLC code.

00:02:46: But the brilliant part is it specifically avoids vendor lock-in by letting user choose their own language model.

00:02:52: Okay, but if AI's writing smarter code locally at machine level that intelligence hits a massive bottleneck.

00:02:58: If the factory brain, MES isn't connected to rest of business

00:03:02: No for sure.

00:03:03: Which brings us to the manufacturing execution system or MES and enterprise integration?

00:03:09: Chacobo Euleric Casal shared recent report surveying one thousand five hundred sixty manufacturers

00:03:14: And numbers were kind Jarring, right?

00:03:16: Very.

00:03:17: Ninety-three percent have an MES but only twenty three percent Have fully integrated it across the enterprise.

00:03:23: So when you ask what's happening across all plants someone just like opens Excel.

00:03:28: pretty much It's a decision problem not a technology problem.

00:03:32: By the way real quick if you're finding this breakdown valuable make sure You subscribe to the show so you don't miss our future deep dives.

00:03:38: But yeah why is that bottleneck so bad?

00:03:41: well William D explained it perfectly.

00:03:43: He said manufacturing often acts like six different planets ERP, MES supply chain management They're all treated as competing tech rather than six connected perspectives on the same business

00:03:55: which is wild.

00:03:56: But vendors are definitely forcing this integration now.

00:03:58: Cullen Massin highlighted that SAP digital manufacturing a sunsetting its legacy MII and ME systems by twenty

00:04:05: thirty.

00:04:05: Right.

00:04:05: There's shedding that custom code debt to capture of The Greenfield running core S for H&A

00:04:11: burning the boats basically.

00:04:12: And SAP is also deploying model context protocol servers on the edge, so local machine vision models can query plant states without cloud latency.

00:04:21: Because a two hundred millisecond round trip to the Cloud is a deal breaker on a fast conveyor belt.

00:04:26: Met and Kaplan actually echoed this shift too noting Siemens Ops Center.

00:04:30: MES is being promoted from A system of record To an AI powered System Of Intelligence.

00:04:36: But wait, John Eunice brought up a pretty harsh operational reality check here.

00:04:40: You can't have enterprise intelligence without standardized data?

00:04:44: Oh absolutely!

00:04:45: He pointed out that if you had the exact same KPI across five plants...you often get five different answers.

00:04:51: Right, because one plant measures downtime differently than another and tools like litmus data catalog are emerging purely to fix these shared metric definitions before they just ruin enterprise reporting.

00:05:02: Exactly!

00:05:02: The AI just inherits the chaos if the data isn't clean.

00:05:06: Okay so once you have standardized data in an integrated Enterprise that intelligence finally gets pushed down into a physical factory floor which brings us to physical AI.

00:05:16: Yeah...and the use cases here are incredible.

00:05:18: Cedric Nike Rev, Lebedarian and Karthik Amirtherajan all discussed the Siemens Air Long & Electronics Factory.

00:05:26: Where they're using NVIDIA Omniverse right?

00:05:28: Yep

00:05:28: to generate synthetic training data for robot arms so that robots can dynamically correct unexpected packaging placements.

00:05:35: They are moving from pre-programmed rules To actual adaptable physical AI

00:05:41: And a sheer scale of infrastructure powering.

00:05:43: this is just mind-bending.

00:05:45: Vladimir Troy and Ali Hadari talked about Foxconn's NVIDIA powered AI factory,

00:05:51: right?

00:05:51: They are achieving two point four eight times faster training on a hundred twenty billion perimeter models

00:05:57: processing at trillion tokens in twelve days And they cut production scheduling hours by twenty five percent.

00:06:03: the speed is what gets me.

00:06:05: Alvin Clark ran this experiment highlighting The Insane Speed of Vision AI specifically.

00:06:10: Oh I

00:06:11: saw that He gave a vision language model, a quality checklist.

00:06:15: Yeah and it answered eight questions in just eighty four point six milliseconds

00:06:19: which

00:06:20: remarkably was only eight milliseconds more than they took to answer single question.

00:06:25: Wait really?

00:06:27: So adding seven complex rules barely touched the latency.

00:06:30: Exactly The heavy lifting is processing image.

00:06:32: initially Rob Kemp noted that NVIDIA VSS blueprint can now deploy visual AI agent under thirty minutes.

00:06:41: I mean, that's great for digital environments but i have to push back a bit on the hype because deploying robots on a messy real world shop floor is completely different story.

00:06:50: Totally fair and Stefan Winkler made a great point about that with the Control-X humanoid robot Roxy

00:06:56: The one that specifically skips having legs right?

00:06:58: Yeah

00:06:59: it uses a mature mobile platform instead Because he needs to work in a real factory today.

00:07:04: Now just look cool doing back flips into viral

00:07:06: video Exactly.

00:07:08: Craig McDonald also highlighted a really practical deployment with the fashion and doctor logistic solution.

00:07:14: They use an AI robot trainer to help vision-based robots adapt to constantly changing fashion packaging

00:07:19: because clothing doesn't hold rigid shape, it's a brilliant application.

00:07:23: but you know putting AI in charge of physical assets and critical processes brings us to The biggest hurdle standing in the way of scale

00:07:31: right security compliance And keeping that human in the loop.

00:07:34: yeah the cybersecurity vulnerabilities alarming.

00:07:37: Zachariah Sankara pointed out that default OT protocols like Modbus TCP, Ethernet IP, IEC- one oh four.

00:07:46: they completely lack basic authentication and encryption.

00:07:49: And Patrick Ho really backed up the severity of this.

00:07:53: he noted that forty six percent of manufacturers were hit by cyber incidents in the past twelve months.

00:07:58: it's a massive blind spot.

00:08:00: then there is the compliance nightmare removing humans from process.

00:08:05: Niraj Kishwani shared a chilling quote from an FDA warning letter regarding the drug manufacturer.

00:08:10: What did it say?

00:08:11: It literally said, The AI agent you used never told you that was required.

00:08:16: Oh wow So the AI wrote production records but just completely skipped process validation.

00:08:22: Exactly

00:08:23: Because general purpose model doesn't inherently know what companies.

00:08:27: specific internal quality systems

00:08:30: brilliant speed reading intern who just confidently ignores federal law because nobody explicitly told them about it.

00:08:37: That is the perfect analogy, and this leads right into Binwang's observation on scaling human expertise.

00:08:43: Oh with ad nox AI enabled real-time operations center

00:08:46: Right It allows their engineers to support two or three times more offshore drilling rigs.

00:08:52: But Wang warrants that greater coverage doesn't automatically mean better oversight

00:08:56: Because of the cognitive load

00:08:57: Exactly.

00:08:58: Can an engineer supporting three times as many assets still properly investigate ambiguous conflicting signals under pressure?

00:09:06: Yeah, that's a huge question!

00:09:07: You can't just stretch human attention infinitely without consequences.

00:09:11: Well if you enjoyed this episode new episodes drop every two weeks.

00:09:14: also check out our other editions on digital construction and connected tools & equipment.

00:09:19: Thank you so much for listening And remember to subscribe.

00:09:22: I Just want to leave you with one final provocative thought to mull over, building on a point from amidst Sharma in the sources.

00:09:29: As machines factories and enterprise systems all become highly intelligent an autonomous every manufacturer will have access to the exact same AI capabilities right.

00:09:39: so In that world how you protect your company's unique alpha?

00:09:43: That specific operational knowledge And those trade-offs they give you of your competitive edge So that AI actually amplifies it rather than just commoditizing?