Best of LinkedIn: Digital Construction CW 36/ 37

00:00:00: provided by Thomas Allgaier and Frennus based on the most relevant LinkedIn post about digital construction in calendar weeks, thirty six and thirty seven.

00:00:09: Frenness is a B to B market research company that supports industrial automation and ICT companies with market intelligence across the construction industry to prioritize segments identify high value accounts and validate use cases.

00:00:23: so imagine handing this steering wheel of you know multi-million dollar bulldozer over two invisible driver.

00:00:31: That

00:00:31: sounds completely terrifying, right?

00:00:32: You just close your eyes and basically hope the foundation gets dug correctly.

00:00:36: but that dynamic is essentially what's happening in construction tech Right now.

00:00:40: today we're exploring The top digital construction trends across LinkedIn for you.

00:00:44: Yeah And specifically We really want to unpack this massive Just glaring contradiction In the industry right Now.

00:00:51: it's huge.

00:00:51: I mean venture capital an AI capabilities are accelerating at breakneck speed But actual on-site productivity is, well it's just stalling out.

00:01:00: It IS a fascinating paradox.

00:01:01: for sure we're building these godlike digital tools but they keep slamming into human bottlenecks.

00:01:08: or you know cash flow realities

00:01:10: and just fundamentally dirty data exactly!

00:01:13: And to really understand this... We have start with the tech that driving all of current hype because its has fundamentally shifted.

00:01:19: Yeah..we aren't talking about chatbots sitting in little side panel anymore You know, where you type a question and it spits back a paragraph.

00:01:25: Right?

00:01:26: The era of AI as just an author or consultant is ending.

00:01:30: It's really becoming an actor!

00:01:31: It IS And the mechanism making this possible Is called Model Context Protocol Or MCP.

00:01:37: Through these MCP integrations, AI is now acting As an autonomous agent

00:01:41: Executing workflows directly inside the software

00:01:44: Exactly Complex multi-step workflows.

00:01:46: This is massive leap.

00:01:48: Look at an experiment recently run by Atomir.

00:01:50: Atomirov.

00:01:51: Oh, the Revit test.

00:01:52: Yeah He wanted to test the absolute limits of what an agent could do in architectural modeling.

00:01:57: So he took GPT-Six Astra, hooked it up directly to Revit via MCP and dialed into its highest reasoning setting...

00:02:05: And what did you give it start

00:02:06: with?

00:02:06: That's the crazy part!

00:02:07: ...he gave a single reference image just one image of floor plan and elevation.

00:02:13: his prompt was simply recreate five story sixteen apartment building

00:02:17: Just from one static image no CAD file or underlying data structure.

00:02:21: It's just the one image he hit.

00:02:23: go in an exactly one hour, nineteen minutes and twenty nine seconds The AI autonomously produced Three hundred and twenty-seven walls.

00:02:32: Wow!

00:02:33: Yeah, in a hundred sixty two doors ninety eight windows And it successfully exported four complete drawing sheets.

00:02:40: That is insane

00:02:42: But the part that should really make you sit up and pay attention Is there was zero manual intervention during an entire hour or twenty minutes.

00:02:48: He literally just sat back to watch the cursor move.

00:02:50: That's terrifying Amazing at the same time, but let me push back on this a little bit.

00:02:55: Sure go ahead building A clean theoretical model from scratch in a controlled environment is impressive sure But that's not the reality of our industry.

00:03:03: no it's definitely messy out there.

00:03:04: Right, navigating an existing massively complex model like a federated model where you have structural MEP and architectural data all clashing together.

00:03:15: That requires a totally different type of spatial understanding.

00:03:18: It really does.

00:03:19: And actually that exact messy reality is what angels say.

00:03:23: tested GPT-Six Astro on?

00:03:25: Oh Really?

00:03:26: Yeah

00:03:27: He didn't ask it to build something new.

00:03:28: he dropped the AI into an incredibly dense, federated data center BIM model right inside a standard web browser.

00:03:36: Okay hold on stop there.

00:03:37: A large language model is fundamentally text predictor.

00:03:40: It doesn't have eyeballs.

00:03:41: So how was an AI flying through a three-D environment in a browser?

00:03:45: Right?

00:03:45: did you write a custom API for that?

00:03:47: Well thats the wild part...he didnt!

00:03:49: He gave it zero custom API access.

00:03:51: The AI was essentially reading the underlying document object model, the DOM of the web browser.

00:03:57: Oh so it just translated to spatial coordinates?

00:03:59: Exactly!

00:04:00: It translated coordinates and component hierarchies into a logic puzzle that could solve... ...it flew through the model entirely on its own inspected element properties and correctly identified cooling systems.

00:04:09: Wow.. Just by reading the DOM.

00:04:11: Yeah, it picked out the rear door heat exchangers and the dedicated air handling units all by itself.

00:04:17: It essentially did the job of a field engineer digging through a model to figure out what's inside is specific data hall And it didn't completely unaided

00:04:26: right.

00:04:26: so The AI can build a model from scratch and it can logically comprehend a dense federated model.

00:04:36: But this brings us right back to that invisible bulldozer problem you mentioned earlier, the

00:04:40: guardrails problem?

00:04:42: Yeah if AI is now executing work directly inside of software at lightning speed how on earth do keep from making catastrophic structural error?

00:04:52: You know, without you noticing until it's way too late.

00:04:54: You can't just let it run wild in a production environment and the major software providers are actually acutely aware of this liability.

00:05:00: Have they said anything about?

00:05:01: Yeah Harlan Brum from Autodesk pointed out that They've just launched new MCP servers for Revit specifically to build guardrails For these invisible drivers.

00:05:09: Oh That

00:05:09: makes sense!

00:05:10: They're introducing what they call trusted right And experimental tiers into the software.

00:05:14: So there layering human oversight back Into the loop

00:05:17: Precisely The AI Can still do all the heavy lifting.

00:05:21: You know, it can model the three hundred walls.

00:05:23: But the software holds those changes in a reviewable state.

00:05:27: A human architect or engineer has to verify what the AI actually did before those modifications are permanently committed to the master file.

00:05:35: And we're already seeing AI agents getting smart enough to realize when they need that Human Safety Net.

00:05:42: Oh really?

00:05:42: Like knowing your own limits

00:05:43: Exactly!

00:05:44: There was great example shared by OMARM R. He noted that Monta AI recently took over twelve hundred generic Revit walls and converted them into a highly detailed design development ready model in just a few minutes.

00:05:57: That's incredibly fast!

00:05:58: Right, it read the project specs proposed correct fire ratings and applied them.

00:06:03: but here is the crucial part.

00:06:05: when It ran into an ambiguous wall condition where the specs conflicted?

00:06:17: you know, from AI as a brute force generator to an intelligent partner.

00:06:22: It forces us to rethink what a building model actually is.

00:06:26: How so?

00:06:27: Well if AI can rapidly execute all these dimensional and property changes the BIM models in digital twins you manage Can't just be three D archives collecting dust on a server.

00:06:37: Oh I see right they have to evolve into active decision engines.

00:06:40: They need to help a project team figure out What To Change And why they should change it.

00:06:44: It

00:06:44: eased the shift from mere data hoarding to true decision intelligence.

00:06:49: We've historically been very good at hoarding data and construction, but um... Very bad at leveraging it in real time.

00:06:56: Michael Janssen actually provided a perfect illustration of this shift with TwinMasters Arch platform

00:07:01: A co-pilot tool.

00:07:02: Yeah!

00:07:02: It integrates directly with tools like Autodesk Revit & Forma To act as a cross discipline Decision Intelligence Co-Pilot

00:07:09: Which is huge because normally those disciplines are totally siloed.

00:07:12: Exactly, think about how this usually works.

00:07:15: you have one team looking at the cost a Totally separate team running carbon analysis and a third team making sure it's up to code And

00:07:21: coordinating.

00:07:21: all that takes weeks

00:07:23: right but Archie applies cognitive reasoning across All of those objectives cost carbon energy compliance all simultaneously.

00:07:33: It finds the optimized balance and pushes those changes directly back into the live model.

00:07:38: By the way, if you want to keep tracking how these digital tools are evolving week-to-week make sure you subscribe To this deep dive so you don't miss our future additions.

00:07:46: Definitely hit subscribe.

00:07:48: but There's a massive trap here when we talk about Digital twins and decision engines isn't there?

00:07:54: There really is.

00:07:55: Santosh Kumar Boda pointed out something completely counterintuitive.

00:07:59: What's that?

00:08:00: If you ask most firms what a digital twin should be they'll tell you it needs to be a perfect God-like replica of the building.

00:08:07: They want to connect every single model, every IoT sensor, Every HVAC return all into one ultimate dashboard.

00:08:14: The single pane of glass fallacy?

00:08:16: Exactly and Boda argues that despite our technical ability To build that most successful digital twins aren't doing it because connecting everything just creates noise.

00:08:24: So what are they doing instead?

00:08:25: ?The strongest implementations Are actually narrowing their scope drastically .They organize the tech around a single highly specific decision loop Like what?

00:08:35: Like inspect, decide repair or monitor predict intervene.

00:08:39: They focus on solving one operational bottleneck first rather than turning the twin into this massive aimless IT integration exercise.

00:08:50: And if we look at the psychology of how these twins are actually used to get even more complicated Schlumbo-Bernardsi highlighted some brilliant research across roughly eighteen hundred people.

00:08:58: Oh!

00:08:59: The behavioral economics research.

00:09:00: Yes

00:09:01: they found that right now When digital twins are built to model or predict human behavior inside a facility, they're completely failing.

00:09:09: Why are they failing?

00:09:10: Because their too rational?

00:09:11: Oh because they assume humans operate like perfect algorithms

00:09:15: Precisely If the twin is build to predict how workers will move through site Or how occupants we use of building's energy systems.

00:09:23: It actually needs to mimic human imperfection And

00:09:25: it need be programmed.

00:09:27: make mistakes

00:09:28: Exactly Needed forget things rely on lazy mental shortcuts, and make irrational errors.

00:09:34: If it just reasons like a flawless superhuman its predictions are useless in the real

00:09:39: world.".

00:09:40: That is a fascinating layer to add but you know let me bring this back to the boardroom for a second because there's a very real anxiety shadowing all of this capability.

00:09:48: if these models an AI agents or becoming this incredibly smart and autonomous got their junior staff?

00:09:56: The people who traditionally did all this manual coordination.

00:09:59: Right, are

00:10:00: they going to be out of a job?

00:10:01: it is absolutely the single most common fear regarding automation right now.

00:10:06: but if we look at history that isn't really how paradigm shifts play out in this industry.

00:10:11: you have an example yeah

00:10:12: Zoltan Zetoth shared and amazing historical parallel.

00:10:15: he looked back at the Eureka Tower project in Melbourne Australia from two

00:10:19: thousand four.

00:10:20: So the absolute infancy for BIM.

00:10:22: Exactly!

00:10:22: It was first skyscraper ever designed entirely in building information modeling and it is done by a thirty person firm using ArchiCAD's.

00:10:29: six point five.

00:10:30: Oh wow, they bet the farm on an unproven tech.

00:10:33: They did And you know what happened?

00:10:36: The technology didn't eliminate single job

00:10:39: so What does that do?

00:10:40: It dissolved the firms internal hierarchy Because the two-D documentation just became an automated byproduct, it removed the traditional wall between older highly experienced design principles and younger tech savvy staff.

00:10:54: Oh so they had to collaborate more closely?

00:10:56: Right!

00:10:56: They all have to sit around their screen working on building together.

00:11:00: The lesson there is that winning tools make your people vastly more capable not optional.

00:11:06: That is such a vital distinction, especially as we transition into the reality check of this entire conversation.

00:11:12: Which

00:11:12: is sorely needed

00:11:13: because We can talk about two thousand four and we Can Talk About GPT six building models in an hour today.

00:11:17: But industry data shows that digital adoption has hit roughly eighty five percent while overall on-site productivity Basically hasn't moved.

00:11:26: it's The multi billion dollar question.

00:11:27: if the tools are This good why Are we still so slow?

00:11:30: Because Fundamentally you cannot automate bad Data and you cannot bypass the humans who actually do the work on site.

00:11:37: Rob Rooster pointed out some stark realities about this...

00:11:40: What did he say?

00:11:41: He noted that AI adoption specifically has hit thirty-seven percent in the sector, but he warns that an AI is ultimately just a machine with the unique ability to be wrong with a straight face.

00:11:53: It has ultimate confidence even when it's hallucinating

00:11:56: Right!

00:11:56: And if you feed an AI late thin, fragmented field data the AI doesn't magically clean it up for you.

00:12:03: It just runs with it and suddenly your data is wrong at an exponential scale.

00:12:07: It

00:12:07: weaponizes bad data

00:12:09: Exactly!

00:12:10: Brewster noted that contractors who are actually winning this tech companies like AC Electric.

00:12:15: they focus heavily on structured clean data capture out in a field before applying single algorithms.

00:12:21: And this brings up a brilliant point by AJ Waters about why the industry keeps failing at this.

00:12:26: He argues that contact marketing has the entire paradigm backward.

00:12:30: How so?

00:12:30: They sell the tech first, process second and people last.

00:12:33: Which guarantees failure?

00:12:34: One hundred percent.

00:12:35: Water says you have to flip that rollout order entirely.

00:12:38: It has be people first because they control adoption Process Second To figure out Why You Do Something.

00:12:44: And Tech Last.

00:12:45: Four out of five digital transformations fail simply because firms reverse that

00:12:50: order.

00:12:51: Right, so let me throw an analogy at you.

00:12:53: Go for

00:12:53: it!

00:12:53: It is like buying a state-of the art water filter but attaching it to a cracked muddy pipe.

00:13:01: The tech itself isn't the core problem.

00:13:04: So what is the actual ROI we should be measuring when we implement these tools?

00:13:09: Because I clearly can't just be a spreadsheet metric.

00:13:12: right.

00:13:12: Kyler Morgan answered this perfectly.

00:13:14: He says the true ROI isn't just hours saved or cost reduced.

00:13:19: The

00:13:19: real tangible value is buying back time for the field teams.

00:13:23: if a superintendent can dictate A daily log instead of sitting in a hot trailer two hours typing it out You just gave them their afternoon

00:13:30: and they could actually walk to site.

00:13:31: Yes, yeah They have time to catch risks and missed details before they become a massive rework claim

00:13:37: using AI to eliminate administrative friction.

00:13:40: I love that.

00:13:41: and speaking of the human element Adam Cooper brought up what i think is The most fascinating point Of this entire discussion.

00:13:45: oh

00:13:45: about tribal knowledge.

00:13:47: yes He argues That, the biggest ai opportunity in construction isn't automated drafting.

00:13:52: it's capturing undocumented Tribal Knowledge.

00:13:55: That Is so true.

00:13:56: every firm has that one veteran.

00:13:58: everyone Calls when a schedule slips

00:14:00: right?

00:14:00: That person just intrinsically knows Which subcontractor to watch or What sequence will fail In the real world.

00:14:05: but what happens When they retire?

00:14:07: all that institutional knowledge Just leaves in their truck

00:14:10: And it crushes the bottom line.

00:14:11: But Cooper says AI can shadow those experts, they can document how they recognize risk and handle exceptions.

00:14:18: you turn that borrowed expertise into a permanent company asset before They leave.

00:14:22: transforming intuition into an asset is incredibly powerful.

00:14:26: but we have to zoom out here because even if a firm perfectly aligns its people cleans Its data and implements Ai flawlessly

00:14:34: still don't operate in a vacuum.

00:14:35: Exactly, they still have to survive the brutal physical and financial realities of the broader market.

00:14:41: The macro constraints

00:14:43: Yeah And right now there's an insane amount of capital betting on this tech.

00:14:47: AEC software startups raised six hundred sixteen million dollars in just the first half Of twenty twenty-six alone.

00:14:52: That is a staggering amount of money

00:14:55: It Is.

00:14:56: Diana K points out that a lot it has tied To this generational buildout for data centers Driven by AI boom itself.

00:15:03: But the physical world is violently pushing back against that ambition.

00:15:08: Rogan Permanentum revealed a massive bottleneck in that exact

00:15:12: sector.

00:15:12: What's The Bottleneck?

00:15:14: Well, while sixteen gigawatts of data centers are slated to come online this year if you look at the reality on-the-ground only about five gigawatths are actually under construction.

00:15:23: Wait an eleven gigawatt gap.

00:15:24: where did all those projects go?

00:15:25: They're paralyzed by local human resistance.

00:15:28: over two hundred and seventy major data center projects are facing a local opposition stalling out at the zoning and permitting stages.

00:15:35: Wow

00:15:36: so communities are pushing back.

00:15:37: exactly.

00:15:38: You can have the best AI generated schedule in infinite capital.

00:15:41: But If The Local Zoning Board Says No you're not putting a shovel in the dirt.

00:15:45: So local politics and permitting are the true modes?

00:15:48: Permitting

00:15:48: is massive, but Fabio Bronson highlighted an even bigger one that the software world completely ignores.

00:15:55: The real bottleneck in construction isn't productivity it's cash flow.

00:15:59: Oh!

00:16:00: The actual money moving around.

00:16:01: Yeah

00:16:02: think about it.

00:16:03: Subcontractors wait roughly fifty six days after submitting a pay application before they see single dollar.

00:16:09: Fifty-six days

00:16:10: Yes That is the slowest days sailed outstanding of any industry on earth.

00:16:16: and On top of that standard retainage, it's ten percent.

00:16:19: but a subs net margin Is only about six to eight percent.

00:16:23: so they're retaining routinely exceeds their entire profit margin.

00:16:26: Exactly as bronze and bluntly points out software can give you a pretty dashboard But he can't fix a fundamentally broken balance sheet.

00:16:34: So

00:16:34: let me get this straight We're pouring hundreds of millions in attack to build faster but the true modes are local politics, permitting and bank lending.

00:16:41: Pretty much!

00:16:42: So can policy actually dictate which software survives?

00:16:45: It absolutely can.

00:16:47: Zulker named Malik brought up a fascinating example from Ontario.

00:16:50: Premier Doug Ford publicly criticized public procurement specs that favored U.S bricks over Canadian ones.

00:16:56: Okay how does it relate to software?

00:16:58: Code doesn't have borders But procurement requirements do.

00:17:02: Malik points out.

00:17:03: this exact same protectionist debate extends When public projects default to US construction tech systems in their master specs, local Canadian Tech gets pushed out before the bidding even begins.

00:17:15: Oh I see it locks them out early on!

00:17:17: Right and because of this SmartMill just launched an initiative demanding that Canadian Construction Tech get fair consideration at the specification stage.

00:17:26: It really all comes crashing back to the real world, doesn't it?

00:17:29: You can optimize virtual building until its perfect but eventually hits municipal laws and actual dollars.

00:17:45: Thank you so much for listening.

00:17:46: We want to leave you with one final provocative thought to ponder.

00:17:49: today, we've talked extensively about AI perfectly optimizing designs and catching scheduling errors.

00:17:54: but what happens when that hyper-efficient AI hits the hard limits of the physical world?

00:17:59: When the model works perfectly... ...but the physical supply chain of concrete or steel simply runs dry.

00:18:05: Exactly!

00:18:06: It forces us to ask In the future, will the ultimate winners in construction be ones with best algorithms or those who control raw physical materials?

00:18:15: Don't forget to subscribe!