Best of LinkedIn: Digital Construction CW 34/ 35

00:00:00: Provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about digital construction in calendar weeks thirty-four and thirty five.

00:00:08: 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.

00:00:22: All right, so welcome to the deep dive everyone.

00:00:24: Yeah

00:00:25: Welcome.

00:00:25: and today we are jumping straight into a really massive paradox That's just kind of staring our whole industry in the face Right now.

00:00:32: it really is

00:00:33: because you know We're seeing all these reports indicating that construction tech adoption Is sitting at like an incredible eighty five percent

00:00:41: which is

00:00:42: huge?

00:00:43: It's massive!

00:00:44: We were pouring literal billions of dollars Into software into machine learning into robotics.

00:00:49: So I guess the question is, why has actual productivity growth on the job site just completely flatlined?

00:00:55: Exactly.

00:00:55: And that's exact questions we're answering today.

00:00:57: We are cutting through all of hype and noise to deliver top digital construction trends.

00:01:02: And we are extracting these directly from a curated stack insights shared by real industry professionals over last couple weeks.

00:01:10: Right because for those who you know, navigating the bleeding edge of smart build and manufacturing.

00:01:16: You already know that tech landscape is evolving way faster than most companies can even process.

00:01:21: Oh absolutely.

00:01:22: So our mission for this deep dive Is to move past that surface level hype.

00:01:27: We're looking at actual realities of AI adoption The sudden rise of these highly specialized digital agents And how all invisible data is finally colliding with heavy hardware out in mud

00:01:40: out in the mud, I love that.

00:01:42: So let's tackle that eighty-five percent adoption paradox first because we're buying all these tools but we are just not getting the output.

00:01:50: why is that?

00:01:51: Well

00:01:51: Daniel Wade actually posted a brilliant breakdown of this exact issue.

00:01:54: oh yeah what's his take?

00:01:56: he argues that pretty much everyone has having the wrong conversation about AI.

00:02:00: like.

00:02:00: the question isn't whether AI is going to change how we design and build...that's a given right!

00:02:04: The train has left the station

00:02:06: Exactly.

00:02:07: The real question is whether the construction industry is actually culturally ready for

00:02:11: it?

00:02:11: Culturally ready,

00:02:12: yeah He says AI readiness isn't a technology problem at all.

00:02:16: It is entirely a culture problem and its built on something he calls managed opacity.

00:02:22: Managed Opacity that's uh That'a great term!

00:02:25: It perfectly diagnoses the issue right.

00:02:27: Yeah Just think about how Construction Traditionally Operates.

00:02:31: Its an environment where data is intentionally siloed.

00:02:34: Oh totally.

00:02:35: I mean Subcontractors are always protecting their margins, right?

00:02:38: General contractors or hiding their schedule float from the owners.

00:02:42: Nobody ever wants to show their entire hands

00:02:44: exactly.

00:02:44: it's a defense mechanism and It's been built over decades and decades of adversarial contracting.

00:02:50: So let me put that in perspective.

00:02:51: then Deploying an advanced AI model overall all that siloed intentionally dirty data I mean That's basically like building a skyscraper on a swamp

00:03:01: Right!

00:03:02: Or, like it's putting a Ferrari engine inside of the nineteen seventies tractor.

00:03:07: Like... The engine itself is brilliant sure but the transmission just gonna shatter into one thousand pieces the second you hit gas.

00:03:15: It absolutely will!

00:03:16: So if we layer AI on top our existing workflows doesn't that give us an even faster version?

00:03:24: That gives you a highly efficient chaos engine.

00:03:28: Wow Yeah.

00:03:30: And this is actually why John Lehman recently laid out a very strict sequence of operations for leadership.

00:03:36: Okay, what's his sequence?

00:03:37: So he noted that contractors are constantly asking you know which AI tool should we buy?

00:03:42: but He says the only question That Actually Matters Is What Does Our Data Look Like When The Tool Finally Gets Here?

00:03:47: Ah Right You Have To Fix Your Data Architecture First Then Your Platforms then and Only Then Do You Actually Bring In The AI

00:03:58: architecture, I mean that sounds like boiling the ocean.

00:04:01: Where do you even begin when half of your project history is locked away in like unsearchable subcontractor PDFs?

00:04:08: You start by just stopping the bleeding...you establish a single operational hub where the ERP, business intelligence dashboards and all of collaboration tools flow seamlessly.

00:04:20: Seamlessly from estimate to budget right out into field?

00:04:23: Exactly with zero duplicate entry!

00:04:26: And crazy thing is that barrier getting there is rarely software itself.

00:04:31: Really what's it then?

00:04:32: Well RICS research backs this up.

00:04:35: they showed top obstacles to AI adoption are actually skilled people at forty six percent?

00:04:40: Wow almost half

00:04:42: Yeah, then system integration is at thirty-seven percent and data quality is that thirty percent.

00:04:47: So leadership has to physically own this transition.

00:04:50: they can't just delegate it to the IT department.

00:04:52: I hope for the best

00:04:53: right?

00:04:53: And speaking of leadership you really can't Just mandate this shift with a company wide email

00:04:58: Right?

00:04:58: no You definitely care because

00:05:00: AJ waters recently highlighted This major psychological trap that leaders are falling into right now.

00:05:06: Oh, the two camps thing?

00:05:07: Yeah exactly!

00:05:08: Leaders are giving this one generic rah-rah keynote to their entire employee base but that base is actually split in a totally different psychological camp.

00:05:18: Right you've got the hyped camp and the scared camp

00:05:21: Exactly And treating those two camps the same Just guarantees failure.

00:05:25: It really does.

00:05:26: because the hyped camp.

00:05:27: they don't need enthusiasm I mean, they're already out there trying to automate their entire workflow on their own.

00:05:32: Yeah They're the ones downloading random apps right

00:05:35: what?

00:05:35: They actually need from leadership or strict

00:05:38: boundaries.

00:05:38: boundaries got it.

00:05:39: yeah, they need to understand that AI isn't Magic it Isn't free and it absolutely cannot be used to siphon proprietary confidential company data into public AI models

00:05:51: which is a huge risk right now.

00:05:53: ah Okay, but then you have the scared camp.

00:05:55: And they're sitting in that exact same keynote secretly terrified at this.

00:05:59: new software is coming for their jobs

00:06:01: exactly and if you force a massive workflow overhaul on them They're gonna reject it purely out of self-preservation.

00:06:07: So he suggests to totally different approach for them.

00:06:09: You give him just one task That the AI does well Just one.

00:06:13: Yeah, just one.

00:06:14: you give them permission to try it?

00:06:15: To fail quickly and most importantly You let them see actual proof that their human expertise is still completely required to validate the output.

00:06:25: I love that.

00:06:26: It's a pacing problem, really!

00:06:28: The companies that succeed won't be the ones just by their most software licenses—they'll take time to read and manage.

00:06:35: the psychology of

00:06:47: does not care that your new app runs on a massive neural network.

00:06:51: They

00:06:51: do not tear at all!

00:06:52: They

00:06:52: only care if it saves them an hour of paperwork, at the end-of-the day.

00:06:55: Yeah Christopher Langeza offered a fantastic reality check on this.

00:06:59: What did he point out?

00:07:00: He pointed out most construction AI pilots fail because they never actually survive outside conference room.

00:07:07: Oh...that is so true Because for field adoption to happen The tool cannot require the superintendent stop what their doing just learn a new interface

00:07:16: Exactly.

00:07:16: The technology has to bend the field, not other way around which is why the real game changer out on this site right now...is voice and multimodal capture.

00:07:26: Yes!

00:07:27: Multimodals are huge

00:07:28: Because instead of fighting with drop down menus in an iPad while wearing heavy work gloves

00:07:32: Which is impossible

00:07:34: It IS So.

00:07:34: instead a superintendent can just talk point-a-camera walk through the site and capture their day.

00:07:41: The friction basically drops to zero.

00:07:43: And when the easiest path for the field becomes the smartest path of business, that is when you finally unlock real operational leverage.

00:07:51: Okay so let's play this out... Let us assume culture has primed right?

00:07:55: The Field interface entirely frictionless and data architecture beautifully clean.

00:08:01: A perfect world!

00:08:02: What actually happens if we move from AI readiness to actual execution on a live project?

00:08:09: Well, we are seeing the answer to that right now with the explosion of specialized AI

00:08:15: agents.

00:08:15: Yeah.

00:08:16: We have moved completely past those generic chatbots that just summarized text.

00:08:21: Right?

00:08:21: The early stuff...

00:08:22: Exactly!

00:08:23: These new agents are digital coworkers designed to execute really complex multi-step workflows.

00:08:30: Omri Stern recently highlighted a massive shift here.

00:08:33: Oh With Procore.

00:08:35: Procore just put a huge stake in the ground by releasing twenty specialized AI agents.

00:08:41: And to be clear for anyone listening, these are not gimmicks!

00:08:45: I mean he highlighted his specific case where Haskell used one of those agents... ...to process their submittal reviews

00:08:51: Which is notoriously painful

00:08:53: Super painful For context.

00:08:55: A submit review usually this highly tedious process Of checking massive stacks of equipment specs against design requirements And you have to do it before anything can even be purchased.

00:09:05: Right, It takes forever Yeah.

00:09:07: But Haskell used an agent To cut a process that normally take seven days down to just ten minutes.

00:09:11: Ten minutes?

00:09:11: I mean That fundamentally alters project economics.

00:09:14: They really does.

00:09:15: When you compress A week of administrative drag Into the time it takes to drink a cup Of coffee You free up your engineers engineer.

00:09:23: Right, and we are seeing these kinds of capabilities emerge across different platforms too.

00:09:28: Oh for sure.

00:09:29: Molly Stites Abbott shared some wild data on Constructables new AI agent.

00:09:34: oh I saw that.

00:09:35: yeah it essentially functions as a highly capable Autonomous personal assistant for project teams.

00:09:42: the specific examples she shared actually blew my mind.

00:09:45: like this Agent was put on a job that didn't even have a spec book.

00:09:49: Wow!

00:09:50: And it's successfully built a complete submittal log straight off the drawings, but here is The Real Kicker...

00:09:57: What's that?

00:09:57: It audited project contracts and found out that twenty-six out of forty four subcontracts were still sitting at ten percent retention.

00:10:06: Okay

00:10:06: let's pause right there to explain why.

00:10:08: this matters for listeners.

00:10:09: Good idea.

00:10:09: Retention is cashed when client holds back from contractor until they end their jobs just so work actually gets finished.

00:10:16: So if paperwork falls through the cracks and you forget to bill for the release of that retention, You are actively starving your subcontractors of their own cash flow.

00:10:24: And this AI agent just went in found twenty-six missing retention releases... ...and fixed all of

00:10:30: them!

00:10:30: That's

00:10:31: incredible!!

00:10:32: It also caught a card access system.. ..that was sitting in a bid scope but never actually drawn on architectural plans.

00:10:39: Oh wow

00:10:40: A classic scope gap

00:10:41: Yeah Which usually results into really nasty change.

00:10:44: order fight later

00:10:46: Definitely, and you know perhaps the most important takeaway from that post was how The tool is actually adopted

00:10:51: right.

00:10:51: They usage surged across the project teams with zero marketing in zero training.

00:10:56: Exactly

00:10:57: when a tool actively removes that level of friction from a Project managers day You don't need to change management seminar they just use it.

00:11:04: Hey, by the way if you are finding this deep dive into these rapidly evolving tools valuable make sure to hit the subscribe button so you don't miss our future updates.

00:11:13: The landscape is moving incredibly fast and we're constantly tracking those shifts.

00:11:18: for you

00:11:18: It really is moving fast And beyond.

00:11:20: these proprietary platform agents were also seeing incredibly specialized construction skills being built directly on top of foundational models.

00:11:28: Oh like the Claude stuff.

00:11:29: Yeah, Hamza Abdul-Jabbar recently catalogued sixteen specific skills for the Claude AI model that are tailored strictly to the built environment.

00:11:37: Okay so these aren't just generic prompts though?

00:11:39: Not at all.

00:11:39: He's talking about turning a five hundred page specification manual into a fully queryable database where every single material testing requirement is cited back to the exact page, or taking flat two D PDF drawings and automatically turning them in price bills of quantities complete with a two witness cross check between project schedule and physical plans.

00:12:02: It can even run health scorecards to grade underlying data quality.

00:12:07: Okay, but let me play devil's advocate here for a second

00:12:10: before I

00:12:10: have seen teams feed.

00:12:12: A massive densely cross-reference project document set into an AI and it completely falls apart.

00:12:18: Oh yeah the hallucinations

00:12:20: right?

00:12:20: It starts hallucinating missing obvious connections or giving you wildly inconsistent results.

00:12:26: so if these tools are so brilliant at parsing single drawing why do they just break projects scale?

00:12:33: That is the multi-million dollar question, and Guido Machiochi explains the mechanism behind it perfectly.

00:12:40: Okay what's his take?

00:12:41: Generalist AI simply fails at AEC scale because of how it processes information.

00:12:48: Standard AI pipelines flatten a project into just raw text in images... Right!

00:12:53: It just strips that all down.

00:12:54: Yeah

00:12:55: But In construction The meaning isn't just in the text its' in relationships.

00:13:00: It's how a specific pump in the spec manual relates to mechanical drawing, which then relates an installation activity on schedule.

00:13:08: So when you flatten documents, you sever all those vital links before AI even begins analyzing them!

00:13:14: I see.

00:13:14: So the AI essentially loses its memory of how a building actually goes together?

00:13:18: Precisely,

00:13:19: he pointed to research showing that on densely cross-reference project documents standard AI retrieval produced complete accurate answers only twenty five percent at the time.

00:13:29: Twenty

00:13:29: five percent!

00:13:30: That's terrible

00:13:31: it is.

00:13:31: but when they used a system that first built a structured graph of the material mapping all those relationships before asking the AI question.

00:13:38: The accuracy jumps Ninety-five percent

00:13:42: versus twenty five percent.

00:13:43: I mean that's the literal difference between a tool you can actually trust to build a hospital and A tool, which just creates a massive legal liability.

00:13:52: exactly Which is why?

00:13:53: The models themselves are not the competitive advantage.

00:13:56: right You could Just rent an AI model.

00:13:57: yeah

00:13:58: the true mode in our industry Right now Is structured ai ready data.

00:14:03: To prove the point, he noted a recent data acquisition that went for eight hundred and forty-five million dollars.

00:14:08: Eight hundred and fourty five million?

00:14:10: Yes!

00:14:11: The market is aggressively valuing structured connected data above literally everything

00:14:18: else.

00:14:19: But how do we actually build that structure so the AI understands our specific industry.

00:14:25: Mark Guza emphasizes that trustworthy construction AI requires deep context.

00:14:29: Right And it gets through a shared construction ontology.

00:14:32: Yes, the ontology is key.

00:14:34: It's essentially a master dictionary That forces different software platforms to speak The exact same

00:14:39: language.

00:14:40: So like think about a standard commercial door To the facility owner...that door Is long term asset.

00:14:45: But to the superintendent scheduling the job, it's an installation activity.

00:14:50: Right.

00:14:50: and To The Architect is a design element with specific fire rating.

00:14:54: And To The Estimator Is Just A Cost Code.

00:14:56: Exactly So If Your Software Systems Don't Have An Ontology Connecting All Those Different Identities.

00:15:02: The AI Gets Completely Confused

00:15:03: It Does!

00:15:04: He Arguees.

00:15:05: This Is Exactly Why Any Answer An AI Gives You Must Be Tied To An Evidence Layer

00:15:11: similar to the architecture used by platforms like Blade.

00:15:14: Yes,

00:15:14: project teams cannot just accept a confidently generated answer on blind faith.

00:15:18: no way.

00:15:19: they need to click a link and see the exact drawing The specific revision or the contract clause backing it up.

00:15:25: because of a confident AI Answer can still be wrong?

00:15:28: The evidence layer is the only mechanism that allows a human expert To instantly verify the truth

00:15:33: makes total sense.

00:15:34: so we've mapped out the digital brain right And we have clean data architecture, ontologies and these highly capable agents executing workflows.

00:15:42: Yeah the digital side is robust

00:15:44: but construction Is ultimately a physical act.

00:15:47: It's about moving dirt and pouring concrete Absolutely.

00:15:51: So how is all of this invisible digital intelligence finally manifesting in the physical world?

00:15:57: Of the job site?

00:15:57: Well,

00:15:58: this is where we see the hardware software convergence and honestly it is moving much faster than that industry realizes.

00:16:04: Oh for

00:16:04: sure Reagan Paraman anthem recently reported on this massive explosion of physical automation.

00:16:10: specifically looking at autonomous excavators

00:16:13: The funding numbers there are insane.

00:16:15: get this three startups bedrock robotics Paraferma and Gravis Robotics, they raised to combine five hundred seventy million dollars in just the last six months.

00:16:25: Over half a billion dollars pouring strictly into heavy machinery automation?

00:16:29: Yeah And this isn't happening at some controlled laboratory either.

00:16:32: Bedrock is already running fully driverless excavators meaning no human operator on the cab AT ALL!

00:16:39: On live US jobsite

00:16:41: Live job sites.

00:16:42: Yes

00:16:43: That includes a massive one point two million cubic yard site work project operating right now with Zachary construction,

00:16:49: which is just wild to think about.

00:16:51: And if you think about the underlying mechanism here this creates a massive data flywheel.

00:16:56: Oh absolutely

00:16:57: It's identical to how autonomous cars learn to drive on public roads.

00:17:01: Exactly!

00:17:02: Like every single unsupervised dig these excavators perform, Every rock they hit, Every slippage they correct... it all creates proprietary training data On site conditions and failure modes

00:17:14: And Earthworks is completely a game of schedule.

00:17:17: compression & fuel efficiency.

00:17:19: The contractor, with the best operators always wins the bid.

00:17:23: So as these fleets learn to self-orchestrate and coordinate across massive sites... ...the autonomous system will fundamentally perform better.

00:17:34: So whoever owns that training data deepens their moat?

00:17:37: Big time!

00:17:37: It potentially creates a winner.

00:17:39: takes all scenario for early adopters.

00:17:41: Okay but let's ground this for general contractor listening right now If you want start adopting robotics today.

00:17:47: You aren't going to go out and buy a fleet of fifty driverless excavators

00:17:51: tomorrow.

00:17:51: No, definitely not.

00:17:52: that would bankrupt your innovation budget instantly

00:17:55: Right?

00:17:55: so how do you actually start?

00:17:57: Charlie Elton offered some incredibly pragmatic advice on navigating this transition.

00:18:02: What's

00:18:02: his rule?

00:18:03: His core rule is Do Not Start With Five Robots!

00:18:07: You start by finding one highly repetitive manual task Like what like A specific overhead drilling application for MEP hangers?

00:18:16: Yeah

00:18:17: You deploy one single robot on a live project, you let it run and then generate a brutal clear-eyed report.

00:18:24: I like that.

00:18:25: prove at first!

00:18:25: Exactly...you measure the exact number of holes completed The amount of rework required, reduction in safety exposure for your workers And compare productivity directly against traditional manual method.

00:18:37: Because once you prove the application works with one machine, the internal conversation shifts.

00:18:41: The question is no longer does the robot work?

00:18:44: The question becomes how many robots do we need to hit our

00:18:46: schedule?".

00:18:47: Yeah that's when you hyperscale the deployment and go to five or ten machines executing flawlessly for sixteen or twenty-four hours a day.

00:18:52: That is the mechanism for unlocking measurable return on investment.

00:18:56: But wait!

00:18:57: To automate moving BERT Or have a robot drill a hole based upon digital model...the machine actually has know what physical environment looks like today.

00:19:06: Oh, right.

00:19:06: You can't automate construction if the robot is blind to reality.

00:19:10: so how are we digitizing the physical site quickly enough to actually feed these machines?

00:19:15: Well that brings us into the evolution of RealityCapture hardware.

00:19:19: Austin Lay recently published a field test for the new Leica RTC-Five Hundred scanner

00:19:24: and it perfectly illustrates how far this capability has advanced.

00:19:27: The mechanical specs on that scanner are staggering.

00:19:30: It delivers one point five millimeter three D point accuracy at ten meter range.

00:19:34: Wow It utilizes a seventy-two megapixel, six camera system to capture four hundred and thirty two megapixels of raw HDR data in about thirty seconds.

00:19:43: Thirty seconds?

00:19:44: So put that into perspective for you listening while your superintendent is like unscrewing the lid their thermos.

00:19:49: take a sip coffee.

00:19:51: this scanner has effectively digitized an entire complex mechanical room down to the millimeter.

00:19:57: That's unbelievable.

00:19:58: And crucially it introduces automatic self calibration during operation

00:20:02: Which was huge deal

00:20:03: right Huge.

00:20:05: In the past, taking the raw scan data and turning it into a usable three-D model —the whole scant BIM process—it required hours of agonizing manual alignment back at office.

00:20:17: Yes pitching all together Exactly!

00:20:19: But this hardware aligns those point clouds automatically eliminating that bottleneck.

00:20:24: It buys back critical schedule time

00:20:27: Before everyone listening runs out to buy fleet of these laser scanners.

00:20:31: we need talk Economics.

00:20:33: Always!

00:20:34: Joseph Trujillo posted a really necessary reality check on the true cost of bringing RealityCatcher in-house.

00:20:41: What's

00:20:41: The Real Cost?

00:20:42: Well, dropping eighty to one hundred and twenty thousand dollars On A Professional Terrestrial Scanner is literally just the down payment.

00:20:48: It's Just The Tip Of The Iceberg.

00:20:50: Really Yeah

00:20:51: You Have To Factor In The Expensive Registration And Processing Software Licenses.

00:20:55: you Need High End Computing Workstations Capable Of Rendering Massive Point Clowns Without Crashing

00:21:00: Which Aren't Cheap?

00:21:00: No.

00:21:01: And you need enterprise-grade data storage, plus you have to pay for the continuous training of specialized labor required even run.

00:21:09: So ultimately...you have to ruthlessly weigh your utilization rate!

00:21:13: If you only actually NEED a millimeter accurate scan ten or twenty times per year You're carrying six figure piece equipment that is just gathering dust in tool crib

00:21:22: Exactly.

00:21:23: You have to price out the total cost of ownership for that full capability and compare it directly against what you would spend just outsourcing The work To a dedicated reality capture partner.

00:21:34: It all comes back to the overarching theme Of the insights we've explored today.

00:21:39: Technology only provides a return when the operating model, data architecture and underlying economics actually make sense.

00:21:46: Perfectly

00:21:46: so!

00:21:46: Which leads me to a final somewhat provocative thought for you to mull over... And this circles back those highly capable AI agents we discussed earlier.

00:21:56: If these AI agents successfully optimize every individual firm's internal workflows making architect engineer general contractor incredibly efficient internally but contracts Coordination between those companies remain completely siloed and adversarial.

00:22:12: Oh, I see where you're going

00:22:13: right.

00:22:14: Do we just end up with thousands of highly efficient private playbooks?

00:22:18: While the actual delivery timeline Of The physical project stays exactly the same?

00:22:23: Wow

00:22:23: that is a great question to end on.

00:22:25: if you enjoyed this episode new episodes drop every two weeks.

00:22:28: also check out our other editions On smart manufacturing in industrial AI and connected tools and equipment.

00:22:34: Thank You so much for joining us on This deep dive.

00:22:36: thanks

00:22:36: for listening.

00:22:37: We'll catch you on the next one.

00:22:38: Don't forget to subscribe!