Best of LinkedIn: Digital Construction CW 44/ 45

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

00:00:09: Freenes 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: so imagine deploying this state of the art AI that confidently tells you your project is perfectly healthy.

00:00:28: oh wow

00:00:28: yeah.

00:00:29: But meanwhile, it entirely misses the fact that your actual finish date just slipped by two whole weeks.

00:00:35: Welcome to The Deep Dive!

00:00:36: By the way our mission today is too well cut straight through the tech hype

00:00:40: right which we really need to do

00:00:42: exactly.

00:00:43: We're unpacking the top digital construction trends across LinkedIn and we are focusing specifically on what Is actually working On job sites you know And in offices Right now

00:00:52: And that is such a critical distinction to make today.

00:00:54: I'm welcome everyone, the industry.

00:00:57: it's rapidly moving from just experimenting with these isolated pilots To treating AI as core infrastructure.

00:01:04: Yeah But the constraint you know It no longer tech itself.

00:01:07: The real bottleneck Is data, context and ultimately people managing at all.

00:01:12: Okay let unpack this Because we have like A massive reality check happening right now around AI adoption.

00:01:18: Everyone says they're adopting AI.

00:01:20: Sure yeah big buzz word.

00:01:21: Right but The promised ROI is violently colliding with the reality of just how messy construction data

00:01:28: truly.

00:01:29: Is it's a massive collision?

00:01:31: Just look at the data that was debated recently by Max Neverman and Renata Pinheiro, they were looking at Bluebeam's twenty-twenty seven outlook.

00:01:38: Oh I saw That one.

00:01:39: yeah.

00:01:40: so right now about twenty eight percent Of firms are running AI in live production And sixty four percent of those firms Are predicting better cost predictability which you know sounds amazing

00:01:51: On the surface, yeah.

00:01:52: Exactly but here's the catch only seventeen percent are reporting any actual reduction in rework on the site.

00:02:01: wait really?

00:02:02: Only seventeen percent.

00:02:03: Yeah

00:02:04: and while forty percent of firms that managed to scale beyond small pilots claim they saved over five hundred thousand dollars those savings aren't necessarily translating into proven margins at the bottom line.

00:02:16: hmm it begs a huge question.

00:02:18: you know Better software forecasts do not automatically mean cheaper, more profitable buildings.

00:02:24: Right and that ambiguity is well it's causing a lot of friction.

00:02:27: if you look at the recent Suffolk in MIT study Morgan Drives was analyzing Oh

00:02:31: yeah The cost savings one

00:02:32: Exactly, it modeled out the potential of these tools and found AI could realistically reduce project costs by like seventeen to twenty percent.

00:02:40: Which is incredible for this industry!

00:02:41: It is.

00:02:42: but Morgan asked The Quiet Part Out Loud are those massive savings actually going to reach the Project owner or are the software vendors just gonna you know absorb all those efficiency gains by jacking up their subscription prices?

00:02:54: Good point Because historically more productivity at the desk doesn't always equal cheaper steel on the ground.

00:03:00: No it doesn't.

00:03:02: And this sheer ambiguity is exactly why our baseline understanding of adoption is so skewed right now.

00:03:08: M Reza Hussaini pointed us out with a white paper explaining the Australian AI adoption figures.

00:03:13: Oh, The one with huge gap in numbers?

00:03:15: Yeah Exactly If you look at reports Adoption confusingly ranges from Amiga's six percent all the way up to forty four percent.

00:03:23: How do we even get that wide?

00:03:25: Well it turns out.

00:03:26: the six-percent metric captures the vast subcontracting pyramid You know, the thousands of small mom and pop businesses doing the actual physical labor.

00:03:34: Oh right.

00:03:35: And then the forty four percent metric.

00:03:36: well that reflects Well resourced corporate headquarters running analytics.

00:03:41: So we are mixing vastly different business models culling them all construction and completely muddying The water.

00:03:47: when the water is muddy the AI gets incredibly confused.

00:03:50: like Chandra, Sir Brumonian shared brilliant, slightly terrifying example from SVUC-twenty six.

00:03:56: Oh I love this example!

00:03:58: Right so a team fed a project schedule to an AI tool and the AI confidently declared Project Health is improving

00:04:06: Classic

00:04:06: And technically by the AIs metrics The schedules quality score had gone up because tasks were being checked off But the overall finish date had slipped my fourteen days.

00:04:17: Yep...the

00:04:17: AI wasn't maliciously wrong it was just underfed.

00:04:21: Exactly

00:04:22: It lacked the contextual architecture of a real project.

00:04:24: A schedule isn't just a list of dates, I mean it's a web of dependencies.

00:04:28: You have concepts like float which is uh...the buffer time before delay impacts final

00:04:33: delivery.

00:04:34: The AI saw tasks being completed and scored their health higher.

00:04:38: It was entirely oblivious to that fact that critical path was severely compromised.

00:04:42: Deploying an AI on bad data is buying Ferrari but only having dirt road drive-on.

00:04:47: All this horsepower absolutely no traction.

00:04:49: That is the crux of the data problem right there.

00:04:52: An AI model doesn't inherently understand construction physics, it just reads whatever raw file you hand it and if you handle raw unversioned or siloed data You get confident fiction.

00:05:04: And this lack of context creates massive liability.

00:05:07: Josh Levy pointed out some Arcadis data showing that while fifty seven percent of firms use AI in project management The top causes of legal disputes haven't changed at all.

00:05:17: Wow

00:05:18: He calls it the carry forward

00:05:19: gap.

00:05:20: The carry-forward gap?

00:05:21: Okay, let me make sure I'm wrapping my head around this.

00:05:24: that's when like a senior superintendent spots a massive risk on site but because they just jot it down in their personal notebook

00:05:30: or Just stays in there had

00:05:32: yeah right It never enters the firm central database.

00:05:35: precisely the risk is found by a person But that knowledge effectively dies with them Or leaves when they retire.

00:05:43: the AI can't learn from something That was never digitized.

00:05:46: And even when things are digitized, I mean the software itself can be a nightmare.

00:05:50: AJ Waters made a great point about how construction platforms are often built from a bunch of unintegrated software acquisitions.

00:05:56: Oh Franken systems

00:05:58: Exactly!

00:05:58: A tech giant buys five different tools slaps one logo on them and calls it an ecosystem but they don't actually talk to each other.

00:06:06: It forces human beings act as API like manually copying and pasting data which actually manufactures administrative cost.

00:06:14: We see this exact breakdown in authoring tools as well.

00:06:18: Devon Shea, Meadow Elden and Verage Votetel were deep into discussion about how AI operates within highly contextual software like Revit

00:06:25: The three-D modeling standard right?

00:06:27: Yeah

00:06:28: And Revit isn't just a drawing tool it's essentially a parametric map of behaviors.

00:06:33: If you move a wall on Revit the software automatically knows that roof slopes need to adjust windows have to shift and HVAC ducts might need to reroute

00:06:41: based on rules.

00:06:42: a BIM professional just carries in their head.

00:06:44: Exactly!

00:06:44: Wait, let me stop you there... So if an AI agent without that deep spatial intuition makes a simple edit to a floor plan it might look visually correct on the surface but could break like fifty hidden dependencies in the background?

00:06:57: Exactly The AI may move adducts to resolve clash But doing so violates mechanical engineering constraint.

00:07:04: It doesn't understand.

00:07:06: Let's push back.

00:07:07: If AI just shifts an expert's time from doing the manual drafting work to meticulously checking the AIs' work, make sure it didn't break the physics of building.

00:07:17: Are we actually saving any time or are we shifting cognitive load from creation and endless verification?

00:07:23: What is fascinating here?

00:07:24: that cost doesn't disappear but changes form.

00:07:28: if promise savings depend entirely on a senior expert spending hours verifying their output then real bottleneck isn't software subscriptions

00:07:36: the data.

00:07:37: Exactly, it's structured data and workflow permissions.

00:07:41: you can't fix the data without fundamentally addressing people who create manage and verify

00:07:47: which means we really need to look at how this is reshaping work force itself because contrary to panic from a year ago.

00:07:54: AI isn't causing mass layoffs in construction.

00:07:56: no not all.

00:07:57: It actually restructuring teams taking direct aim at administrative bloat.

00:08:02: John Lieman highlighted BRG survey of construction executives that perfectly illustrates this.

00:08:07: Forty-six percent of these execs have actively changed their team structures because of AI, and thirty eight percent have invested heavily in retraining their staff.

00:08:16: but the headcount stays the same right.

00:08:18: that is the vital statistic.

00:08:20: only eleven percent has actually reduced their head count.

00:08:23: The people aren't leaving the industry.

00:08:25: Their daily roles are just mutating.

00:08:27: And Andy Pritchard had a fascinating observation on why that mutation is so necessary.

00:08:31: He pointed out, back in nineteen ninety-one the ratio of builders on tools to office staff was three point two to one.

00:08:40: But by twenty twenty-one That ratio dropped from one point eight.

00:08:45: We didn't suddenly stop needing bricklayers and carpenters.

00:08:48: we just added massive layers middle management to report check document legally protect builder.

00:08:55: And that's exactly where the AI comes in.

00:08:57: Right, AIs true target is that exact paperwork layer things like RFI.

00:09:02: you know they endless requests for information and massive compliance packs.

00:09:07: The goal was to automate that administration to push that ratio back up

00:09:11: To put a framework to that.

00:09:12: Kenneth Wu offered a highly practical way to view this transition.

00:09:16: He sources daily BIM tasks into three distinct levels based on automation.

00:09:21: How does he break it down?

00:09:22: So level one involves clear rigid rules where the AI just executes.

00:09:27: The task flawlessly.

00:09:29: Level two is option nearing, the AI generates five different layout options But he makes the final human call.

00:09:36: and level three

00:09:37: level Three is fully autonomous decision-making which detect simply isn't that yet.

00:09:43: but He noted something profound sixty percent of his tasks in those first to levels are already gone handled entirely by machines

00:09:51: Weaving just judgment.

00:09:52: Exactly what is left for him to do, his pure judgement and strategy.

00:09:56: And we are seeing that human judgment being aggressively applied To managing digital team members.

00:10:01: now Steve Bowie predicts That very soon every estimator & project manager Will act less like an individual contributor And more Like a manager overseeing A small team of AI agents.

00:10:11: It's happening fast

00:10:12: it really Is.

00:10:14: Matt Perot recently launched SUPA Which handles construction admin natively via text message in WhatsApp.

00:10:20: Oh nice.

00:10:20: And Sarah Buckner's trunk tools brings AI directly into Outlook and has an iPhone app to instantly draft those RFIs right from the site.

00:10:29: Chris Copeland also noted how Valve uses a plain English AI agent in pre-construction, to rapidly cross check thousands of written specs against drawings

00:10:38: So they find discrepancies before a shovel ever hits the dirt?

00:10:41: Exactly!

00:10:42: If you look at the mechanics these tools aren't asking workers to log onto some new clunky dashboard.

00:10:48: They are intercepting the communication where it already happens naturally.

00:10:52: In WhatsApp and emails?

00:10:54: Right, And they're performing administrative translation in the background!

00:10:58: They are functioning as actual team members rather than just passive software interfaces.

00:11:02: So the PM is really becoming an orchestra conductor managing these specialized ADI agents instead of spending six hours a day drafting e-mails.

00:11:11: Oh by the way if you want to keep tracking how those roles and technologies evolve make sure you hit subscribe so catch our future deep dives.

00:11:18: Definitely subscribe.

00:11:20: But

00:11:20: this whole transition from doing the work to managing in the agents makes me wonder, how does this affect very top of the funnel?

00:11:28: Does it supply to architects and engineers designing these buildings in first place?

00:11:33: It absolutely does!

00:11:34: And its driving a massive disruption on the authoring tools themselves.

00:11:38: We're watching a fundamental platform shift away from manual point-and-click drawing toward AI generated concepts an open source alternatives

00:11:47: alternatives that challenge the legacy monopolies.

00:11:50: Like, The massive elephant in the room here is Autodesk right?

00:11:54: Oh big time!

00:11:55: Day on Ninov Andrea Nagnost and Peter Russo shared some intense insights recently when Autodesks CEO confirmed their new industry cloud platform Forma will eventually replace Revit entirely.

00:12:08: A huge pivot.

00:12:08: Yeah they're pivoting hard from a desktop drawing app to connected cloud-based design to operate platform because decisions made in early design carry their context all the way to physical field.

00:12:20: Though, as Conrad K. Silban and Lee Snyder pointed out when you start dealing with cloud model cutoffs and recovery tools transitioning decades of proprietary legacy data into a new cloud system

00:12:32: it's gonna be a nightmare

00:12:33: A monumental heavy lift for lot of architectural firms.

00:12:37: yes

00:12:37: And that friction with proprietary data is exactly what fueling open source rebellion?

00:12:42: Absolutely.

00:12:43: Irving Gersendi's and Regis Benoit Brycen highlighted a surge in open-source alternatives like FreeCAD, Bonsai... ...and newly launched projects like BIMcraft & Kriabim.

00:12:54: Open source for the win!

00:12:56: They're being engineered from ground up so that AI agents can drive them programmatically.

00:13:01: However they noted critical mechanical limitation

00:13:04: Which is?

00:13:05: None of these opensource tools can reliably open native RVT files Revit's proprietary file format without data loss.

00:13:12: yet.

00:13:13: So things like real-time collaboration and work sharing, the deep moats protecting the incumbents are still there.

00:13:19: But the actual act of drafting is undeniably changing regardless of the software.

00:13:24: Brendan A shared a tool called Synapse.

00:13:26: two point oh And Dennis Panuta highlighted snap true to agent these aren't tools where you draw lines

00:13:33: right.

00:13:33: they're prompt based exactly.

00:13:35: They are turning plain text prompts in rough napkin sketches directly into complete three D buildings and usable construction drawings.

00:13:43: And Victor Perez showed how CREA allows architects to rapidly iterate on different visual renders without losing the fixed underlying design specifications.

00:13:51: Meaning, The AI respects the structural constraints?

00:13:53: Yeah!

00:13:54: It can generate a beautiful render of a lobby but it won't hallucinate doors that lead into solid brick walls...

00:14:00: ...and we're seeing this agentic behavior being hacked in traditional tools as well.

00:14:04: Evan Cyrilli showed an AI agent who literally built a custom egress checking app inside Revit based on a single text request.

00:14:13: To calculate fire escape routes?

00:14:15: Automatically, yes!

00:14:17: And Marco Intolasano demonstrated a custom server that allows an AI to drive the highly complex structural engineering software, Sophistic all away from basic hand sketch... ...to fully calculated structural safety report.

00:14:31: Okay here's where it gets really interesting for me.

00:14:34: Alistair Lewis made this provocative argument online that architects should just start sharing their closely guarded Revit templates and standard details with their direct competitors.

00:14:43: That's bold!

00:14:44: Right, his logic is the real value of an architecture firm in complex design thinking and problem-solving not digital administration placing standard doors on a model.

00:14:54: But if he's right anyone who has a laptop can generate three hundred viable building designs from text prompt ten minutes.

00:15:01: how do firms actually differentiate themselves?

00:15:04: Is the commercial value now entirely in human judgment rather than digital generation?

00:15:08: That is exact paradigm shift.

00:15:10: When a generation of assets becomes cheap, instant and ubiquitous Human judgement become premium.

00:15:17: highly priced product Makes sense And as design phase increasingly automated and commoditized into cloud The real battleground moves out to physical execution on job site Which

00:15:28: takes us right to the physical frontier.

00:15:30: We are seeing AI and robotics step out of the software, take physical ownership.

00:16:00: that makes an imperfect slow robot suddenly look like a brilliant solution.

00:16:04: We're seeing this with companies like ReFrame Systems, which my year mystery shared.

00:16:08: Oda Automated Framing?

00:16:09: Yep

00:16:09: they are automating the framing and sheathing of modular homes through a tight combination design software feeding directly into physical factory robotics

00:16:17: And technology capital is becoming hyper-specialized too.

00:16:21: Li Xiaolong noted company called BuildAughts just raised one hundred thirty million dollars specifically to scale their AI construction monitoring but explicitly tailored for data centers.

00:16:32: They aren't trying to build a generic tool?

00:16:33: Nope, they are proving that contact is verticalizing because the mechanics and tolerances of building a massive hyper-skilled data center in Texas require completely different methods than building up a bespoke hospital in London.

00:16:46: But the most disruptive shift in the entire industry right now Is who's actually doing the building?

00:16:52: Yeah this is crazy.

00:16:53: Daniels Gapplix and Keytake Hymn issued a fascinating warning about venture capital-backed firms like Unlimited Industries, and Zero RFI.

00:17:02: These are not software companies trying to sell fifty dollar monthly subscriptions...

00:17:06: Now they're building!

00:17:08: They are using AI in heavy robotics—like Icon's massive three D concrete printers or bedrock robotics autonomous heavy earth moving equipment —to sell the finished physical building directly

00:17:21: By passing the traditional AEC studios entirely?

00:17:23: Yes.

00:17:24: Wait, I have to challenge this!

00:17:26: It's a difference between selling of bakery-a slightly better oven and just delivering the baked cake directly to customer.

00:17:33: These VC firms are delivering the cake

00:17:35: They ARE

00:17:36: But aren't they taking on an insane amount of

00:17:38: liability?!

00:17:39: If they take over the highly repeatable designs, data centers and residential blocks does traditional construction just become a bespoke boutique service for highly complex one-off projects like museums?

00:17:51: But if you use automation to sell the finished completed stage of a building, You own the risk.

00:18:15: Yes but you also owned The entire profit margin

00:18:18: and we are seeing that deep integration into the harsh realities Of job sites right now.

00:18:23: Beyonce Park shared an incredible example from China.

00:18:26: They used a heavy lift cargo drone To install two kilometer pilot rope across A massive gorge for suspension bridge

00:18:33: Two kilometers.

00:18:34: yeah

00:18:34: Historically, a job like that required coordinating barges closing commercial navigation channels hours of dangerous vessel towing.

00:18:42: A drone just flew the high-tension cable across the gap in minutes.

00:18:45: It's entirely about rehearsing and managing physical reality before it happens.

00:18:50: Harold Huang demonstrated an agentic four D planning system

00:18:53: where forty means adding time to the three d model right?

00:18:55: Right visual schedule.

00:18:57: But this system rehearses steel erection using simulated digital workers that are strictly constrained by actual OSHA

00:19:04: rules.

00:19:05: The safety regulations?

00:19:06: Yeah,

00:19:06: if a digital worker in the simulation isn't properly tied off with a safety harness... ...the AI physically won't let the virtual crane lift happen showing planners the real schedule impact of strict safety compliance.

00:19:19: And when it comes to human workers actually executing on site today Ricardo Kahn highlighted onsite insights.

00:19:27: They are delivering multi-lingual audio instructions triggered by GPS and Bluetooth beacons on the

00:19:33: site."

00:19:33: That's

00:19:33: incredibly practical!

00:19:34: It is, a worker walks into a specific structural zone...and their phone or smart glasses automatically play the latest installation instructions in their native language.

00:19:43: Stopping rework right at the source?

00:19:45: Exactly plus.

00:19:46: Ahmad Saroya and Sean Lewis discussed autonomous scan to BIM, to GIS.

00:19:52: Where drones and robotic dogs scan a site in AI agents convert those point clouds into usable Revit models and map them into geographic information systems within forty-eight hours.

00:20:03: Prating an instant digital inventory which really brings this entire conversation full circle with AI pushing senior professionals away from being the creators of manual drawings And heavy robotics finally arriving on site to fill the massive labor gap.

00:20:23: What exactly happens to The Junior Builder?

00:20:26: How do we build deep intuition, spatial awareness and vital judgment in next generation of engineers when that knowledge used come entirely from years of manual drafting?

00:20:37: If

00:20:38: you enjoyed this episode, new episodes drop every two weeks.

00:20:42: Also check out our other editions on Smart Manufacturing and Industrial AI And Connected Tools & Equipment.

00:20:49: Thank-you so much for joining us On This Deep Dive!

00:20:52: Don't forget to subscribe To catch the next one.

00:20:54: Keep building, keep learning.

00:20:56: We'll see ya next time.