Best of LinkedIn: Digital Construction CW 32/ 33

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

00:00:08: Furnaces is a B to B market research company that supports industrial automation an ICT companies with market intelligence across the construction industry to prioritize segments identify high value accounts and validate use cases.

00:00:22: Yeah And imagine spending eight hundred forty five million dollars in an industry where profit margins you know they barely scrape three percent.

00:00:30: Right, it's wild.

00:00:31: It is.

00:00:32: but that is exactly what is happening in construction right now

00:00:34: because we are really trying to cut through all of the software marketing hype today.

00:00:38: I mean you constantly hear about digital transformation But were looking at raw reality.

00:00:43: Exactly so.

00:00:44: In this deep dive We're going examine top Digital Construction Trends specifically focusing on What Is Genuinely Changing How You Manage A Job Site and how You Run The Back Office.

00:00:55: Yeah Because Friction Out There palpable.

00:00:58: if you step onto any major job site today, You'll see this massive disconnect.

00:01:02: Oh absolutely It's a disconnect between the tech companies trying to sell digital tools and you know The superintendents actually pouring the concrete.

00:01:11: Kenneth Brabham A general contractor down in Houston highlighted This beautifully.

00:01:15: what did he say?

00:01:16: Well He pointed out that while dozens of AI startups are raising these massive venture capital rounds.

00:01:23: Almost none Of them Are led by people who have Actually built a building right Right.

00:01:27: And the result is just this flood of highly fragmented tools, like a startup will build a hyperspecific AI that just does blueprint takeoffs and then they hand

00:01:44: unpaid integration layer.

00:01:45: Exactly, think about the mechanics of that.

00:01:47: for a second use an advanced AI to extract quantities from a PDF which sure it takes thirty seconds seems

00:01:53: great on paper

00:01:54: right.

00:01:55: but then you have to export that data as a CSV manually reformat the columns to match your scheduling software import and

00:02:02: then realize this scale was misread by factor ten.

00:02:05: yes And you have to do it all over again.

00:02:08: So the software executed one micro task faster, but The human is still carrying the mental load of stitching the entire project together across like nine different incompatible applications.

00:02:20: so It's basically like buying a dozen Incredibly specialized expensive kitchen gadgets that just completely refuse to talk to each other.

00:02:28: That's

00:02:28: a great way to put

00:02:29: And you are still the one doing all of the prep work, cooking and cleaning.

00:02:34: It really raises a question whether these vendors actually understand the physical physics of their job they're trying to disrupt?

00:02:41: Well buyers are definitely losing patience which is creating this completely new evaluation standard before anyone signs software agreement.

00:02:49: now testing domain expertise for it.

00:02:53: Oh interesting like a quiz

00:02:55: Kinda yeah.

00:02:56: Derek Bergen shared this brilliant litmus test.

00:02:59: He suggested asking a software sales rep to explain the difference between two pipe and four-pipe hydronic system.

00:03:05: Oh, man Let me guess if they start sweating in looking for their notes you just walk away.

00:03:09: basically yeah

00:03:10: Because, I mean for those listening a four pipe system has separate supply and return lines.

00:03:16: For both hot and chilled water meaning you can heat one side of the building while cooling the other right?

00:03:21: Right!

00:03:21: And two-pipe systems cannot do that.

00:03:23: so if software developer designing your new AI routing tool doesn't know basic mechanical reality they're going to generate massive collision in the ceiling plenum.

00:03:32: Yeah there are what he called software tourists

00:03:35: Software tourists.

00:03:36: it is perfect label and this backlash is actively reshaping how content companies hire too.

00:03:42: Oliver Paul, who recruits for these specific roles noted a massive shift in candidate success rates.

00:03:48: Really?

00:03:48: What kind of shift?

00:03:50: Well... A software account executive could walk into an interview with a hundred-and-forty percent quota attainment history and they still lose the job to someone with lower numbers if that second candidate understands the physical constraints of

00:04:20: absolutely cannot troubleshoot a cloud sync error while a fleet of concrete tracks is waiting on your approval.

00:04:26: You

00:04:26: just can't.

00:04:28: the physical environment breaks fragile software and because field teams have been burned so many times by apps that you know they work in a pristine demo but fail-in-the-dirt general contractors are rejecting AI marketing claims entirely,

00:04:44: Just writing them off completely.

00:04:45: yeah.

00:04:45: Rachel Manina observed that the industry is aggressively shifting back to trust and relationships.

00:04:51: I mean, you cannot fake peer validation when one project engineer tells another that a tool actually solves A problem without creating three new ones.

00:04:59: That carries so much more weight than any enterprise Kitchdeck

00:05:02: totally.

00:05:03: But you know, if the field guys are rejecting the tech because it's fragmented we really need to look at the back office.

00:05:08: Because they're dealing with the exact same headache just in a different format.

00:05:12: I mean They aren't wrestling with muddy iPads but they are wrestling with digital models that Are completely disconnected from The operational reality of the building

00:05:21: right?

00:05:21: Because the digital model is supposed To be this single source of truth.

00:05:25: BIM building information modeling was supposed to solve coordination.

00:05:28: Supposed

00:05:29: do, yeah.

00:05:29: But

00:05:29: Flipe Bostos audited a series of major projects and found that what people call a BIM model is often just beautiful three-D drawing.

00:05:37: Just the picture exactly

00:05:39: it contains zero structured data No parameters no coordination logic.

00:05:45: he labeled at bin Building Information nothing

00:05:48: Building information, nothing.

00:05:49: I love that!

00:05:50: So it's just a very expensive hollow picture?

00:05:53: Exactly

00:05:53: Because if you move the structural column in true BIM environment The HVAC system should logically understand how to reroute itself around this new obstacle.

00:06:02: But If data is flat geometry... ...the pipe sits there clipping right through steel.

00:06:07: Yeah and lack of structured parameters becomes massive liability when introducing artificial intelligence into the mix.

00:06:14: Metal Eldon pointed out the danger here.

00:06:16: The industry is super eager to unleash AI inside Revit, assuming that algorithm will just automatically clean up a messy model.

00:06:24: Right

00:06:24: Just hit the fix it button.

00:06:26: But Revit is deeply parametric dependency rich ecosystem.

00:06:31: It's like letting an AI change single cell in massive Excel financial model.

00:06:36: but the AI doesn't understand complex formulas attached to that cell.

00:06:40: Perfect analogy.

00:06:41: It changes the diameter of a primary water line, and suddenly... ...the entire mechanical system on the roof shrinks by half because some hidden constraint.

00:06:48: Yes!

00:06:49: The cascading failures would be catastrophic.

00:06:52: That's why industry must adopt continuous model compliance similar to CICD Continuous integration and continuous deployment in software engineering.

00:07:00: Okay so how will that work?

00:07:02: Basically every time an AI or human makes a change An automated script needs run into background Checking that change against strict set boundary

00:07:10: rules.

00:07:11: The drift has to be detected immediately inside a safe sandbox.

00:07:16: Which brings up a massive issue of leverage because if we need strict guardrails for the AI who actually owns the track it's running on?

00:07:22: Oh, right!

00:07:23: I mean... If your building data is fragmented across a dozen different proprietary platforms you don't control boundaries

00:07:30: And that is why Data Sovereignty is becoming the most critical conversation in construction Right now.

00:07:35: Adam Stark argued having line-in-your contract.

00:07:38: saying We own our data is completely meaningless if a software vendor controls the underlying platform architecture.

00:07:44: Right, because true ownership means you define how the information is structured.

00:07:48: You define which system acts as the authoritative source

00:07:51: Exactly

00:07:52: And control where historical operating data actually resides.

00:07:56: If the vendor dictates the architecture they dictate How your AI is allowed to interact with their own building.

00:08:02: They hold keys for future innovation.

00:08:04: It's wild.

00:08:05: And by the way, if you want to stay ahead of how data ownership and artificial intelligence are fundamentally rewriting the rules of this deep dive.

00:08:17: Definitely, because this realization that controlling the platform architecture is the ultimate end game it really explains the sheer volume of money moving around right now.

00:08:27: The big players aren't acquiring startups just to add a new button to their dashboard?

00:08:31: No not at all.

00:08:32: They're

00:08:32: buying complete control over workflow data.

00:08:35: We are witnessing an unprecedented wave of consolidation and valuations reflect absolute desperation We mentioned at the very beginning, Paul Arpino highlighted Procore's eight hundred and forty five million dollar acquisition of drone deploy.

00:08:51: Yeah that number is staggering!

00:08:53: It is.

00:08:54: Contractors take on years of extreme financial and physical risk for a three percent profit margin while software company pulls in near billion-dollar valuation.

00:09:02: But you know unpack what was actually purchased there.

00:09:04: they didn't just buy tool that flies quadcopters around to site.

00:09:07: They bought the underlying physical site data, captured across thousands of massive infrastructure projects over the last decade.

00:09:14: Exactly!

00:09:15: Scythian Metha analyzed this strategy behind it.

00:09:18: noting that Procore is transitioning from a system record to action.

00:09:24: by layering BIM tools, scheduling AI and visual intelligence into one ecosystem they are constructing highest level ontology for built environment.

00:09:35: Okay, let's break down how that ontology actually functions mechanically.

00:09:38: Because so a drone flies over a site and creates a point cloud millions of XYZ coordinates mapping the exact current state of a concrete pour right?

00:09:47: The system then overlays that point cloud directly onto the three D Revit model.

00:09:52: if the physical concrete is two inches out of alignment with the digital design... ...the system doesn't just record the error it

00:09:58: acts on

00:09:58: exactly.

00:09:59: the AI recognizes the discrepancy checks the schedule, realizes this is going to delay the steel erection next week and autonomously drafts an RFI into the structural engineer.

00:10:08: That's exactly

00:10:09: what it looks like between a system of record...and action.

00:10:13: Jama Drake observed shift in venture capital behavior too.

00:10:17: Funding no longer flows into applications that digitize paper processes such as turning a daily log into iPad form.

00:10:26: Capital is aggressively pursuing technology that eliminates human workflow entirely!

00:10:31: I have to challenge the reality of that though.

00:10:34: The venture capital thesis sounds great, but are firms actually achieving this level of autonomous action today?

00:10:39: Well...the data tells a very sobering story about the gap between adoption and actual automation.

00:10:47: Li Xiaolong shared statistics showing while AI usage & property management recently spiked to fifty-eight percent only eight per cent of those firms successfully automated complete process from end.

00:10:58: Wow, a fifty percent gap between using the tool and getting the promised result.

00:11:02: Yeah.

00:11:03: I'm assuming that comes right back to the messy data problem we were talking about?

00:11:05: A hundred percent!

00:11:06: An AI can draft beautiful email but if you ask it to reconcile a broker's poorly scanned PDF against a lender's rent role in a partner's custom Excel sheet... The AI hallucinates or fails because every single document defines basic terms differently.

00:11:25: The massive M&A race we just discussed is the industry's attempt to force all those different documents and definitions under one roof.

00:11:33: Because once that data has actually integrated, then nature of AI changes fundamentally.

00:11:38: It stops being a chatbot, you query.

00:11:40: And it becomes the digital employee that actually executes.

00:11:43: and this cross is a major threshold.

00:11:46: Zach Korber presented A Real World Case Study at a recent AGC tech conference.

00:11:50: That completely redefines The back office.

00:11:53: Oh This was fascinating.

00:11:54: Yeah He introduced a system operating under the persona Wendy.

00:11:58: Wendy Is an AI agent running administrative operations for An active general contractor?

00:12:02: She operates autonomously.

00:12:04: She tracks down missing insurance certificates, she processes weekly certified payroll documents and she actively emails subcontractors who are lagging behind.

00:12:11: That's crazy!

00:12:12: And the subcontractors email her back often without realizing they're negotiating with an algorithm.

00:12:17: that is wild...and it's not just basic admin work either.

00:12:21: The pre-construction phase has seen the exact same shift.

00:12:24: Michael Edelin detailed how his open takeoff platform is engineered with AI agents treated as first-class users.

00:12:31: Meaning

00:12:32: what exactly?

00:12:32: The software isn't just a tool for human estimator, it's a workspace for the agent!

00:12:37: The mechanics of that are fascinating... ...the agent ingests a massive set of architectural plans autonomously confirms this scale traces the geometry every room calculates material quantities and exports fully marked up sets drawings.

00:12:53: But the critical feature is the art of ability.

00:12:55: Every single quantity generated by an agent carries a provenance receipt,

00:12:59: which means that human senior estimator can click on a number and see exact mathematical logic.

00:13:04: in this specific polygon AI used to arrive at that material count.

00:13:08: Exactly!

00:13:08: You aren't just blindly trusting black box.

00:13:11: you are managing the AI's work output like you would manage a junior human estimator.

00:13:16: And these agents move upstream into complex design too.

00:13:20: Michael Hopp documented an instance where a large language model, specifically Clawed AI was used to model the structural framing of John Hancock Center directly inside Revit.

00:13:30: That just blows my mind because A Large Language Model is fundamentally a text prediction engine.

00:13:36: But by interacting with the Revit API, it's proving that can understand three-dimensional spatial logic structural load paths and parametric constraints.

00:13:45: It isn't just generating a flat two D rendering of building is actually writing code that generates mathematically sound three D structure.

00:13:54: productivity

00:13:54: implications are staggering.

00:13:56: but There is a massive often ignored risk hiding inside all of this automation.

00:14:03: Oh,

00:14:03: I see where you're going with us If digital employees like Wendy are autonomously modeling the structure doing the quantity takeoffs and chasing the insurance certificates.

00:14:11: What happens to the human junior employees

00:14:13: exactly?

00:14:14: The

00:14:14: entire industry is built on an apprenticeship model.

00:14:17: You learn how to build a building by grinding through the tedious work.

00:14:20: Olivia Grindel raised the alarm on this exact crisis.

00:14:24: We have a generation of AI-native young workers entering an industry where the entry level jobs had been completely seniorized.

00:14:33: The traditional method of absorbing tribal knowledge, you know.

00:14:36: spending three years reviewing hundreds of repetitive RFIs manually highlighting specs finding the physical clashes on the drawings.

00:14:44: that path is being automated out of existence.

00:14:47: So A Kid Fresh Out Of An Engineering Program has handed us Senior Level Task On Day One.

00:14:52: They're told to review the AI's provenance receipts and judge whether the algorithms quantity takeoff is accurate.

00:14:58: But how can they possibly judge the AIs output if they have never suffered through the underlying work themselves?

00:15:04: They cannot, you simply won't have intuition spot subtle errors that only come from experience.

00:15:10: Grindel warns unless firms deliberately construct new training frameworks that force young workers learn alongside the AI industry will hit a catastrophic skills cliff.

00:15:20: in decade The pipeline of senior talent will simply evaporate.

00:15:24: Which totally flips the value proposition of AI on its head, right?

00:15:28: We've been talking about AI as a tool to replace hours worked.

00:15:31: but Aditya Locke argued that the ultimate value of AI in construction isn't just raw productivity.

00:15:37: It's about knowledge

00:15:38: Exactly!

00:15:40: Its true purpose is to capture the invisible tribal knowledge of your veteran superintendents and turn it into permanent queryable organizational memory.

00:15:49: Think about the knowledge lost every time a thirty-year superintendent retires.

00:15:54: All of lessons on how specific soils react in a freeze, or which subcontractors consistently underbid an over bill.

00:16:01: If AI is integrated into workflow it records all those thousands of micro decisions.

00:16:06: So five years from now during design phase of new project The system analyzes soil report remembers a failure from past project that no current employee was even around for and autonomously flags the structural engineer to adjust foundation design.

00:16:21: Exactly, it creates an institutional memory that scales.

00:16:24: It connects patterns across thousands of data points so no single human brain could ever retain at once.

00:16:29: It fundamentally changes what construction company actually is...

00:16:32: ...it really does.

00:16:33: What if true competitive advantage your firm in future isn't physical steel or glass you deliver?

00:16:40: What if the physical building is just a byproduct of highly tuned invisible data engine?

00:16:46: That's profound thought.

00:16:47: The companies that will dominate the next decade are the ones focusing on that invisible learning loop right now.

00:16:52: They're enforcing structured data discipline, controlling their platform architecture and using AI to institutionalize their tribal

00:17:12: knowledge.".