Best of LinkedIn: Digital Construction CW 28/ 29

00:00:00: Provided by Thomas Allgaier and Frenus, based on the most relevant LinkedIn posts about digital construction in calendar weeks twenty-eight and twenty nine.

00:00:08: Frenous 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.

00:00:21: Yeah so imagine just letting your smartphones autocomplete algorithm design of load bearing bridge.

00:00:30: I mean, it sounds completely terrifying.

00:00:32: Yeah that sounds like a massive lawsuit waiting to happen

00:00:34: exactly.

00:00:35: but today artificial intelligence is actually actively writing the blueprints The schedules the estimates for our next generation of skyscrapers.

00:00:44: So welcome to the deep dive.

00:00:45: glad you're here with

00:00:46: us yeah.

00:00:46: And if you joining this?

00:00:47: You know the construction industry is well.

00:00:49: It's notorious for being late to the tech party.

00:00:52: based on the field reports we're analyzing today, that error is basically over.

00:00:56: We are cutting through all of the vendor fluff to look at you know...the actual nuts and bolts about how digital construction is working right now On The Dirt.

00:01:03: And let's just jump right into the deep end here, which is pre-construction.

00:01:07: We're seeing AI officially graduate from those shiny corporate pilot decks straight in to the trenches with project managers... Which

00:01:16: where it actually matters?

00:01:17: Exactly!

00:01:18: It all about speeding up heavy lifting.

00:01:20: So there was this great example of sources.

00:01:22: when someone mapped out a workflow They took this massive I mean three hundred page technical specification document and they fed that whole thing.

00:01:32: Oh,

00:01:32: wow.

00:01:33: And I mean historically that means an estimator sitting there for days just manually reading highlighting cross-referencing...

00:01:42: Just hoping they didn't miss something?

00:01:44: Right!

00:01:45: crazy specialized material requirement buried on like page two, fourteen.

00:01:49: Yeah exactly.

00:01:50: but they didn't just say you know hey AI summarized this.

00:01:53: They actually attached a highly specific Croatian cost database to it and a blank standardized Excel template.

00:01:59: so the essentially built strict guardrails around The large language model

00:02:03: they forced into play by the rules.

00:02:04: yeah

00:02:05: yes.

00:02:05: And then I ran as a background process matching the spec text to the database line items.

00:02:13: Twenty-nine division conceptual estimate with a risk register in like twenty minutes.

00:02:18: That is wild, but you know the mechanism.

00:02:20: there's what really key.

00:02:21: they aren't asking The AI to just guess the cost of steel.

00:02:25: They're using its semantic ability right.

00:02:27: it's reading comprehension Yeah To parse the text and map it to us.

00:02:30: strictly deterministic database Right.

00:02:33: And major platforms are seeing this and just hard coding it.

00:02:36: Trimble actually Just announced their applying This kind Of Ai to Their MEP Estimating Tools

00:02:40: for Mechanical Electrical and Plumbing

00:02:43: Yeah, and for anyone outside that specific engineering space MEP drawings are They're just incredibly dense.

00:02:50: Just layers and layers of lines in symbols.

00:02:52: yeah a human has to basically count every single junction box Every foot of ductwork clicking one by one on a PDF.

00:02:58: it takes forever

00:02:59: Exactly.

00:02:59: But now Trimble system uses pattern recognition To detect I think they said eighty percent Of those count of objects automatically It sees the geometry of a symbol knows its VAAV box and tallies it.

00:03:13: Estimators are cutting their time by up to sixty percent, then you've got Outbuild launching an AI schedule generator from scratch.

00:03:20: Okay wait hold on though.

00:03:20: let me put back this a little bit Because an estimate or a schedule, that is the financial bedrock of a project.

00:03:26: And large language models at their core are essentially high-powered guessing machines right?

00:03:32: They predict the next logical word they're probabilistic

00:03:36: but

00:03:36: construction is fiercely deterministic.

00:03:38: I mean you can't pour a concrete foundation on a probabilistic guess.

00:03:42: isn't this incredibly dangerous?

00:03:44: it feels a bit like letting us spell checker.

00:03:46: design a bridge.

00:03:46: Oh your totally right!

00:03:47: That's the crucial bottleneck.

00:03:49: And a post from Bill Trace analyzed this perfectly.

00:03:53: They pointed out that if you run the exact same prompt with the same drawings through a raw LLM three different times, You

00:04:00: get three different answers.

00:04:01: Exactly, you get three slightly different estimates and an industry with razor thin margins just cannot rely on fluid assumptions.

00:04:10: So Bill Trace solves this by entirely separating the tasks that use the AI purely for semantic extraction.

00:04:16: so just reading them messy human input

00:04:18: translating human language into structured data yes

00:04:21: And then that structure data is handed off to a deterministic causal chain?

00:04:26: So bill trace keeps like ninety percent of its systems actual logic hard-coded.

00:04:31: The math is just traditional software math the AI's basically.

00:04:34: Just do it with a data entry got It.

00:04:36: so you use the AI to get the estimate too.

00:04:38: maybe eighty percent.

00:04:39: But a human expert still have to validate that final gap

00:04:42: Which honestly requires?

00:04:43: A massive cultural shift on the actual job site because the firms succeeding With this they aren't managing it from some corporate IT silo.

00:04:50: like That post about Suffolk construction

00:04:51: John Fisch.

00:04:52: yeah, their CEO John Fish made the call To literally embed an AI engineered directly into the jobsite trailer.

00:04:59: that

00:04:59: Is So smart

00:05:00: right?

00:05:01: Not at headquarters.

00:05:03: This engineer is sitting in the dirt wearing boots, sitting on weekly coordination meetings with actual superintendents.

00:05:11: See that bridges this gap because software engineers usually try to solve problems builders don't actually have and builders really know what software's capable of doing so.

00:05:22: by putting a tool maker into exact same room as the tool user you bypass all theoretical stuff.

00:05:30: The divide is definitely no longer about who believes in AI.

00:05:34: It's about Who Is Taking Action on it this year.

00:05:36: Yeah, exactly.

00:05:37: And the teams that are taking action today?

00:05:39: they're pushing way past just simple text extraction.

00:05:43: We're moving into a phase where AI is actively creating geometry Its becoming an agent within VIM environments.

00:05:49: Oh!

00:05:49: A genetic BIM.

00:05:50: Did you catch that Geopogo example?

00:05:52: I did.

00:05:52: yeah

00:05:53: Yeah...the technical implications of their Claude to Revit connector Are just fascinating

00:05:57: Right?

00:05:57: They had that demonstration On the Autodesk store

00:06:00: where the user just uploaded a simple flat, two-D photograph of building into Claude.

00:06:07: And the AI didn't describe the photo.

00:06:10: it actually understood architecture and identified walls windows roof lines.

00:06:14: Then translated that to native Revit API commands?

00:06:18: Yes!

00:06:18: It actively recreated this building as fully editable BIM geometry directly inside the modeling software.

00:06:26: We're literally looking at this shift from generative text to generative workflows, and it's not even just drafting from photos.

00:06:33: there was that mere example where they connected Autodesk, Forma, Revit and Microsoft co-pilot studio right?

00:06:40: They created these agent work flows that actually automate heavy engineering modeling.

00:06:44: It is wild!

00:06:46: But wait its great.

00:06:47: that can generate a layout faster sure but if team relies on for real multi million dollar project actually approved

00:06:54: it.

00:06:54: That's a million dollar question,

00:06:55: right?

00:06:55: What happens if the AI script breaks something hidden inside of the three D model and no one notices until they fabricate

00:07:16: enterprise grade governance.

00:07:17: Yeah, it needs accountability.

00:07:19: exactly

00:07:20: a Revit automation isn't real software.

00:07:22: until It has strict permissions version control runtime logs and A definitive human owner.

00:07:29: right you basically need a black box flight recorder for your AI.

00:07:32: precisely You Need to know?

00:07:35: Exactly which geometric constraints were checked by the algorithm at what specific time And Which Human Engineers signed off on it.

00:07:42: The governance layer isn't sexy, but it's required.

00:07:45: Exactly generating geometry gets all the likes on LinkedIn But proof and accountability are the only things that make it trustworthy on a live project.

00:07:53: Yeah

00:07:53: absolutely.

00:07:54: And hey if you rely on this kind of trust worthy practical analysis to stay ahead Of the curve take a quick second right now To just hit subscribe On your app.

00:08:01: so don't miss our future editions

00:08:03: Definitely!

00:08:04: And actually talking about filtering through the noise is A perfect segue to Our next theme

00:08:08: Interoperability.

00:08:09: Yes, yeah

00:08:10: because if we have all these powerful AI agents working inside BIM The whole ecosystem basically collapses If that data is trapped in a proprietary silo.

00:08:20: Yeah the data absolutely must flow freely.

00:08:23: This is the ultimate architectural priority right now and Arcadius Gasky had a great post illustrating this war against Proprietary lock-in.

00:08:32: Oh with the railway right?

00:08:34: He brought a Bentley project-wise IFC model, and for those listening who might not be in the weeds.

00:08:41: OpenPDF format for three-D building data.

00:08:44: He brought that model directly into SREs ArcGIS.

00:08:47: So bringing a hyper detailed engineering models straight Into a geospatial mapping environment

00:08:52: Exactly to create a digital twin of the thirty kilometer railway.

00:08:55: The whole point is that Data shouldn't just belong To whoever generates it.

00:08:58: Right

00:08:59: and we saw That same philosophy from PropEx.

00:09:01: They shared an open schema IFC workflow running From Autodesk straight into Blender BIM, and they use pattern matching in AI to clean up all that notoriously messy raw BIM data.

00:09:11: Right

00:09:11: standardizing

00:09:12: it yeah.

00:09:12: so they generate perfectly clean metadata which means they can run automated ESG reporting for like carbon tracking without a human manually fixing the data first.

00:09:23: but let me ask you this Why not just avoid all these integration headaches entirely?

00:09:28: why Not, Just buy one giant enterprise software suite that does everything like sign a massive One million dollar Palantir contract and force everyone to use it.

00:09:38: Well the posts from crew in AEC foundry systematically dismantled That whole idea first off.

00:09:44: ninety five percent of The industry just cannot afford a million-dollar contract true but more importantly generic all-in-one platforms fail because they try to fit the average workflow of a generic firm, which means that you don't fit your specific firm.

00:09:57: So the real solution they propose is this single affordable unified data foundation.

00:10:02: You don't buy one app and it tries do everything okayish...you build one secure data layer.

00:10:06: then your teams plug in highly specialized best-in class point tools for scheduling or modeling

00:10:11: the value compounds in The Data Foundation.

00:10:15: And that foundation is honestly crucial when we talk about where the real work is happening, which is existing structures.

00:10:21: Retrofits?

00:10:22: Yeah!

00:10:22: Which brings us to reality capture because most work today isn't new builds it's taking the messy physical world and bringing into our clean digital foundation.

00:10:31: yeah...and one of sources mapped out this really interesting twenty-twenty six existing building tech stack.

00:10:37: they priced at about.

00:10:41: And while Revit is still the spine, The real future in tools like Material Index for material reuse.

00:10:48: Right, because the carbon reduction argument and cost arguments are finally aligning.

00:10:52: Exactly But you need data first!

00:10:55: I loved The Auckland Night Grid project for this.

00:10:57: Oh that was incredible

00:10:59: A master student using municipal lidar to infill missing Z-axis height data For OpenStreetMap Because ninety seven percent of building footprints were flat two D shapes.

00:11:09: So they used a lidar blanket To give everything its true height

00:11:13: And they optimized it so well That the entire three d mesh fits into six megabyte

00:11:17: file.

00:11:18: That's just brilliant reality capture infrastructure.

00:11:21: It is, but I hear the term digital twin thrown around constantly with stuff like this.

00:11:26: isn't a Digital Twin without a solid maintenance plan basically?

00:11:29: Just a really expensive three D screensaver?

00:11:31: Yes

00:11:32: exactly it's The biggest trap.

00:11:35: A source laid out three strict questions owners have to answer before scoping a twin.

00:11:40: one Who owns the updates

00:11:43: right?

00:11:43: who pays to maintain it next year?

00:11:45: yep two What decisions will this twin actually drive?

00:11:49: And three, what is the minimum viable data set.

00:11:52: Meaning you don't always need a photorealistic mesh.

00:11:54: Rarely do you need that.

00:11:56: Asset IDs and simple maintenance links are way better than a photo-real model.

00:12:00: it takes ten minutes to load.

00:12:01: Okay so we've gone from estimating to modeling to data to reality capture.

00:12:06: Now the loop finally closes.

00:12:08: Digital intelligence is driving physical autonomous action on job site Physical

00:12:12: AI.

00:12:13: It's finally here.

00:12:14: Look at red men robotics, they were laying out a hyperscale data center using HP site print.

00:12:19: it is this robotic layout tool that prints the digital blueprint directly onto concrete floor.

00:12:24: Yeah They mapped forty five hundred bolt locations ten times faster than human crew with sub three millimeter accuracy.

00:12:32: And we're moving beyond just drawing lines too.

00:12:35: Do you all partnered August Robotics on Dale?

00:12:38: An autonomous drilling robot

00:12:39: Oh!

00:12:40: The ceiling driller.

00:12:42: It navigated a site and drilled two hundred thirty thousand pilot holes overhead with ninety-nine point nine seven percent accuracy.

00:12:49: it saved one hundred ninety weeks of schedule, And then Team UAV showcased these confined space drones With thermal sensors and lidar

00:12:57: So you don't have to send human into dangerous sewer or tank?

00:13:01: Exactly!

00:13:01: Completely removes the human risk.

00:13:03: But practically speaking if I had cleaning robot from vendor A Drilling robot from Vendor B and drone from vendor C How on earth do I manage all of them across a hundred buildings without needing a dozen different iPads?

00:13:15: That is the exact logistical nightmare.

00:13:17: But, The solution emerging in this source is called the Ambient Permission Plane.

00:13:20: Okay unpack that.

00:13:22: So QuickBot has product called Quicksync.

00:13:24: Basically elevators and security doors run on closed software.

00:13:27: A DeWalt robot doesn't know how to speak To a Schindler elevator

00:13:30: so it just gets stuck In hallway.

00:13:32: Right but This ambient permission plane acts as universal translator And air traffic control.

00:13:36: The Robot pings the server The serval translates the request to the elevator, holds the doors open and securely orchestrates everything in real time so that robots don't collide.

00:13:46: That

00:13:47: is just wild!

00:13:48: We are moving so fast...

00:13:49: You really are.

00:13:50: So before we wrap from all these sources what was most vital takeaway for you?

00:13:55: You know it's actually a provocative thought of one post.

00:13:58: they asked Is technology replacing job site conversation?

00:14:01: Oh

00:14:01: interesting

00:14:02: Yeah because we have BIM AI digital QA dashboards.

00:14:07: Yet we'll sit there and send emails instead of just walking twenty meters across the site to solve a problem face-to-face.

00:14:12: The tech becomes a buffer.

00:14:13: Exactly, no algorithm can replace trust you know?

00:14:16: And NO DAC board replaces true leadership.

00:14:19: So the most valuable skill in an AI powered future might literally be keeping our humanity.

00:14:25: That is a great point to end on.

00:14:27: If you enjoyed this episode, new episodes drop every two weeks!

00:14:31: Also check out our other editions of Smart Manufacturing and Digital Power Tools.

00:14:36: Thank-you so much for joining us.

00:14:37: Remember to subscribe And we'll see ya next time.