Best of LinkedIn: Digital Construction CW 20/ 21

00:00:00: This episode is provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about digital construction in calendar weeks twenty-and-twenty one.

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: you can find more info

00:00:25: Right, so welcome everyone.

00:00:27: We are jumping right into our deep dive today extracting You know the absolute top digital construction trends that have been blowing up across LinkedIn recently.

00:00:36: Yeah

00:00:36: specifically from calendar one twenty and twenty-one.

00:00:39: And the mission here is really just to figure out how all this talk about AI and data, and lean practices is actually hitting the job site.

00:00:47: Like how was it affecting the bottom line?

00:00:49: Exactly because I mean if you're listening to this You already know the gap between a sleek software demo And uh A muddy Tuesday morning concrete pour.

00:00:58: It's massive Oh!

00:00:59: Its two different realities We are going keep natural in focus today.

00:01:02: No fluff Just looking at what experts say

00:01:05: Sounds good.

00:01:06: So let's get into the first big theme which is really about AI infrastructure and workflow integration.

00:01:10: Yeah, because what we're seeing Is that AI?

00:01:14: It's no longer this separate tab or you know some tool You have to go open up.

00:01:18: it's actively moving inside The established software stack

00:01:22: right Which is huge.

00:01:23: in James life actually had a really great post About this.

00:01:26: he pointed out something That dropped on April seventh which was the release of Revit twenty-twenty seven.

00:01:30: oh yeah That was a massive update, but it was so quiet.

00:01:34: So quietly if you just skim the notes You'd miss it.

00:01:37: But they built in an mcp server.

00:01:39: so model context

00:01:41: protocol right MVP

00:01:42: And this basically unlocks something incredible.

00:01:45: Instead of being stuck in this closed vendor system, you can point an AI like Clawed or ChatGPT directly at a Revit model.

00:01:54: Wait so just talk to the model?

00:01:55: Literally!

00:01:56: You use natural language to find say all the unrated fire doors Or automate whole model audit.

00:02:03: That's wild because normally that is just hours of clicking and checking properties, right?

00:02:07: Exactly!

00:02:08: And now it's a prompt.

00:02:09: This is happening everywhere.

00:02:11: Jonas Strimholtlau posted about how Bluebeam doing the exact same thing.

00:02:15: Oh really?

00:02:16: Bluebeem building MCP support into review.

00:02:19: Yep So users can literally talk to their drawings.

00:02:22: You just ask it to pull markup data across the whole set.

00:02:25: Wow,

00:02:25: I mean that makes so much sense.

00:02:27: and Rishi S actually gave another example he mentioned.

00:02:29: being human is bringing MCP construction finance So you can instantly query vendor spend?

00:02:35: Just asking your finance stack where the money's going?

00:02:38: Yeah exactly.

00:02:39: but looking at all this i gotta ask a quick sharp question here.

00:02:43: shoot if The AI Is already moving inside the tools we use every day Why are firms still paying for all this redundant tech?

00:02:51: Like why are they buying separate AI

00:02:53: platforms?

00:02:54: that is the million dollar question And Grace Norris brought up.

00:02:57: This exact point.

00:02:58: she noted That most construction firms or already paying for AI through

00:03:02: their normal licenses, right

00:03:03: exactly like Microsoft.

00:03:04: three sixty five her Google workspace.

00:03:06: I already have it

00:03:07: yeah

00:03:08: but They're only utilizing maybe twenty percent of its capability.

00:03:12: where that is I mean, that's just throwing money away.

00:03:14: Yeah they're panicking and buying new stuff when they haven't even used what They have.

00:03:18: it's like.

00:03:18: well think of this way.

00:03:19: before AI was a consultant in A separate building you had to force your team To walk all the Way over there to ask a question.

00:03:28: right completely disconnected from The work.

00:03:30: but mcp is basically Like giving That ai security badge.

00:03:34: now It can freely Walk around Your existing office.

00:03:37: look at the files You are already working on Right At your desk.

00:03:40: thats great analogy.

00:03:41: But, you know having AI in the tool is great but we have to talk about data itself because construction data is incredibly complex.

00:03:49: Oh for sure it's not just techs?

00:03:51: No no and horizontal AI like your standard chat GPT really struggles with it.

00:03:57: which brings us to our second theme vertical AI and translating the real world.

00:04:02: Yeah, and Guido Machiochi has been doing some really interesting work on this with Document Foundry.

00:04:07: He pointed out this critical gap between data being machine-readable And Machine Interpretable.

00:04:12: Okay break that down.

00:04:13: what's the difference?

00:04:14: So machine readable is when an AI looks at a PDF and it sees you know lines in text You can read the letters But machine interpretable Is knowing That A specific polygon On say sheet S one oh One and it adjoins a hallway.

00:04:30: So actually understanding the spatial relationship?

00:04:32: Exactly,

00:04:33: its context

00:04:34: which is so hard for AI we're not up in here.

00:04:37: I'll actually share this brilliant summary of an article by deep tea.

00:04:40: any ready at boon ai.

00:04:43: oh i saw that one really good stuff.

00:04:45: yeah deeply makes us point that construction drawing isn't to photograph.

00:04:49: What's she call it?

00:04:50: A compressed symbolic projection.

00:04:52: A compressed, symbolic projection.

00:04:54: that is a dense phrase but its so accurate!

00:04:56: Right because standard object detection just reads pixels But...a good estimator- a human.

00:05:02: They look at those symbols and they infer the whole building in their head Because

00:05:05: they know how a building actually goes together.

00:05:07: Exactly So AI has to learn symbol grammar And cross referencing.

00:05:12: It can't just look at it..it has understand it.

00:05:14: That perfectly highlights what Victor Agustia was talking about.

00:05:18: he had this realization about estimators.

00:05:20: He found that they are absolutely refusing to give up Bluebeam!

00:05:24: They won't use the new AI interfaces?

00:05:26: Nope, and honestly...they shouldn't have do.

00:05:28: The AIs should replace bluebeam or the estimator's judgment?

00:05:31: Right because the human knows the real world context.

00:05:35: Exactly, The AI should just handle the eight hours of tedious clicking and counting you know?

00:05:41: Just do the grunt work!

00:05:42: Leaving the exception handling... And value engineering to human.

00:05:46: Exactly let the expert be an expert

00:05:49: Which ties into inside.

00:05:50: from Goulnor She pointed out that best AI results they don't come for writing smarter prompts

00:05:56: Like telling it act as engineer.

00:05:58: Right.

00:05:58: That doesn't really work for construction.

00:06:00: Yeah The best results come from real project constraints.

00:06:03: You have to feed it actual budgets, hard timelines and structured engineering logic

00:06:08: you Have to ground in reality.

00:06:10: but that brings us To a huge issue at our third theme.

00:06:13: let me guess What if the reality is a mess?

00:06:17: bingo data quality governance?

00:06:20: And the real bottleneck because of AI Is making decisions based on your project data.

00:06:25: what happens If your data is terrible?

00:06:28: yeah That is a scary thought.

00:06:29: Dan Howard highlighted this alarming stat from fmi it's uh one hundred and seventy seven billion dollars.

00:06:36: Billion with a B?

00:06:37: With the b, one hundred seventy-seven billion is lost annually in u.s construction just to inefficiencies and bad data.

00:06:44: that is insane!

00:06:46: And out of that thirty one billion is purely avoidable.

00:06:48: rework wow.

00:06:50: and dan's warning was very clear ai does not fix bad data It just scales the consequences of

00:06:55: it.

00:06:56: Yeah, Michael Pink had a great really blunt quote about this.

00:06:58: He said AI can't fix bad schedules messy data or weak project controls?

00:07:02: It Just makes bad information move faster.

00:07:04: make that information move fast.

00:07:06: That is exactly right

00:07:07: because if your baseline as chaos ai just automates The Chaos

00:07:11: and he gets worse when we talk About Ai agents Nate Fuller posted some Really eye-opening research about This.

00:07:16: who was an NVIDIA in Stanford study on AI agent Vulnerability.

00:07:20: okay

00:07:20: so like When an AI Is acting On its own.

00:07:23: yes And everyone worries about hallucinations, right?

00:07:26: The AI making things up.

00:07:28: But the study showed that hallucinations only cause nine percent of failures.

00:07:33: Wait!

00:07:33: Really?!

00:07:34: Only nine percent Yeah...the

00:07:36: real threats—making up ninety-one percent were things like gold drift tool misuse and state manipulation Tool

00:07:43: misuse…that sounds dangerous.

00:07:45: Think about it.

00:07:46: What happens when a confused AI agent has standing access to your contract repository in the box?

00:07:52: Oh my god, it could just email out a totally flawed contract of twenty subcontractors before anyone notices.

00:07:57: Exactly!

00:07:58: It didn't hallucinate but misused tools that were given access too.

00:08:02: That is terrifying and brings up all human elements on the field side.

00:08:06: Zulkanin Malik, z-u-l-q on LinkedIn gave a really good reality check about this.

00:08:10: Oh

00:08:10: but the foreman right?

00:08:11: Yeah you spend sixty grand on some new software.

00:08:14: The Foreman opens it once on a muddy site swears at because its too complicated and just goes back to using a paper notebook.

00:08:20: And Right there...the data chain is broken

00:08:23: Completely broken.

00:08:24: If tech adds friction for crew working in rain or dust your profit leaks out and adoption dies instantly

00:08:32: which is why we have to look at this from a lean perspective.

00:08:35: Eduardo Fernandez de Freitas and Felix Urrumbin both touched on this.

00:08:38: Yeah, what was their take?

00:08:39: They noted that construction schedules in projects they don't usually fail because of execution errors on the ground... ...they failed due to weak production systems an upstream design or client factors

00:08:50: So problems happening way before the shovel hits the dirt.

00:08:53: Exactly Governance & Lean practices absolutely must precede technology.

00:08:58: You can just slap AI onto a broken process.

00:09:01: Okay.

00:09:01: So to truly fix that disconnect between the office data and field reality, we have look at our fourth theme.

00:09:07: Digital Twins & Physical AI.

00:09:10: Yes

00:09:10: because the job site is not a document.

00:09:12: Right.

00:09:13: Jeevan Kalanithi made this exact point in the ENR Future Tech conference.

00:09:17: He said all these agentic ai conversations are way too focused on text and documents

00:09:21: But construction is physical.

00:09:23: Exactly it's constantly changing physical fields.

00:09:26: so we need what he calls physical Ai.

00:09:29: It needs to be grounded in spatial, time-stamped reality data.

00:09:33: Moving

00:09:34: from what someone reported... ...to what actually happened?

00:09:37: Exactly!

00:09:37: And I mean that is the ultimate dream of digital twins right?

00:09:40: Yeah

00:09:40: absolutely.

00:09:41: But Peter Mitev pointed out a huge problem.

00:09:44: The market for Digital Twins is projected to hit seventy three point five billion dollars by twenty thirty.

00:09:50: Huge number

00:09:51: Huge.

00:09:52: But right now, only twenty-two percent of AEC organizations have fully implemented foundational BIM level two.

00:09:58: Twenty-two

00:09:59: per cent?

00:09:59: Wow!

00:10:00: So the basics aren't even there for most people...

00:10:02: Right so there's this massive execution gap between the hype and reality.

00:10:07: How do we actually close that DAP?

00:10:08: FlooringHumor offered a really good solution.

00:10:10: He laid out five step blueprint For scalable digital twin architecture.

00:10:15: What is his approach?

00:10:16: His biggest warning was how you treat data.

00:10:19: You have to treat BIM as a spatial database, extracting the metadata via IFC.

00:10:23: Not just a three-D picture?

00:10:24: Exactly.

00:10:25: and you have to decouple your live data streams from The Three D Model

00:10:29: because if you don't de-couple it

00:10:30: Right!

00:10:30: If hard code LiveData into the visuals... ...the twin becomes too heavy and dies.

00:10:36: You have deploy the visual tech like VR or WebGL.

00:10:41: last

00:10:42: Get the Database right first make it pretty second.

00:10:45: Exactly!

00:10:46: Well, looking at everything we've covered from these two weeks... It leaves us with kind of a provocative thought to mull over.

00:10:53: I love a provocative though.

00:10:54: lay

00:10:55: on me.

00:10:55: well if AI agents require this perfectly closed loop highly observable systems just to function without breaking things yeah perhaps ai's greatest legacy in construction won't be the algorithms themselves.

00:11:07: okay what will it be then?

00:11:08: maybe that real revolution is they're just preparing for.

00:11:12: Ai is finally gonna force the construction industry to standardize, clean up and industrialize its messy human workflows once-and for all.

00:11:21: Oh

00:11:21: wow!

00:11:22: Yeah it's like The Ultimate Forcing Function.

00:11:24: You can't use this shiny new toy until your room is spotlessly clean.

00:11:27: Exactly

00:11:28: I think that a perfect place to leave if you explore on your

00:11:42: own.