Best of LinkedIn: Digital Construction CW 38/ 39

00:00:00: provided by Thomas Alguair and Frennis, based on the most relevant LinkedIn posts about digital construction in calendar weeks thirty-eight and thirty nine.

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

00:00:22: so what if i told you that big tech companies like we're talking meta Google players who deal almost entirely in the digital realm are currently spending half a billion dollars to teach human beings how to pour concrete.

00:00:37: Wait, really?

00:00:38: That sounds completely backward

00:00:41: right But if you're a smart builder manufacturing professional listening to this You probably already know that the physical world is you know hitting a massive bottleneck.

00:00:51: Yeah And that's exactly why we're here today on The Deep Dive our mission just cut straight through the noise, no fluff.

00:00:59: Exactly!

00:00:59: No fluff.

00:01:00: We are filtering out all of the endless hype and distilling absolute top digital construction trends from the front lines

00:01:08: Because we really need to look at how data is functioning on job sites right now.

00:01:13: I mean...the measurable impact AI has beyond those flashy tech demos.

00:01:18: Right And maybe most importantly The underlying business mechanics.

00:01:22: Yeah, because right now the economic structure of construction is either going to drive this digital transformation forward or it's completely choked

00:01:31: out.

00:01:32: Okay so let us unpack from the ground up Before we even entertain the idea of autonomous AI agents designing skyscrapers.

00:01:40: We really have talk about the unglamorous plumbing of construction

00:01:43: The boring stuff

00:01:44: Exactly!

00:01:45: The absolute foundation Data and interoperability

00:01:49: which is the single most critical prerequisite.

00:01:51: I mean, you simply cannot build a highly intelligent predictive system on top of fractured chaotic

00:01:56: data.".

00:01:57: Right and there was a massive yet seemingly boring update brought up by John Phillips recently.

00:02:03: he noted that Autodesk has been quietly unifying its coordinate systems across Forma Revit and Civil Three D.

00:02:09: Oh yeah

00:02:10: Yeah they went in and standardized over two thousand coordinate labels

00:02:15: thousand different ways.

00:02:16: the software was essentially trying to describe where something is located in physical space.

00:02:21: It's crazy,

00:02:22: which I know sounds incredibly dry until you understand that catastrophic cascading effects of what happens when those coordinates just don't perfectly align.

00:02:31: So let's actually visualize that for a second.

00:02:34: if you're managing a massive commercial project You have civil engineering teams mapping the terrain and civil three D while The architects are designing the structure and Revit.

00:02:44: right Historically, because those platforms calculated spatial language differently.

00:02:49: when you merge the files.

00:02:51: The digital building might literally drift a few feet from its intended location which

00:02:55: is terrifying.

00:02:57: and the mechanical engineer who routes a massive HVAC system based on that slightly drifted model?

00:03:02: well they have no idea.

00:03:03: there's a problem

00:03:04: none at all.

00:03:05: nobody realizes it.

00:03:06: until the physical concrete Is actually poured on the job site?

00:03:09: That duct work arrives And it suddenly intersecting with a load bearing column.

00:03:13: yeah, that's a nightmare.

00:03:15: So when he points out this invisible plumbing fix, He's highlighting something that saves more actual project timeline and budget than like any flashy generative AI feature possibly good.

00:03:25: It's fixing the boring infrastructure

00:03:28: Exactly!

00:03:29: And this disconnect isn't just spatial it happens with the actual component data too.

00:03:35: I want to bring in a point Gertjuan Adelman highlighted regarding P&ID diagrams

00:03:39: The piping-in instrumentation diagrams?

00:03:42: Right those and how they relate to three-D models.

00:03:45: The diagram dictates what a pipe does, the flow rate, pressure where it connects.

00:03:51: but the three D model dictates were that pipe physically sits in building.

00:03:54: So are describing exact same physical pipes But two completely siloed environments?

00:04:01: Yes And keeping those two realities connected is usually just a very fragile manually entered tag number

00:04:09: two people who have drifted apart into marriage.

00:04:11: I mean, they are talking about the exact same thing just using completely different words and nobody realizes there's a total breakdown in communication until the divorce papers or printed out late on a table.

00:04:22: that is a brutally accurate analogy.

00:04:25: right if a designer accidentally renames or deletes That tag-in one system The connection just vanishes.

00:04:31: the three D software won't throw an error it Just silently breaks

00:04:35: which terrifying from a quality control standpoint.

00:04:38: A three D clash detection tool will sound the alarm instantly if two pipes occupy this same physical space, but we lack robust systems that raise a red flag when to different engineering documents describe the exact same pipe in fundamentally different ways.

00:04:53: and That lack of cohesion it directly feeds into a massive behavioral problem.

00:04:57: We see how teams actually build these models.

00:05:01: Milgin on Jalevich called out his epidemic over modeling.

00:05:05: Oh, the over-modeling trap.

00:05:07: Yes when offices transition to BIM software there's this overwhelming temptation for beginners to model absolutely everything.

00:05:14: they'll spend hours modeling the exact profile of a skirting board every single bracket Every tiny piece of decorative trim.

00:05:21: because you know placing three D geometry on a screen gives me This dopamine hit of visual progress

00:05:26: exactly and it looks impressive to a client.

00:05:28: But it completely slows down the actual purpose of model, which is generating readable documentation for people building things.

00:05:35: Think about what happens when you take a hyper-detailed architectural model and print out on an one to hundred scale PDF.

00:05:41: It's a mess!

00:05:42: All those perfectly modeled brackets and trims just bleed together into giant useless black smudge of ink... What actually matters is structured information….

00:05:51: That's crucial distinction right there.

00:05:53: A simple basic geometric wall that carries embedded metadata, like it's fire-rating.

00:05:59: It's acoustic glass its manufactured details.

00:06:01: That is infinitely more valuable to a contractor than a photorealistic wall that basically just a hollow shell.

00:06:08: Exactly structured information dictates reality.

00:06:12: excess geometry Just slows down the computer.

00:06:14: Yeah.

00:06:15: And if we pull back and look at the bigger picture of these interoperability failures it completely explains the sheer frustration in structural engineering right now.

00:06:24: Marco Goretta pointed out that structural engineers are trapped in this absurd cycle where they're essentially drawing the exact same building three separate times.

00:06:34: Wait, walk me through how actually happens at a modern workflow?

00:06:38: Okay so... They get their initial plans and draw the structural layout to Tutti AutoCAD.

00:06:44: Okay, then they have to manually rebuild that entire structure inside finite element method software or FIM To actually run the physics and stress analysis

00:06:53: right?

00:06:53: The physics check.

00:06:54: Yeah And once the physics checkout They have to redraw this structure a third time Inside a BIM tool like Revit just to coordinate with the architects.

00:07:03: three times

00:07:05: Three times because exporting a complex structural model between those platforms usually results in such a garbled mess of broken data.

00:07:12: It's genuinely faster for the engineer to just start from scratch.

00:07:16: Yeah, that is a massive failure of the software ecosystem

00:07:19: tool-of-failure.

00:07:20: but The author actually built a plugin called mg struct To bypass this entirely.

00:07:25: it takes a single two D Auto CAD drawing Recognizes what a wall is versus a slab or an opening and automatically generates A pristine three d structural model That perfectly ports into both the physics software And the vim software.

00:07:39: oh wow Yeah,

00:07:41: you define the load bearing structure once and data just flows seamlessly.

00:07:46: But the fact that an independent engineer had to build a custom plug-in To achieve what should be the baseline standard for entire industry?

00:07:53: I mean... That's incredibly telling!

00:07:55: It really is.

00:07:56: And it leads to a very uncomfortable reality check that surfaced recently.

00:08:00: Josephine Kleiner moderated the panel at BIMworld Copenhagen, discussing major industry report on European construction.

00:08:07: Oh I saw this!

00:08:08: Yeah they asked construction professionals rate their confidence in their own project data on scale of one-to ten

00:08:13: and resulting number is genuinely hard to stomach.

00:08:17: It was six point seven out of ten.

00:08:19: Yikes!

00:08:19: When she asks audience if anyone would give them perfect ten not single hand went up

00:08:24: Of course not

00:08:25: Because the larger and more complex the project, the lower confidence score dropped.

00:08:30: You simply cannot hide weak data governance when you have dozens of subcontractors and thousands of design files!

00:08:36: Wait I need to stop here... If the industry's own internal confidence in its data is essentially a D+.

00:08:43: Aren't we putting the cart before the horse by obsessing over AI?

00:08:48: How on earth can machine learning model generate anything useful if it's training on a six point

00:08:53: seven?!

00:08:54: is the exact friction point that industry is hitting right now.

00:08:58: And it transitions us perfectly into our second theme, The actual measurable impact of AI when it does work.

00:09:04: Right?

00:09:05: Renata Pinheiro shared findings from a joint MIT and Suffolk construction study.

00:09:10: That directly answers your question...

00:09:19: Wow!

00:09:21: A twenty percent cost reduction on like a hundred million dollar commercial build is staggering.

00:09:26: It is, but the MIT researchers attached a massive caveat.

00:09:31: naturally The absolute prerequisite for those margins is clean, centralized project data.

00:09:36: You cannot reduce what you do not

00:09:38: know.".

00:09:38: Make sense?

00:09:39: "...the company seeing those twenty percent gains aren't just buying off the shelf AI.

00:09:43: they spent the last five years painstakingly cleaning their data architecture so that the AI actually has a reliable foundation to analyze...".

00:09:51: Which makes sense when we look at the sheer velocity of these tools can produce if it's structured correctly... Victor A ran an experiment using a new tool called Astra.

00:10:01: Okay, what did it do?

00:10:03: He feted some basic site screenshots and a detailed text brief for a project.

00:10:07: And in under thirty minutes without ever opening traditional CAD software the AI generated fifty-two preliminary technical sheets and nineteen presentation sheets.

00:10:18: Fifty two technical sheets in half an hour.

00:10:20: Yeah That represents weeks of human drafting time.

00:10:23: Now obviously a massive package generated that fast requires heavy engineering review.

00:10:28: A licensed human professional still has to validate the code compliance and stamp the drawing.

00:10:32: Sure, but it completely shatters The traditional bottleneck of early stage conceptual design.

00:10:39: It really shifts the human role from creator To editor exactly.

00:10:43: But succeeding in that new roll requires a completely different skill set.

00:10:48: Adeline Chan brought up an incredible concept called token optimization.

00:10:51: token optimization

00:10:53: Yeah, she shared an example of a designer using a model called DeepSeq to draft a forty by forty meter hotel CAD plan.

00:11:00: The total compute cost for generating that entire plan was exactly one US dollar.

00:11:06: Wait!

00:11:07: A ONE DOLLAR HOTEL PLAN sounds like magic.

00:11:11: How are they actually controlling the AI without it hallucinating a buncha useless geometry?

00:11:16: That's where token optimization comes in.

00:11:18: So every time you ask an AI to process or generate information.

00:11:22: It costs tokens, which basically translate to computing power and money.

00:11:26: Right?

00:11:26: This designer didn't just type design a hotel.

00:11:29: they gave incredibly rigid mathematical instructions specifying a strict five by five column grid with an eight meter span.

00:11:36: Okay when the AI inevitably made a slight error in attempted four-by-four grid The designer caught it instantly and course corrected prompt.

00:11:44: They knew exactly how to steer the machine to patch the drawings without wasting tokens on endless regenerations.

00:11:50: Oh, I see.

00:11:51: So the real survival skill for a digital construction professional isn't just buying an AI license.

00:11:57: it's understanding how to architect the workflow to minimize the compute costs while maximizing the precision of the output

00:12:03: Precisely!

00:12:05: And hey speaking of optimizing workflows if you are finding this deep dive valuable for your own strategy Just make sure hit subscribe so that we don't miss our future breakdowns.

00:12:14: We release these specifically to help navigate those exact kinds industry shifts.

00:12:19: Yes definitely subscribe because understanding these shifts explains why some companies are pulling so far ahead.

00:12:26: Like Sarah Buckner recently laid out a framework for AI adoption, arguing there seven distinct stages of AI maturity in construction.

00:12:34: Seven stages okay?

00:12:36: Yeah

00:12:36: she noted that the vast majority of general contractors right now or stuck-in stages two or three

00:12:42: which usually looks like ad hoc usage right

00:12:46: exactly?

00:12:47: A project manager might use an AI co-pilot to draft an email, or a VDC Manager plays with the plug in on pilot projects.

00:12:55: But field crews usually reject it because they bolted onto their existing workflow meaning that just feels like extra administrative work.

00:13:03: It's just a nuisance for them.

00:13:04: Meanwhile market leaders have already pushed into stages five and six.

00:13:08: For them, AI is no longer separate application.

00:13:11: you open up your desktop.

00:13:13: It is the invisible engine that shapes how they bid for jobs, how they calculate their risk margins and exactly who they hire.

00:13:21: It's just baked in?

00:13:22: Yes

00:13:23: over eighty percent of workforce interacts with it daily largely without even realizing.

00:13:28: but breaking through that barrier from stage three to stage five is incredibly difficult because of the nature of technology itself.

00:13:36: Elan Alexander Radkin brought up a crucial technical limitation.

00:13:40: Oh what.

00:13:41: Large language models, the tech behind chat GPT and most current AI struggle massively with construction data.

00:13:48: LLMs are probabilistic text engines.

00:13:51: they're designed to guess the most likely next word in a sentence.

00:13:54: But construction isn't a novel.

00:13:57: it's its spatial mathematical reality.

00:13:59: I mean a load-bearing steel beam either exists at coordinate X or doesn't.

00:14:04: Exactly You cannot have a probabilistic approach to structural engineering.

00:14:09: The author points out that the industry desperately needs what he calls, a deterministic data infrastructure.

00:14:15: We need systems that operate on absolute mathematical certainty with one hundred percent data.

00:14:20: fidelity

00:14:21: Makes perfect sense

00:14:22: Because until we build that layer... ...we are essentially asking very advanced text predictor To solve multi-dimensional spatial engineering problems.

00:14:30: And even Kalanathi echoed that exact sentiment from a physical perspective.

00:14:35: Construction happens in the dirt, In The Rain on dynamic job sites.

00:14:39: they change shape every single day.

00:14:41: It's messy

00:14:42: Very!

00:14:42: If we are going to deploy AI agents To actually manage a site They cannot just be text-based chat windows sitting On A server...They need visual and spatial awareness..they need eyes Right.

00:14:54: They have to comprehend the physical reality of space they're analyzing

00:14:58: which forces us to ask the uncomfortable question that Vincent Poon raised.

00:15:02: If we take these incredibly powerful AI agents and plug them into the fragmented, six point seven out of ten data infrastructure we currently have aren't we just automating a broken process so it fails faster?

00:15:16: That is such a good point!

00:15:17: if the AI hallucinated design flaw that causes a clash on job site who was actually legally or financially accountable for this outcome?

00:15:26: That is the multi-million dollar question.

00:15:28: And it forces us out of a theoretical technology and straight into our final theme, The Brutal Business Mechanics Of The Built Environment.

00:15:35: The money part?

00:15:37: Yeah!

00:15:37: Because the technology's clearly arriving but... ...the traditional business models of construction are actively fighting its implementation.

00:15:45: It all boils down to who pays for innovation Who absorbs risk and who actually swings hammer.

00:15:52: AJ Waters broke down the reality of AI stakeholders brilliantly.

00:15:55: He argues there are four distinct groups fighting for control over construction AI right now.

00:16:00: Okay, who are they?

00:16:02: You have the software vendors pushing the licenses The GC executives trying to protect their profit margins... ...the IT departments terrified of data security and the field crews Who have to figure out if this tool actually helps them pour concrete faster.

00:16:15: So four groups with completely conflicting incentives

00:16:18: Exactly.

00:16:19: And here's the kicker.

00:16:21: All four of those groups walk away.

00:16:22: the second the ribbon is cut on The New Building.

00:16:24: Oh, wow!

00:16:25: The owner—the entity that actually has to operate maintain and live with both a physical asset And digital data for next fifty years Is almost entirely missing from conversation.

00:16:36: That's

00:16:37: so true.

00:16:38: We are optimizing our technology For temporary two-year construction schedule Instead building a digital twin for A fifty year asset life cycle.

00:16:46: That misalignment is crazy.

00:16:48: And when you look at the financial reality of this, subcontractors.

00:16:51: You know that people actually executing work The dysfunction gets even worse.

00:16:56: Oh absolutely

00:16:57: Fabio Bronzine wrote a phenomenal piece analyzing the economics of retainage.

00:17:02: Subcontractor's typically operate on razor thin net margins usually around six to eight percent.

00:17:08: Okay Yet it is standard legal practice for general contractor To withhold ten percent Of every payment until the entire massive project officially closes out.

00:17:18: Wait, I want to make sure that math on those lines... The amount of cash being legally withheld from a subcontractor is larger than the entire profit margin they will make on-the-job.

00:17:27: Exactly!

00:17:28: They are basically paying to build this building.

00:17:31: Added to it, a fifty six day average wait time for payment after an invoice was submitted?

00:17:36: That's

00:17:36: insane!!

00:17:37: It is the slowest collection cycle in any major industry on earth.

00:17:41: Subcontractors are essentially acting as zero percent APR involuntary, unsecured lenders to the multi-trillion dollar construction supply chain.

00:17:49: Wow!

00:17:50: They front the cash for the labor they front a cache from materials and just wait.

00:17:55: You can hand that subcontractor at most advanced predictive AI dashboard on planet.

00:18:00: but if their cash flow is trapped in retainage all this dashboard will tell them exactly how broke it was on day.

00:18:06: fifty six

00:18:07: Software cannot fix a fundamentally broken cash flow model.

00:18:11: Nope!

00:18:11: And this economic friction directly explains why we are seeing so much trouble pricing these new digital services.

00:18:17: Rogan Parantham analyzed the push for outcome-based pricing in AEC design.

00:18:21: What did he find?

00:18:23: Well, everyone wants to stop billing by hour and start billing for value of final design.

00:18:27: But there two massive structural traps.

00:18:30: First is verification deficit.

00:18:32: Okay AI design tools are great, but edge cases still require about twenty percent human remediation.

00:18:38: There is no automated instantaneous way to definitively prove that.

00:18:41: an AI-generated design Exactly.

00:18:53: And the second issue is what the author calls, The Co-Production Tram In construction design... ...the architect doesn't create final product in a vacuum.

00:19:01: They co-produce it with client through endless feedback loops.

00:19:05: Oh!

00:19:05: The dreaded feedback loops

00:19:06: Right.

00:19:07: So if your new AI tool cuts a design task from ten hours down to two hours, but the client goes on vacation and delays the review process by three weeks.

00:19:16: The vendor's massive productivity gains completely vanish from the P&L statement... That

00:19:21: makes total sense!

00:19:22: ...the operational efficiency is swallowed whole by human delays completely out of their control.

00:19:27: The software did its job—the productivity gain was real —but economic value simply evaporated….

00:19:34: Yeah

00:19:35: Yet despite all these massive structural bottlenecks, the big capital markets clearly see a path through.

00:19:41: Yaqir Sudri highlighted that build-outs recently raised a staggering hundred and thirty million dollars for their AI driven data platform.

00:19:49: And they aren't raising money to just build another simple dashboard.

00:19:54: Their goal is absolute data consolidation.

00:19:56: They want to pull in every single data set that influences a project like contracts, three D models budgets even weather patterns.

00:20:03: so the platform's analytical judgment effectively mirrors intuition of a thirty year veteran superintendent.

00:20:09: They are trying digitize human experience

00:20:12: pretty much.

00:20:13: but maybe the most shocking capital movement we saw isn't going into software at all.

00:20:18: Diana K highlighted massive development.

00:20:21: big tech companies realizing they have physical world problem meta Google, Lowe's.

00:20:27: These massive corporations have committed over five hundred and forty-five million dollars towards skilled construction trades training.

00:20:34: Wait tech companies are funding plumbing in carpentry schools?

00:20:37: Yes because they looked at their roadmaps and realized They can spend billions designing the most advanced AI data centers in the world.

00:20:44: But they still need human beings to physically pour the concrete slabs And run high voltage electrical conduit.

00:20:50: That is a huge shift.

00:20:51: Google alone is targeting three hundred thousand skilled workers with their initiatives.

00:20:55: Lowe's committed two hundred and fifty million dollars, they are slamming head first into the hard constraint of the physical world which a massive labor shortage

00:21:04: Which brings us to an incredibly profound shift in power dynamics.

00:21:07: I mean If the traditional AEC industry's margins are so heavily trapped in broken systems like retainage that they can't afford to innovate, and Silicon Valley is stepping with half a billion dollars to literally train next generation of construction workers who will actually own the future.

00:21:26: We spent this entire deep dive talking about optimizing design but building takes two years constructs.

00:21:36: I want you to ask yourself, if your company's digital and AI strategy doesn't ultimately serve the person holding the keys on day seven hundred of that building life.

00:21:45: Are you actually innovating or are you just building a really expensive temporary dashboard?

00:21:50: If you enjoyed this episode new episodes drop every two weeks.

00:21:54: also check out our other editions On Smart Manufacturing & Industrial AI And Connected Tools & Equipment.

00:21:59: Thank You so much for taking time to join us today.

00:22:01: Remember To Hit That Subscribe Button.

00:22:03: Keep questioning the process on your own projects, and we will catch you in our next deep dive.