How AI is redefining work in construction: the tools are ready, the data is not

Who this is for

• You have seen the demos and want an honest account of why the results are not showing up on your projects.
• You are deciding what to train people on, and need the difference between the project manager's competencies and the BIM coordinator's.
• You are trying to get subcontractor data into a central system without asking anyone to expose their pricing.

What the episode says

Transcript

Transcribed from the recording and lightly edited: mis-heard technical terms and broken sentence boundaries are corrected, nothing said was changed or added to. Views expressed are the presenters’ own. The voices in the recording are synthesised with AI, produced with Kamilla Drożdż; the words are the presenters’.

Welcome to Tylko Advisors. We empower construction material manufacturers with cutting-edge digital solutions. And now, welcome to our podcast, where we share insights from the world of smart construction technologies. Frantically rewriting spreadsheets, calling subcontractors, and just trying to salvage a timeline that is completely falling apart.

Just running on coffee in pure panic, yeah. Exactly. But today, An AI algorithm looks at that mud, calculates 10,000 alternate futures, and spits out a revised, perfectly optimized schedule in less than 60 seconds. I mean, the era of the digital hard hat is actually here. It is. And that 60-second recalculation is, it's completely changing what it means to be a professional in this space.

I mean, we're moving way past the novelty of chatbots. Yeah, you're talking real stakes now. Right. When an algorithm is dictating the critical path of a half billion dollar infrastructure project, The human beings running that project, they have to possess a completely different set of skills than they did even, say, five years ago.

And that is exactly our mission for you today. If you're anywhere in the architecture, engineering, and construction industry, the AEC sector, we're providing you with a shortcut to becoming an AI-ready professional. Which is so critical right now. It really is. We're going to unpack how artificial intelligence is actively rewriting job descriptions across building information modeling, project management, and just daily construction operations.

And to do that in this deep dive, we have to start by looking at what this actually looks like in the dirt and the steel. Yeah, because before you can adapt your skills, you have to understand, you know, the mechanics of the tools you're actually adapting to. Take that delayed steel scenario you mentioned. Historically, adjusting a schedule meant dragging bars around on a Gantt chart and relying on human intuition to guess the cascading impacts on I don't know, the plumbers, the electricians, the drywallers.

Just hoping you didn't forget anybody. Exactly. But now we have generative scheduling platforms like Alice Technologies. See, generative scheduling sounds like having a chess grandmaster on site who can foresee 10,000 moves ahead when that steel delivery is delayed, rather than a human just, you know, shifting a couple of blocks on a spreadsheet.

A chess grandmaster is close, but it's actually messier than chess. Really? How so? in chess, the board never changes shape, right? In construction, Alice is playing a game where the rules change every hour based on weather, supply chains, labor shortages. It uses this constraint-based logic and Monte Carlo simulations.

Okay, so you're feeding it the rules of the site. Right, you tell it a crane can only swing this fast, concrete needs this many hours to cure, this crew can only work 8 hour shifts, and the algorithm maps out thousands of fluid alternate realities to find the mathematical path of least resistance. Wow, so it basically turns a static timeline into a living organism.

Exactly. And this shift from static to living, it's happening in design too. I mean, think about clash detection in a 3D BIM model. In the past, an engineer would manually cross-reference architectural drawings with mechanical drawings, right? Just to make sure an HVAC duct didn't run straight through a load-bearing steel beam.

Which was incredibly tedious. Oh, completely. Now AI parses the geometric data of those models, finds the spatial conflicts, and resolves them mathematically before a single shovel even hits the dirt. Yeah, and then you take that conflict-free digital model and link it to the physical world using a digital twin. Right, the digital twins.

Digital Twin isn't just like a pretty 3D rendering, it's a virtual replica wired up to real-time telemetry. So it's actively pulling data from the site. Constantly. You have strain gauges embedded in the actual concrete, temperature sensors in the HVAC, GPS trackers on the heavy equipment, all constantly feeding data back to the digital model.

The AI analyzes that incoming telemetry to predict when a component will fail before it actually breaks. It's basically giving the building a nervous system. That's a great way to put it. And moving on onto the active site, the computer vision applications are just wild. Tools like SmartVid.io are acting like 1000 extra sets of eyes.

They ingest live video feeds from job site cameras and use edge detection algorithms to spot safety risks. The mechanism there is fascinating, actually. The computer vision AI is trained to recognize specific pixel patterns and geometric shapes of, you know, hard hats, safety vests, guardrails. So it knows exactly what a safe worker looks like.

Yes. So if a worker steps within 6 feet of an unprotected elevator shaft, the software doesn't just record a video of it - it calculates the spatial relationship in real time, assigns a risk score, and literally pings the safety manager. Oh, wow. So it shifts safety from a reactive investigation to a proactive intervention.

Exactly. Even the paperwork is getting overhauled. I mean, anyone who has ever worked on a major project knows the absolute misery of digging through 1000 page PDF of engineering specifications just to verify, like a material grade. Oh, yeah, it's the worst. Right. But platforms like civils.ai use natural language processing to ingest mountains of unstructured text contracts, RFIs, specs, and turn them into a highly structured queryable database.

You ask it a question and it extracts the exact compliance clause you need in like 3 seconds. But this brings us to a massive glaring contradiction in the industry right now. Okay, what's that? Well, we just outlined AI tools capable of generating perfect schedules, predicting mechanical failures, spotting safety violations instantly, and reading thousand-page specs in seconds.

So why are major construction projects still plagued by massive cost overruns and chaotic delays? right. It's because the industry is currently tripping over its own fragmented data. If you peel back the curtain, you realize these brilliant AI tools are starving. I mean, AI fundamentally requires clean, structured, standardized data to learn and make accurate predictions.

But AEC data is notoriously siloed. It's everywhere. Right. The project managers will live in their spreadsheets. The architects are in their design software. The supply chain is running on a totally different ERP system. And literally none of these pipelines talk to each other. And when data is trapped in inconsistent formats like that, it's completely useless for training machine learning models.

You just can't read it. Right. Imagine trying to train an algorithm to forecast the material costs of a new hospital. If the data from your last five hospital projects categorizes structural steel 5 completely different ways, the AI cannot identify the pattern. It just sees noise. And it's not just formatting either. It's the physical reality of the environment literally destroying the data collection.

Oh, for sure. Look at the mega projects happening in the Middle East right now. They're trying to deploy these cutting edge computer vision safety systems, but they have these massive dust storms. and the dust just degrades the lenses on the site cameras. Which breaks the underlying mechanism of the AI entirely. Really, just from the dust.

Yeah, because computer vision relies on high contrast edge detection to map three-dimensional space. When a dust storm coats the lens or fills the air, that contrast drops to zero. The image just becomes a flat, blurry mess. Oh, I see. So the algorithm can no longer distinguish between a human worker and, say, a piece of scaffolding.

So it starts missing severe safety violations entirely. It's the ultimate illustration of garbage in, garbage out. Exactly. Another massive failure point is the lack of as-built updates. You start with this pristine, geometrically perfect BIM file. during the design phase. But once you're in the mud, things change. They always do.

Right. A crew hits unexpected bedrock and has to pour a footer 3 feet to the left of where it was designed. If they don't update the central BIM model to reflect that physical change, the digital twin is now just a fantasy. And if you then feed that outdated fantasy BIM model into an AI to calculate future material needs or structural load distribution, the outputs are going to be dangerously incorrect.

So here's a question then. If an AI model gives a project manager bad advice on resource allocation because it was trained on an out-of-date BIM file that a human forgot to update, whose fault is that? I mean, is the AI making a bad call, or is this just exposing our own sloppy data habits? it exposes the human element entirely.

The industry suffers from a severe lack of digital maturity. When teams don't enforce strict information architecture, they inevitably get burned by bad AI insights. And when the insights are bad, the field crews lose trust in the automated systems and the entire technology initiative just collapses. The bottleneck isn't the software, it's data stewardship.

Which means the human workforce has to fundamentally evolve. They do. To bridge that gap between our messy reality and these algorithms, construction professionals need a completely new set of core competencies. I mean, if you're listening to this, your job is no longer just moving dirt or checking geometric alignments.

You are essentially becoming a data auditor. And the absolute foundation of that is data literacy. You have to stop viewing data as just a passive byproduct of building a wall. Data is a highly valuable strategic asset. You need to understand how structured information flows through a common data environment, or CDE. And A CDE isn't just like a glorified cloud folder where you dump PDFs, right?

Well, not at all. It has to be treated as the living, breathing central nervous system of the project, where every single revision is tracked and validated. You have to understand the workflow of the tools feeding into it. Take a platform like Buildots, for example. Oh, they're doing great stuff. They mount 360-degree cameras on the hard hats of the site superintendents.

As the super walks the site, the camera captures raw visual data. But you need to understand what happens next. The software uses photogrammetry to stitch those 2D pixels into a highly accurate 3D point cloud. Right. It then overlays that point cloud onto the 4D BIM schedule to automatically flag variances. If you don't understand how that data pipeline translates a photograph into a schedule variance, you cannot blindly trust the output.

Which brings us to the most critical skill of all. Interpreting AI-generated outputs. It is not enough to just know which buttons to click. You have to apply Human-in-the-Loop, or HITL, oversight. Consider an AI system monitoring the structural health of a bridge. Let's say the AI detects structural fatigue. It flags a crack in a load-bearing component.

Okay, so it sees a problem. Right. And based on its pattern recognition of previous cracks, it automatically generates a work order for a superficial epoxy repair. So it just assumes it's a surface issue. Exactly. But an engineer with strong human-in-the-loop oversight does not just blindly approve that work order. They physically investigate the crack, analyze the surrounding telemetry, and realize the crack is actually a secondary symptom of a much deeper hidden foundational shift.

Oh, wow. Trusting an AI output without human-in-the-loop oversight is kind of like putting a brick on the gas pedal of a heavy excavator and just hoping it digs a perfect trench. That is a terrifying analogy, but yes. Right. You don't need to know how to build the excavator's engine from scratch or how to write the code for its hydraulic sensors, but you absolutely must know how to steer the tracks, read the dashboard, and know exactly when to hit the emergency kill switch.

That mindset is absolutely mandatory. But how that mindset is applied depends heavily on your specific role. The technical depth required for a project manager is very different from what is required of a design engineer. Let's break those roles down a bit. If you're a project manager listening right now, you're what the competency frameworks classify as an AI leader.

Yes, the AI leader. Your required skills are actually highly non-technical. You aren't sitting there writing Python scripts. Your job is orchestration. You need to identify which AI solutions are actually viable for the specific constraints of your site. You need to use the AI's output for dynamic resource optimization, all while managing the expectations of stakeholders who think AI is just like a magic wand.

Right. A project manager uses the AI to manage the complexity of the math so they can free up their time to manage the people. Which makes sense. But on the other end of the spectrum, you have BIM coordinators and systems engineers. These are the AI professionals. Their required skill set is deeply technical. They are the ones deep in it. Yes.

They need to understand the underlying architecture of how machine learning actually integrates with BIM. They have to set up the automated metadata extraction, build the data feature models, and rigorously evaluate whether an algorithm is actually performing within strict engineering tolerances. And this highlights a really crucial distinction between what we should automate and where we still desperately need human judgment.

Absolutely. Think about a company like Dusty Robotics. They make these small autonomous rovers that drive around a freshly poured concrete slab and print the floor layouts, the walls, the electrical drops with millimeter precision directly from the digital model. It's amazing to watch. It is. And printing lines on concrete is a highly repetitive manual task.

It relies purely on geometry. Automate it completely, let the rover do it. But compare that to a scenario where a general contractor and a subcontractor are in a heated dispute over, say, a three-week delay. Well, if we look at the trajectory of the industry over the next three to five years, that window for human intervention is shrinking fast.

Right now, we're working alongside AI co-pilots that suggest ideas or flag errors. The next phase is a jump to highly autonomous agents that actually execute decisions on their own. The pace of that shift is just staggering. And to survive that shift into autonomous agents, Data governance and interoperability are going to become the most important skills on the site.

I mean, future professionals will have to champion open data standards. You're going to hear terms like IFC and ISO 19650 constantly. And they're critical. They're essentially the universal translators for construction. Right. IFC ensures that a highly complex concrete column designed in an architect's proprietary software is still recognized as a concrete column with all its specific load-bearing math intact when it is opened by the structural engineer's completely different software.

Without that universal translation, the autonomous agents will just constantly misinterpret the data. Wait, I'm stuck on this idea of data governance across different companies, though. The frameworks talk about enforcing domain data contracts. Yes. If you have 50 different subcontractors on a mega project, plumbers, glaziers, steel workers, how do you mathematically force them to share their data with the central AI without handing over their proprietary trade secrets or pricing models.

I mean, they wouldn't do it. And that is exactly what a domain data contract solves. It's a strict programmatic rule set. It acts almost like a cryptographic filter. Okay, how does that work? Well, it ensures that the Glazier software automatically feeds the precise weight and installation timeline of the glass into the central AI, but it mathematically blocks the central AI from seeing the Glazier's internal profit margins or proprietary manufacturing.

manufacturing techniques. Oh, clever. Right. It allows secure collaboration in a highly competitive zero trust environment. But implementing all of this mathematical governance and AI integration sounds like it requires a massive industry-wide reskilling effort. That's a huge lift. And construction isn't exactly known for pivoting on a dime.

How do construction firms avoid leaving their veteran superintendents behind? I mean, the people who might not know how to write a domain data contract, but who have 30 years of hard-won intuition built up in their heads. It's a great question, but those veterans are actually the most valuable asset in the AI revolution.

Really? Why is that? Because of domain knowledge. You cannot train an AI to understand the nuances of pouring concrete and freezing rain, or how to phase a complex excavation around live city utilities just by feeding it generic textbook data. Machine learning models need the hard-won intuition of experienced builders to actually know what right looks like.

So the AI needs the veteran to teach it the messy reality of the physical world. Precisely. And more importantly, the veteran's domain expertise is the ultimate safeguard against naive AI. Naive AI. Yeah.

A junior engineer who grew up on iPads might look at a flawless, fully rendered AI simulation and trust it implicitly. A 30-year superintendent will look at that exact same simulation and immediately know it will fail, because they know how the local labor force actually moves on a Tuesday afternoon. The machine calculates, but the human understands.

I love that. The machine calculates, the human understands. So let's pull all of this together. Based on the realities of the job site and the mechanics of these algorithms we've been unpacking, what defines the AI-ready construction professional today? It's clearly not someone who sits in a trailer trying to outcode a software engineer.

No, definitely not. It's a hybrid thinker. It's a professional who combines deep construction domain expertise with rigorous data stewardship. Someone who treats data like a physical asset. Exactly. They understand that clean data is just just as critical a building material as steel or concrete. They provide uncompromising safety oversight, and they have the collaborative mindset to orchestrate a site where heavy machinery, autonomous robots, and human crews are all interacting constantly.

They're essentially the translators. They take the mathematically perfect digital ideal, and they know how to force the AI to adapt when it hits the actual mud of the physical job site. The tools will continue to evolve, absolutely, but the need for human accountability, physical context, and critical engineering judgment is only going to increase.

Well, if you take one thing away from our deep dive today, let it be that future-proofing your career in AEC means mastering the flow of information just as well as you master the physical materials. Couldn't agree more. I want to leave you with a final thought to ponder as you look at the future of your own projects. If artificial intelligence eventually standardizes the perfect building process, if it totally optimizes the generative scheduling, eliminates every geometric clash, and automates all the safety monitoring, will the construction companies of the future compete not on how well they actually pour concrete, but entirely on the quality, the depth, and the exclusivity of their proprietary data?

Think about it. That's it for today's episode. We hope you enjoyed exploring how digital workflows are transforming the construction industry. Don't forget to subscribe and join us next time for more insights into building smarter. Head over to our website. You'll find everything you need to keep building smarter. To give you the best listening experience, our audio is enhanced with the support of AI technology.