Every night, construction crews go home and a robot takes over. It walks the site, scans everything at millimetre accuracy, and produces a report by morning. The technology works. The question is whether anyone opens the file.

This week we spoke with Chinn Lim, founder of dConstruct.ai, a Singapore-based AI robotics company working across site inspection, progress monitoring, and facility management. Two weeks ago, in The AI You Buy Isn't the Advantage, we featured his point that the data layer is the real 10-year moat in construction. This edition goes a level deeper: what actually happens when robots arrive on site, and why adoption is failing on grounds that have nothing to do with capability.

Chinn is one of the smartest people we’ve spoken to on Bricks and Bytes. His entrepreneurial success began when he sold his first startup to Autodesk. After that, he was poached by the Singapore Government to lead their Smart Cities initiatives.

Now he is back, with his new robotics company dConstruct.

And trust us when we say you will know A LOT more about this company soon.

Ask Chinn whether a robot can do a site inspector's job today, and the answer is yes, with a qualification that cuts to the core of the adoption problem.

A robot scanning a site at one to three centimetre accuracy does not just find problems; it finds everything. That includes deviations that are perfectly within engineering tolerance but that a financier or property owner, confronted with a deviation flag, may not interpret with the same nuance. The report becomes evidence. Contractors, quite rationally, are wary of evidence.

That is not technophobia. It is a risk calculation made by people who understand how disputes actually unfold on construction projects. The legal and contractual frameworks governing construction have not caught up with what LiDAR can see.

The counterweight is timing. Chinn's framing: when the construction team goes home, let the robot do its thing. The next morning you get a report, errors are flagged, and you can correct them before they compound. Used as a partner to the process rather than a governance instrument, the data serves the contractor as much as it does the client.

The Building and Construction Authority projected construction demand of between SGD 47 and 53 billion in 2025, up from SGD 44.2 billion in 2024, with major contracts for Changi Airport Terminal 5 and Las Vegas Sands' USD 8 billion integrated resort still in the pipeline. At the same time, the government tightened foreign worker quotas by 4.2% from January 2024. Corestaff puts the sector's workforce shortfall at close to 15%, with safety officers and site engineers the hardest roles to fill.

That combination is a policy, not a temporary imbalance. Singapore is deliberately engineering the conditions that make robotics adoption structurally necessary, which is why it is the most instructive live test case for physical AI in construction anywhere in the world. Chinn's choice to build dConstruct there is not incidental.

Worth noting: dConstruct's CTO spent over a decade at Pixar on films including Wall-E and Toy Story 3. The mathematics of animating movement and navigating a robot through an unstructured environment are closer than most people assume.

Chinn is more useful than most on this because he does not oversell it.

What dCconstruct does well: design-versus-built comparison at one-to-three-centimetre accuracy using LiDAR, superimposed on BIM to show progress and flag deviations. Safety monitoring in parallel: PPE compliance, intrusion detection, blocked walkways, unattended items. Facility management rounds across large estates, identifying cracks, spalling, and meter readings at a frequency no skeleton crew can match.

One capability that stands out is object search. Since the environment has been fully scanned, the system can locate a specific fire extinguisher, track its movement, and flag any discrepancies. A Google search for the physical world, in Chinn's framing.

Bricklaying and painting robots exist, but they require structured environments that most sites do not provide. In Singapore, prefabricated construction dominates, making those use cases largely irrelevant. Focusing on inspection involved disciplined market selection.

Most construction companies have no robotics department. Chinn's procurement guidance, drawn from what he has seen go wrong:

Chinn closed with an analogy worth sitting with. Before the elevator, the ground floor was prime real estate because nobody wanted to climb. The elevator made height accessible, buildings grew taller, and the penthouse became the premium. The physical infrastructure of movement reshaped the economics of space.

Autonomous vehicles are horizontal elevators. When moving across a city requires no active effort, the logic of the central business district weakens. The same shift, applied horizontally.

The construction implication is more immediate. Buildings designed for robot accessibility, with lift interfaces, door protocols, and floor tolerances that allow autonomous navigation, are a design brief that does not yet exist in most practices. Virginia Tech and Procon's MARIO project, published in March 2026, is already coordinating humanoid, quadruped, and aerial robots across sites under single-operator supervision. The multi-robot future is being built in labs now. The architects designing buildings today are not yet designing for it.

The overnight scan works. The question is whether the commercial and contractual environment has evolved enough to use that accuracy without it becoming a liability. 'Not yet' is the honest answer. However, the firms that are currently collecting data are establishing a position that continues to grow. The ones waiting for certainty will be buying someone else's history at a premium.

Bricks & Bytes covers the people and ideas shaping construction's transformation. Forward this to a colleague or find the full archive at bricks-bytes.beehiiv.com.

POWERED BY: