AI's Double-Edged Sword: Construction And VC

May 20, 2025

AI reshapes construction and venture capital by optimizing processes, but struggles to replace the physical nature of building and the human intuition behind non-consensus investing.

Patric Hellermann

Shub Bhattacharya

AI's Double-Edged Sword: Construction And VC

The physics of building

Construction is about atoms, not algorithms, so AI can enhance workflows but won't replace physical building

The conversation kicks off with a first-principles argument that's grounded in physical reality: construction is fundamentally about atoms. It's about moving and stacking physical materials. And while AI continues to advance, those atoms still need to be physically manipulated.

Construction involves moving atoms, stacking atoms, and building infrastructure that people need. This observation helps put some of the AI hype in construction tech into perspective.

Short of a Star Trek-like "replicator" that could reorganize matter at the atomic level (which might require artificial superintelligence, not just AGI), AI may struggle to fundamentally replace the physical nature of construction. AGI might excel at rapidly computing or taking decisions off of data, but AI itself isn't likely to be doing the actual production.

This distinction is important. While AI continues to transform how we plan, design, and manage construction projects, the core activity - the physical building process - largely remains bound by the laws of physics.

Until the rules of physics can be bent in a different way or replaced by another set of rules, we will be bound by these physical constraints. This suggests the construction industry might not face the same level of existential threat from AI that some other industries potentially could.

What seems more likely is AI enhancing and optimizing the processes that surround construction, potentially making them more efficient, while not replacing the fundamental need for physical work. It might change aspects of how we decide what to build and how to build it, but something still needs to stack those atoms.

The physical nature of construction could provide something of a natural moat against complete AI disruption that knowledge-only industries might not have.

Trust Networks in Construction

AI might eliminate intermediaries in construction supply chains but won't solve fundamental trust issues

The discussion shifts to how AI might impact the many intermediaries in construction - those people who manage relationships, check documents, and generally keep the complex supply chains and production lines functioning.

AI could potentially diminish the importance of intermediaries who manage "hyper-local relationships" in construction supply chains. These are the people who know exactly who to call when something specific happens - the human decision matrices that embody tribal knowledge.

But there's an interesting counterpoint: it's a trust problem. You could put all stakeholders on the same platform, but that would still not solve the trust issue.

This gets to what might be a core reason why construction has been somewhat resistant to technological disruption. The industry isn't inefficient merely because of poor communication or non-standardized software. It's inefficient partly because construction involves significant financial risk, complex interdependencies, and a need for trust and track records that might not simply be solved by better information flow.

It would still not solve trust and track record issues. It would also still not solve working capital and inventory challenges for just-in-time delivery.

This perspective offers something to consider. While AI might help orchestrate processes better, perhaps through more automated marketplaces, it may face challenges with the fundamental human elements of trust and reputation that help keep construction functioning.

The conversation takes an interesting turn when discussing document checking - a task that seems potentially suitable for AI automation. In many setups, the reason why there's still so much inertia against replacing it with technology or automation or AI is because you sort of always want a person to catch hold of if something goes wrong.

This appears especially relevant in markets like the US, where the liability system often requires contractors to independently verify work to be held liable. This creates a situation where certain inefficiencies persist because humans may need to remain in the accountability loop.

AI might optimize information flow in construction, but the industry's challenges around trust, liability, and working capital could require more than just algorithmic solutions.

VC's human edge

Non-consensus early-stage VC investing might be the last refuge from AI disruption in capital allocation

The conversation eventually turns to venture capital itself and how AI might affect this industry.

VC can be broken down to its essential function: early-stage VC exists because otherwise the acquisition of capital would be very opaque and complex. VCs are like gas stations for oil companies - they help allocate capital to what they believe are high-quality companies.

The real differentiator is the individual partner a founder chooses to work with. The brand is the vessel to inexperienced founders or founders who are not as prepared to make a good capital decision... but the true vessel is always the person.

This sets up an interesting question: If AI could facilitate effective capital allocation between startups and those with capital, might that potentially make some aspects of VC less necessary?

One compelling perspective is that it's been proven repeatedly that the best returns come from non-consensus bets. And by definition, AI, which is trained on existing data and patterns, might possibly struggle with truly non-consensus investing.

Non-consensus investing might be difficult to power with AI because it goes against what AI is supposed to represent. Non-consensus is supposed to be idiosyncratic, unscalable, etc.

Where human VCs might maintain an edge is in their ability to recognize and value things they haven't seen before - precisely what statistics-based AI models could struggle with. AI systems tend to bias toward what has worked in the past, while some of the best VC investments often come from spotting something entirely new.

While much VC capital already follows consensus patterns that could potentially be replicated by AI, the exceptional returns often come from those contrarian bets that may require human intuition and comfort with the unprecedented.

Growth-stage investing, where companies are more "legible" through their metrics and performance, might be somewhat more vulnerable to AI disruption than early-stage investing where human judgment of founders and novel ideas potentially plays a larger role.

The defense against AI disruption in early-stage VC isn't just being contrarian for its own sake, but building organizational structures that enable and reward thinking that diverges from consensus views.

Conclusion

As we consider the potential impacts of AI on both construction and venture capital, a pattern emerges around what might be called a Physical-Digital Equilibrium Framework. This approach suggests that AI will likely transform many decision-making processes in construction and capital allocation, but may face limitations when confronting physical realities or highly subjective human judgment calls. The areas that could prove most resilient to AI disruption appear to combine physical constraints with human trust relationships that AI might struggle to fully replicate or replace. For organizations looking to thrive in this emerging landscape, identifying their core atoms and trust elements while strategically leveraging AI to optimize surrounding processes could create sustainable competitive advantages. Both construction tech and venture capital will likely find a new balance where AI handles the more predictable patterns while humans continue to manage exceptions, innovations, and the physical world. This framework isn't about resisting AI adoption but rather about understanding its natural boundaries and complementary relationship with human capabilities.