AI Isn't The Bathtub, It's Just Running Water

April 16, 2025

AI is like water—it flows freely, but without a vessel, it's wasted. In construction tech, AI features alone aren’t enough. The real winners will build the “bathtubs”: robust tools and data systems to capture, refine, and scale AI's value.

Patric Hellermann

Shub Bhattacharya

AI Isn't The Bathtub, It's Just Running Water

Ever tried to catch water with your bare hands? No matter how tightly you cup them, most of it still slips away. That's exactly what's happening with AI in construction tech right now. Everyone's excited about the water flowing from the tap, but few are thinking about the vessel needed to make it truly useful.

This Week On Practical Nerds - tl;dr

AI features aren't enough without systems to contain them

Authoring tools and data infrastructure are harder to build than AI

Companies owning the "bathtub" will win over pure AI feature players

🎧 Listen To This Practical Nerds Episode

AI features aren't enough without systems to contain them

Why are data infrastructure companies more valuable than AI features?

The world of AI feature companies has exploded since early 2022. Everyone's rushing to incorporate AI into their offerings, but there's a fundamental insight being overlooked. When it comes to complex construction documents, designs, or processes, the first AI-generated version is never the final version.

Think about what happens when you ask an AI to create construction documents or 3D models. You get a statistical representation of what might work—but it's rarely exactly what you need. The real value emerges in the editing phase. And here's where things get interesting: is that editing better done through more AI prompting, or through purpose-built interfaces designed for precision?

Stan from Rayon shared a perfect metaphor at a recent event that Patric found particularly resonant. AI is like tap water—ubiquitous, soon to be a commodity, and essential. But without a vessel to contain it—whether that's a sink, a bucket, or better yet, a bathtub—the water's utility remains limited. You can cup your hands to catch some, but that's inefficient and temporary.

This metaphor perfectly encapsulates our thesis about the true value in construction tech. The companies that will ultimately win are those providing the "bathtubs"—the systems of record, ERPs, authoring tools, and data infrastructure that can properly contain, structure, and leverage AI's capabilities.

We've seen this pattern before. Remember when VCs had website sections devoted to "internet" or "consumer internet" investments? Today, that would be meaningless—every relevant company is an internet company in some form. The same transformation is happening with AI. Eventually, AI won't be a differentiator on its own, but rather a standard component of every successful tech stack.

What will differentiate winners is how deeply they embed themselves in workflows, how effectively they master specific use cases, and crucially, whether they own the vessels that contain and direct AI's power. The water is flowing—who will build the best bathtubs?

Authoring tools and data infrastructure are harder to build than AI

Is it easier to add AI to existing tools or build new infrastructure?

When we look at construction tech investment opportunities, two paths emerge: existing "bathtub" companies adding AI capabilities, or AI-first companies attempting to build infrastructure around their features. The second path is significantly more challenging.

Take authoring tools for architects and engineers. These interfaces for manipulating geometries are incredibly complex to build—far more difficult than implementing AI generation capabilities. Why would anyone believe that generating a geometry with AI would be sufficient without needing to author on top of it? It's simply not realistic.

The same applies to ERPs for construction companies. In the US, there are dominant players, while Europe has segment leaders. But globally, there are tons of sub-segments without proper cloud ERPs—presenting major opportunities.

What about data infrastructure? This is perhaps the most underappreciated "bathtub" of all. Data infrastructure typically combines data pipes connecting different systems with some sort of data structure or ontology. The common misconception is that AI eliminates the need for structured data. That's incorrect. Better prepared and structured data leads to much better statistical results, especially when deterministic outcomes are desired.

Before AI, data infrastructure allowed automation between systems that couldn't communicate effectively on their own. With AI, this becomes even more powerful. Some argue that agents can replace data infrastructure entirely—simply tapping into systems back and forth without needing integrations. But our conviction is that this is just a lazier version of data infrastructure, and ultimately, they'll converge back.

Think about it this way: you could build a two-wheeled vehicle, but four wheels might make more sense in most cases. Similarly, properly structured data infrastructure with AI enhancement will outperform pure agent-based approaches in most construction tech applications.

Looking at the competitive landscape, it's revealing that mega tech companies are already far advanced in deploying agentic AI. These companies generate substantial revenue from agents—but crucially, they already have systems of record in-house. If you're going down the agentic path, ask yourself: do you have an unfair advantage compared to these giants?

The real opportunity lies in better structuring data to enable superior inference—particularly in construction, where data remains relatively unstructured. This is the uncharted territory that remains difficult for incumbents to access.

Companies owning the "bathtub" will win over pure AI feature players

How do you build a lasting construction tech company in the AI era?

One founder pitched Patric an agentic framework for real-world infrastructure projects like highways and bridges. When asked how this differed from what UiPath already offers customers, the founder argued that UiPath workflows are cumbersome to maintain. But when pressed on why their own agents wouldn't require similar maintenance, there was no answer.

This reveals a critical insight: UiPath requires maintenance not because its software is changing, but because customer systems are constantly updating. Agents would face the exact same problem. The systems of record, ERPs, and data infrastructure are what's truly valuable to own—and these will eventually embed agents themselves.

So where are the opportunities to build these crucial "bathtubs" in construction tech? For authoring tools, multiple companies are vying for dominance, and we believe there's room for several winners. For geometric data infrastructure, there are two or three high-quality contenders.

The picture varies by region. In the US, Procore is the incumbent for the construction execution phase of enterprise general contractors. Europe lacks a clear leader. In Asia excluding China, the field is wide open, while within China, Glowdown has made significant inroads.

The pre-construction stage for enterprise general contractors represents another opportunity. We've seen at least one promising company with aspirations to become the system of record in this space, and we expect more to emerge.

Even CRM presents possibilities. While Salesforce dominates generally, there's room to build specialized systems of record on top of their platform for construction-specific workflows.

Overall, the playing field remains relatively open, particularly in regions like Asia (excluding China) where monetization may be challenging but the opportunity to become the dominant system of record exists.

Conclusion: The Vessel Value Framework

Identify whether you're building the water (AI features) or the vessel (systems of record)
If you're an AI-first company, chart a path to becoming infrastructure or risk commoditization
For existing "bathtub" companies, focus on embedding AI capabilities that enhance your core value
Remember that structured data remains crucial for effective AI implementation in construction
Look for opportunities in regions and segments where dominant systems of record don't yet exist