You Can’t Vibe Code Your Way to Trust: Why AI in Construction Needs More Than a Weekend Hack

Two weeks before this conversation, a competitor released a free AI contract review tool for construction. Within days, someone else shipped another one. The chatter was loud and predictable: contract review is done, commoditized, a feature any large language model can handle out of the box.

Josh Levy has heard it all before. He’s spent seven years building Document Crunch and over 20 years in construction risk management. When we asked him to respond, he didn’t start with product specs or competitive positioning. He started with Theodore Roosevelt.

“It’s not the critic who counts,” Levy quoted, reading from a framed piece in his office. “The credit belongs to the man who’s actually in the arena, whose face is marred by dust and sweat and blood.” It set the tone for everything that followed, a founder’s unfiltered take on why shipping fast and earning trust are two very different things in this industry.

What Actually Happened: The Free Contract Review Tool Debate

Two tools, one commoditized feature, and a whole lot of noise

A ConTech AI competitor released a free contract review tool. Then a second player built another one (with a charitable twist, to be fair to them, every review triggered a donation). Social media ran with it. The argument: contract review as a standalone feature is now a commodity. Any decent LLM can do it. The moat is gone.

Levy’s pushback was blunt. “I don’t think you can vibe code your way to trust in this industry.” He made the distinction between shipping a solution and earning the confidence of legal departments and risk teams at major contractors. Those teams aren’t going to swap their proven construction risk management software for a free tool someone assembled over a weekend, no matter how slick the demo looks.

The broader question here matters for anyone watching AI in construction right now. If foundational models keep getting better (and they will), what actually separates a purpose-built vertical platform from something any decent engineer can spin up with a prompt and an API key?

Why Vibe Coding Falls Short in Construction Risk Management

Technology is good. Trust is earned inch by inch.

Vibe coding, the practice of building software by describing what you want to an AI and letting it generate the code, has exploded since Andrej Karpathy coined the term in early 2025. It’s changed how startups prototype, how founders validate ideas, and how internal tools get built. For construction technology in 2026, it means the barrier to shipping a basic AI tool has essentially collapsed.

Levy doesn’t dispute any of that. What he disputes is the leap from “I can build it” to “contractors will trust it with their risk exposure.”

“Good luck trying to enter the risk management space and earn the trust of legal departments and risk teams of these big contractors. I’m sure they’re just thrilled and comfortable to let folks go use free tools to do major, major, major risk work.”

He was being facetious, obviously. But the point lands. Construction is an industry where a single contract oversight can trigger disputes worth millions. The companies managing that risk aren’t optimizing for clever, they’re optimizing for reliable.

This maps to a pattern showing up across vertical SaaS more broadly. The AI layer is getting commoditized. The domain knowledge, the customer relationships, the regulatory understanding, those aren’t.

The Boomerang Effect: 70% There Isn’t Where the Magic Happens

Contractors are trying DIY AI. Then they’re coming back.

Here’s where the conversation got really interesting. Levy shared that Document Crunch is already seeing what he called “the boomerang effect,” companies that tried to build their own AI contract review tools internally and then circled back.

“We’re already seeing it. People that try to do it themselves. Oh, it’s not as easy as you thought. Oh, we vibe coded it 70-80% of the way there, but that’s not where the magic happens.”

That 70-80% figure is worth sitting with if you’re a contractor weighing the build vs. buy decision for construction AI tools. Getting a basic contract summary out of an LLM is straightforward. Getting it to reliably flag the specific risk provisions that matter to your company’s tolerances, across 17 different document types, integrated into your actual project workflows? That’s the remaining 20-30%, and it’s where the real cost lives.

For construction firms already operating on razor-thin 2-3% margins, that gap between “pretty good demo” and “production-grade risk platform” can be the difference between catching a liability clause and missing one.

The Data Moat: Why Feedback Loops Beat Feature Launches

Thousands of contracts, years of iteration, and an unfair advantage

Levy made a compelling case for something that’s becoming a recurring theme across ConTech AI: data defensibility. Document Crunch has processed thousands of construction contracts over five years. Every review generates feedback. Every customer interaction adds another data point about how real teams use the tool on real projects.

“Imagine that every one of those little interactions is just another node that you just start connecting the dots. Somewhere along that path, you start to look up and realize just how big of a gap exists between you and whatever else is in the market.”

This tracks with what CEMEX Ventures noted in their recent 2026 AI trends report: “In 2025, capital rewarded AI integration. In 2026, capital will reward data defensibility.” Algorithms can be replicated. Proprietary, structured, high-volume construction datasets can’t.

The specification analysis story was a sharp example. Levy described how his VP of engineering spotted a new foundational model with expanded context windows and had it integrated into the product within 24 hours. That speed only matters because the team already understood the problem deeply enough to know what to do with the new capability. A year of iteration later, Document Crunch’s specification analysis is, by Levy’s account, the strongest commercially available option in the space, and one of their fastest-growing features.

From Contract Analyzer to Project Platform

90% legal users became 90% operations users. That shift changes everything.

Levy dropped a stat that caught our attention. A few years ago, Document Crunch was used 90-95% by legal and risk management teams. Today, over 90% of usage comes from pre-construction, project teams, project controls, and operations.

That’s a complete inversion of the user base, and it signals a broader shift. The company is rolling out what Levy described as “really the new version of Document Crunch,” a project-based risk management platform covering 17 different document types across the project lifecycle. Think mobile-responsive, voice-responsive, with cross-document intelligence.

For the construction technology market in 2026, this kind of platform expansion matters because it shows what happens when a vertical AI tool earns enough trust to spread beyond its original use case. Legal teams validated the accuracy. Project teams saw the value. Now it’s embedded in daily workflows across the organization.

The Build vs. Buy Question for Contractors

Your team can probably build 80% of it. The question is whether 80% is enough.

The honest version of the build vs. buy debate for construction AI tools goes something like this: yes, your team (or a competent developer with Cursor and a weekend) can probably build a contract analyzer that handles the basics. The LLMs are good enough now. That part is real.

The question is what happens next. Who maintains it? Who validates the outputs against your specific risk tolerances? Who handles the 17 different document types your pre-construction team actually works with? Who builds the feedback loops that make it better with every review? Who takes the call when it misses something on a $50 million project?

“Be intellectually honest. Yes, the technology is good. You can maybe vibe code an isolated workflow. But if you really want deep, differentiated solutions for big meaningful problem spaces, I think it’s going to take more than that.”

Comparison: Free AI Contract Review vs. Vertical SaaS Platforms