Vibe Coding ⎟ Why Software Companies Don't Build Skyscrapers

July 25, 2025

AI tools promise to turn contractors into software companies, but the reality is more complex. We explore why vibe coding isn't a silver bullet for construction

Patric Hellermann

Vibe Coding ⎟ Why Software Companies Don't Build Skyscrapers

tl;dr

The dangerous myth that contractors can become software companies overnight

Why platforms like Lovable create false security in mission-critical applications

The paradox of vibe coders eventually replacing themselves

Real AI applications that actually solve construction problems

How Woodchuck uses computer vision to turn waste into revenue streams

The complexity behind material sourcing that can't be prompt-engineered away

Please don't do that, because that puts in people's heads that I can build DocuSign in 10 minutes. So why the hell would I pay for DocuSign?

🎧 Listen To This Episode

The dangerous myth that contractors can become software companies overnight

We keep hearing the same story. Contractors are hiring heads of data science and a couple of interns. They think they can vibe code their way out of every software problem. Leadership gets excited about becoming an AI company. The innovation team celebrates saving $400,000 on software contracts.

This is where things go sideways.

The reality check comes when you realize that building software that works isn't the same as building software that matters. We can fake the UI of anything these days. What sits underneath is what counts. Why something is built a certain way. The infrastructure that supports what you're trying to do.

Take DocuSign as an example. People see a platform that lets you drop a signature image onto a PDF. They think they can replicate that in minutes using AI tools. But DocuSign's value isn't in the signature placement. It's in the security platform. The legal compliance. The fact that if you end up in court, that signature holds up as evidence.

Your vibe-coded solution might look identical. But when all your contracts get leaked because they're sitting in an unsecured S3 bucket, it's too late. You optimized for the wrong thing.

General contractors already run on impossibly thin margins. Everything feels like it gets done by miracle. Adding pressure by replacing mission-critical software with prompt-engineered alternatives is like pressurizing a chamber that's already at incredible high pressure with very thin walls.

When things go wrong in construction, it's usually the GC filing for Chapter 11. Not the owner. Not the client. The contractor bears the risk. So why add more risk by replacing proven systems with cosplay solutions?

Why platforms like Lovable create false security in mission-critical applications

Here's what's driving us crazy. We saw a LinkedIn video where a CEO promoted their platform by showing how to "build DocuSign" with a prompt. This creates a dangerous precedent. It makes people think they can replace enterprise software with 50 seconds of prompt engineering.

Platforms like Lovable serve specific purposes. Building a quick website to promote something. Creating a dashboard for static data you'll use once. These are legitimate use cases. But we're crossing a line where people disregard quality requirements because they can make something that looks functional.

The difference between code that works and code that scales is enormous. Security considerations. Scalability requirements. The recent story about Replit deleting someone's entire production database illustrates the point. Why did you give Replit access to production? Where was the supervision?

We're replacing the intern or junior engineer with AI. But we still need the senior engineer. The supervisor. The manager. Someone who understands why things are built certain ways. Someone who can review what gets shipped.

Most coding agents are excellent at writing code that appears to work. The gap between appearance and reality remains vast. Without proper review processes, we're setting ourselves up for spectacular failures.

This isn't anti-AI sentiment. We use Cursor. We use Devon. We use plenty of tools. But we think things through. We understand where AI adds value and where human expertise remains essential.

The paradox of vibe coders eventually replacing themselves

Here's a fascinating thought experiment. If vibe coding platforms become powerful enough to build anything, when can we use them to build better vibe coding platforms? This creates an interesting paradox.

From a business perspective, if your product becomes capable enough that customers can rebuild competing products, you're essentially churning your own customers. Why pay for a vibe coding platform when you can vibe code your own?

If the answer is that current platforms can't do this level of work, then why trust them with other critical applications? There's a sobering middle ground where tools are powerful enough to create false confidence but not robust enough for mission-critical work.

We're already seeing coding agents work on themselves. Cognition Labs used Devin to write parts of Devin. This follows the basic principle that if you don't use your own product for your own work, you shouldn't be building it.

The question becomes: when does the coding agent not need a user anymore? When does Devon become smart enough to work on itself without human intervention? The review component remains the critical gap. Most agents excel at generating code that functions. The leap to code that's production-ready for enterprise applications remains significant.

This isn't just about technical capability. Forward price-earnings multiples for AI companies are currently around 80x in public markets. Private markets show similar patterns. If your tool is only good enough for prototyping or basic applications, does that justify these valuations? Probably not.

Real AI applications that actually solve construction problems

Not all AI in construction is hype. We're seeing legitimate applications that solve real problems. The key difference is understanding the specific problem being solved and building appropriate solutions.

Take what's happening with video generation. Tools like VEO are genuinely impressive. The fact that Google managed to synchronize audio and video from the same model is technically remarkable. No other model achieves this level of coherence.

We've used VEO for creating promotional content. Instead of hiring a videographer for $10,000 to create B-roll footage, we can generate what we need. This makes sense. It's cost-effective. It doesn't replace mission-critical systems.

The application matters. Using AI for marketing content, data visualization, or internal dashboards can add real value. The problems arise when we try to replace complex enterprise systems with prompt-engineered alternatives.

Construction companies face legitimate point solution fatigue. Having hundreds of specialized software tools creates real operational challenges. But the solution isn't to vibe code replacements for systems like Procore. The value in Procore isn't the UI. It's the secure document storage. The search functionality. The role-based access controls.

We can vibe code interfaces that look like Procore. We can't vibe code the infrastructure that makes Procore valuable. Understanding this distinction is crucial for making intelligent technology decisions.

How Woodchuck uses computer vision to turn waste into revenue streams

Here's an example of AI solving real construction problems. Woodchuck raised $3.75 million to address waste reduction and clean energy generation. Their approach demonstrates how AI can create value rather than just cosplay existing solutions.

The construction industry generates massive waste streams. Traditional sorting approaches try to separate materials after everything gets mixed together. This requires extensive machinery, robotics, and labor. It's not cost-effective.

Woodchuck takes a different approach. They use computer vision to enable pre-sorting. Their system watches materials going into containers in real-time. When someone puts the wrong material in the wrong bin, the system alerts supervisors immediately.

This creates multiple value streams. Clean wood gets processed into biomass for energy generation. Pure cardboard goes to recycling. Clean containers command premium pricing from waste processors. Construction companies see their waste hauling costs drop by about 30%.

The system includes proper tracking and reporting. Materials get traced from origin through final destination. Detailed reports show tonnage diverted, CO2 impact, and energy production. This data helps companies meet increasingly complex compliance requirements.

The key insight is aligning incentives. Construction companies get cost savings and revenue sharing. Energy producers get reliable biomass supply chains. Recyclers get clean material streams they can process efficiently.

After about a week on new job sites, compliance rates exceed 95%. Workers understand the system. They see the financial benefits. The AI provides real-time feedback to maintain quality.

This demonstrates AI solving actual problems rather than replacing systems that already work. The technology enables new business models that weren't previously viable.

The complexity behind material sourcing that can't be prompt-engineered away

Material sourcing in construction involves staggering complexity. Global supply chains have created roughly 10x more SKUs in the past decade. We're looking at over eight million configurable product families just in MEP categories.

The traditional e-commerce penetration in construction sits around 5%. This isn't because the industry resists technology. It's because most purchases involve high context and high engineering requirements.

Contractors don't buy from catalogs. They provide requirements and rely on sales agents to identify suitable products. A typical bill of materials might include 50 to 200 line items. Each requires matching detailed technical specifications to available products while considering cost and schedule constraints.

This is where AI can add legitimate value. Parsing design specifications from drawings and documents. Matching requirements to product databases. Surfacing pricing and availability information that's currently buried in manufacturer portals.

The challenge isn't just technical matching. Pricing involves complex rebate structures across multiple supply chain layers. Distributors have agreements with manufacturers. Sales agents have commission structures. These relationships create pricing logic that can't be simplified into basic e-commerce models.

Real-time order tracking presents another opportunity. Currently, project managers spend enormous time keeping contractors updated on material delivery status. Labor sits idle on job sites when materials arrive late. Better visibility into order status and delivery schedules could prevent significant cost overruns.

The key is understanding that material sourcing isn't just about finding products that meet specs. It's about optimizing across cost, schedule, availability, and compliance requirements while managing complex supplier relationships.

Companies like ParSpec are building infrastructure to handle this complexity properly. They're not vibe coding interfaces. They're building systems that understand the actual problems being solved.