My Vertical AI Thesis: Why Horizontal AI Products Fail

There's a pattern I keep seeing in AI startups that I think is fundamentally broken. A team builds a general-purpose AI tool ("AI for writing" or "AI for productivity" or "AI for research"), ships it, gets initial traction from the novelty factor, and then watches as users drift away. The tool does everything okay but nothing exceptionally well, and eventually users go back to ChatGPT or Claude because the general-purpose models are getting better every quarter.

I've a strong opinion on this: the future of AI products is vertical, not horizontal. And my entire product strategy is built on this thesis.

What Horizontal AI Gets Wrong

Horizontal AI products try to serve everyone. They build a general-purpose interface on top of a general-purpose model and compete on UI, speed, or price. The problem is that all three of those advantages are temporary.

UI advantages disappear. Every good UX pattern gets copied within months. The chat interface, the sidebar, the inline suggestions: these are commoditized.

Speed advantages disappear. Model providers are constantly optimizing inference speed. Your latency advantage today is everyone's baseline tomorrow.

Price advantages disappear. Model costs are dropping exponentially. Competing on price in AI is like competing on price in cloud storage: it's a race to zero that only the largest players win.

What horizontal AI products can't compete on is domain expertise, because they deliberately chose not to have any. They are jacks of all trades and masters of none.

The Vertical AI Advantage

Vertical AI products go deep instead of wide. They pick an industry, understand its problems intimately, and build AI solutions that are deeply tailored to that domain.

The advantages of going vertical compound over time:

Domain-Specific Data

A vertical AI product accumulates domain-specific training data, evaluation benchmarks, and user feedback that a horizontal product never will. A legal AI tool's understanding of small claims court procedures improves with every user interaction. A general-purpose AI tool has no mechanism to accumulate that kind of specialized knowledge.

Workflow Integration

Every industry has its own workflows, its own tools, its own data formats. A vertical AI product integrates into these existing workflows. It speaks the industry's language, connects to the industry's systems, and fits into the industry's processes. A horizontal product asks users to adapt their workflow to the AI.

Trust and Credibility

In regulated industries like law, healthcare, and finance, trust isn't optional. Users need to know that the AI understands their domain's constraints, compliance requirements, and professional standards. A vertical product can build and demonstrate that trust. A horizontal product, by definition, cannot.

Defensible Positioning

Perhaps most importantly, vertical AI creates a defensible market position. When OpenAI makes GPT-5 twice as good, horizontal AI wrappers lose their value proposition overnight. But a vertical product with deep domain integration, specialized data, and workflow-specific features has moats that a better general model doesn't wash away.

Lessons from My Own Products

I didn't arrive at the vertical AI thesis through abstract thinking. I learned it by building products.

My aviation portfolio (Aviation Infinity, AvioSharing, New Pilot Shop, Want To Be a Pilot) succeeds precisely because each product goes deep into a specific niche within aviation. Aviation Infinity isn't "a content platform that also covers aviation." It's an aviation-first platform built from the ground up for aviation enthusiasts. The content categories, the community features, the UX patterns: everything is designed for that specific audience.

When I started exploring AI products, I initially considered building something horizontal. An "AI assistant for entrepreneurs" or something similarly broad. But my experience with aviation taught me that the specificity is the product. The more specific you are, the more valuable you become to the people in that niche.

Applying the Vertical Thesis

This is why I believe the future of AI products lies in collections of vertical AI agents rather than single horizontal platforms.

A vertical AI legal tool should not try to be an AI for everything with a legal feature. It should be built from the ground up for people navigating the legal system. Every interaction pattern, every document template, every workflow should be designed for legal use cases.

Similarly, an AI coding assistant should not be "ChatGPT but for coding." It should be a coding agent that understands multi-product architectures, TypeScript monorepos, and the specific patterns of building and maintaining a portfolio of products.

Each product should go deep into its vertical while sharing technical infrastructure horizontally. The vertical depth is the product. The horizontal infrastructure is the efficiency.

When Horizontal Works

I should be fair: there are scenarios where horizontal AI works. If you're building a developer tool or an infrastructure product, horizontal makes sense. Vercel's AI SDK is horizontal by nature because it needs to work across industries. Model providers themselves are horizontal because they're the foundation layer.

The distinction is between infrastructure and applications. AI infrastructure can and should be horizontal. AI applications, the things end users actually interact with, should be vertical.

The Market Is Moving My Way

I am seeing increasing evidence that the market agrees with this thesis. The AI products that are gaining real traction and retaining users, not just getting sign-ups from hype, are the vertical ones. Harvey in legal, Abridge in healthcare, Vanta in compliance, Ironclad in contract management. These products win because they understand their users deeply, not because they have better models.

The horizontal AI wrapper graveyard is growing. Products that launched to great fanfare in 2023 are quietly shutting down or pivoting because they can't differentiate from ChatGPT plus a nice UI.

What This Means for Builders

If you're building AI products, my advice is direct: pick a vertical and go deep. Talk to users in that industry. Understand their workflows, their pain points, their language. Build something that a general-purpose AI could never replicate because you have invested in understanding the domain.

The AI model isn't the product. The domain expertise wrapped around the model is the product. And that expertise takes time, research, and genuine engagement with the people you're building for.

That's the bet I am making with my product strategy, and it's the same bet that has worked for me across 16 years of building products in aviation, travel, and beyond. Go deep. Go vertical. Let the horizontal products fight over the scraps.