Most of What You're Building Will Be Replaced by a Better Model. Here Are the Five Layers Between You and Irrelevance.

Read the original article

Most of What You're Building Will Be Replaced by a Better Model

Main Thesis

Unless you're Anthropic, OpenAI, or Google, building AI products on top of foundation models is existentially risky. A better model will simply erase your competitive advantage overnight. The author argues that durable value on the future internet lives in exactly five layers that AI structurally cannot replace.

Key Findings

The Collapse of the Build Layer

• The AI app builder category (Lovable, Bolt, Replit, Shipper, etc.) appeared safe but is now collapsing.
• Lovable raised $330M at a $6.6B valuation and hit $400M ARR — yet the entire category is fragile.
• Most of these companies are thin wrappers around the same foundation models, giving them a moat roughly one week deep.
• When a better model ships, the wrapper becomes irrelevant.

The Escape Hatch

• Companies like Replit, Vercel, and Notion survive because they offer something beyond model access — infrastructure, workflow integration, or platform depth.
• The distinction: does a better model make your product redundant or more valuable? If redundant, you're on the wrong layer.

The Five Durable Verticals

These are the things AI structurally cannot provide on its own:

1. Trust — Verified identity, reputation, and credibility that users and institutions rely on.
2. Context — Deep, proprietary knowledge of a specific domain, user, or workflow that a generic model lacks.
3. Distribution — Established audiences, relationships, and channels that take years to build.
4. Taste — Curation, aesthetic judgment, and editorial voice that resonates with a specific community.
5. Liability — Accountability and regulatory compliance that requires a human or institution to stand behind outcomes.

The Real Map

• The future web will reorganize around these five layers.
• Builders who anchor their products to one or more of these verticals are insulated from model upgrades — a better model actually strengthens their position rather than threatening it.

Practical Takeaways

• Positioning Audit: Ask yourself which of the five verticals your product occupies. If it occupies none, you are a wrapper and you are at risk.
• Agent-Readiness Stress Test: Evaluate whether an AI agent can fully replicate what you've built. If yes, your moat is gone.
• Build on layers, not models: Your product should become more valuable as models improve, not obsolete.
• The right question isn't 'what can AI do?' but 'what does AI still need from humans or institutions to deliver real value?'