Five Durable Verticals of the Web: What AI Cannot Replace

SUMMARY

Source: Video transcript · ~26 minutes Topic: The future of the web in an AI-saturated economy — identifying the five layers of value that persist regardless of model capability

The Collapse of the Build Layer

At least a dozen companies are racing to build platforms where users describe an app and it appears. Lovable has raised $330 million at a $6.6 billion valuation, running at $300 million+ ARR with 100,000 new projects created on its platform every day. Vercel's V0 has 4 million users; Replit has roughly 25 million developers. Smaller players — Bolt, Shipper, Base 44 — compete on essentially the same pitch.

Since OpenAI's operator-style capabilities arrived, these platforms have expanded their promise from "we'll build your app" to "we'll build your entire business." But most of them are functionally thin wrappers around the same base models (Claude, ChatGPT, Gemini), differentiating only on UI, pricing, and surface-level features.

The Middleware Trap

When your product is a UI layer on top of someone else's intelligence, your moat is as deep as the time it takes to replicate the UI — which, with Claude Code and similar tools, is about a week. The conventional escape is training your own model: Cursor has done this for code, Replit has trained code-completion models with Databricks, and Vercel trained a custom autofix model with Fireworks AI. But model training is not what separates survivors from casualties. The companies that make it through the middleware trap own something structural that model providers cannot replicate.

Replit escapes because Claude cannot execute code — Replit owns the runtime where applications live. Vercel escapes because it has deployment infrastructure already hosting production applications for OpenAI, Anthropic, Nike, and PayPal. Notion doesn't even try to train a model; it offers a model picker and bets that 100 million users have built the largest structured knowledge graph of organizational information on the planet.

The pattern: AI commoditizes production. The companies that survive build on layers production cannot replace.

Vertical 1: Trust

The web is being flooded with millions of AI-generated apps, storefronts, and content streams daily. Most are indistinguishable from each other, many are garbage, and some are actively malicious. When anyone can generate a professional-looking checkout page in seconds, visual legitimacy means nothing.

The companies that become the verification layer — confirming that an app won't steal your credit card, that a service does what it claims, that content was produced by someone accountable — capture tremendous value. Stripe's position strengthens in an AI-saturated web because "Powered by Stripe" functions as a trust signal after processing over a trillion dollars in transactions. The same applies to Shopify, Apple's App Store review process, and Vercel's deployment infrastructure.

In the agentic economy, trust becomes even more critical. AI agents transacting autonomously need trust signals to determine which payments are safe, which services are verified, which APIs won't steal data. Trust providers become the routing layer for responsible web traffic. If an agent cannot verify a service, it won't transact — and in many cases won't be allowed to.

Vertical 2: Context

The most valuable resource on the internet is not compute or prompting ability — it is specific situational data: company data, customer relationships, medical records, meeting notes. AI is a general tool; to be useful, it needs context unique to each situation. The companies that become the authoritative store for context, and the permissioning layer governing where context gets served, own the choke point of the internet.

Notion understood this deeply and built custom agents that run autonomously across individual workspaces. Tens of thousands (or hundreds of thousands) of custom agents have been built by users. The context is what makes those agents valuable — Notion recognized that its data layer was the secret sauce and that bringing any AI model into that context creates enormous value for users.

This is the same structural data play that makes Salesforce durable, Epic durable in healthcare, and Palantir durable in security. Other players in the context space include Snowflake, Databricks, and potentially Apple and Google if they succeed with local AI. Google recently launched a context layer for Maps, underscoring how many ways it has to win the AI race: as a model provider, a foundation player with TPUs, a context player, an ecosystem player, and a devices player.

Vertical 3: Distribution

An app can be generated in seconds, but someone still has to see it. The bottleneck was never building — it was always distributing. When the supply of digital products increases by 10x or 100x, an edge in distribution becomes essential for any signal to reach customers. When supply is infinite, curation becomes the scarcest resource.

Google, Apple's App Store, TikTok, and YouTube are distribution monopolies that AI makes more powerful, not less — gatekeepers strengthen when the flood grows. For the agentic economy specifically, agent discovery is a massive unsolved problem. If every business has AI agents, something needs to help those agents discover where to do business with one another and with humans. An agent-native app store or discovery mechanism is an emerging category.

What makes a business viable for agent transactions is more than deploying an MCP server. Transaction speed, API clarity, ease of selection, and simplicity of fulfillment all need rethinking with agents at the center. Almost no businesses are thinking this way yet.

Vertical 4: Taste

When producing software is free, what you choose to produce becomes the entire game. Product decisions, design sensibility, editorial judgment about what is worth building — these are human skills AI can assist with but cannot replace, because they require a point of view on how humans do business with humans. Taste is a conviction about what should exist in the world that is not easily derivable from training data.

The analogy is music production after GarageBand and now Suno: the producers who thrive are not the ones with the most expensive studio but the ones with an ear for what will connect with an audience. The same is about to happen to software. A vibe coder who ships an app in minutes hasn't done the hard part — figuring out how it will deeply connect with an audience.

On the agentic web, taste manifests as orchestration quality. The winning agent systems will be the ones where a human with deep domain expertise has carefully tuned prompts, designed workflows, chosen tools, and made thousands of small editorial decisions. Even as auto-research and agent self-evolution emerge, the human remains accountable for what the agent does and how it participates in the larger economy.

Vertical 5: Liability

Someone must be on the hook. When an AI-generated financial plan loses money, when an AI-built medical app gives bad advice, when an AI-generated contract contains a defective clause — "the AI did it" is not an answer that survives in court. Regulated industries (healthcare, finance, legal, insurance) are fundamentally liability niches; the professionals in them sell accountability.

Counterintuitively, the better AI gets at sounding plausible, the more important authentic accountability becomes — because the mistakes made with plausible-sounding AI are far more serious. In the agentic economy, liability becomes a governance layer: defining boundaries for autonomous agents, auditing their actions, and bearing ultimate responsibility.

Players in this space range from Deloitte and McKinsey (repositioning as AI assurance providers) to ElevenLabs (offering insurance for voice agents), regulated SaaS platforms like Veeva and Elation, and individual AI professionals providing safety and vetting protocols.

The Future Web in Layers

Model providers (OpenAI, Anthropic) own the bedrock intelligence layer — enormously valuable, increasingly commoditized relative to each other. Wrapper companies (Lovable, Bolt, Shipper) don't own anything durable; most will die, some will be acquired, and a few with enough data and momentum may become platforms. Infrastructure players (Vercel, Replit, Stripe, Shopify) own trust and execution layers and become the picks and shovels of the AI gold rush. Context owners (Notion, Salesforce, Snowflake, Databricks) own data gravity and become the permissioning layer. Distribution gatekeepers (Google, Amazon, Apple) own human and potentially agent attention.

Humans — founders, professionals, operators — provide taste, judgment, and accountability: the connective tissue that makes everything work.

The Actionable Question

Ask yourself: what do I own that still matters if AI gets 10x better? If the answer is nothing, change your positioning now. If a better model makes your product more valuable — because you own trust, liability, context, or a distribution position — you have something durable to build on.

The five verticals (trust, context, distribution, taste, liability) have always mattered on the web. AI is the forcing function that makes them matter more. And AI cannot take their place.