How Do We Grow the GDP of the Internet?

Taste is the next frontier.

Stripe grew the GDP of the internet by making payments frictionless. Reduce the cost of transacting, and transactions that wouldn’t have happened suddenly do. The pie grows.

The next friction layer is taste. We have payments, and attention is a game of trends. Taste is something different: the underlying preferences that determine what people actually want. Make taste legible, portable, and owned by the people who have it, and you unlock an entire layer of economic activity that currently doesn’t exist. Billions of dollars in transactions fail to happen every year because the right match between preference and product never gets made. The information exists, but it’s fragmented across platforms, locked in proprietary databases, inferred from surveillance rather than shared through relationship.

---

The Agent Economy is Already Here

This may sound futuristic, but personal agents are already making this real. Claude Code runs as a personal chief of staff on millions of computers, spawning sub-agents to handle research and tasks while you focus on something else. Visa has launched a Trusted Agent Protocol for cryptographic verification of AI transactions. PayPal and OpenAI built Agent Checkout Protocol so you can purchase things without ever leaving a conversation. Google’s AP2 standard has Mastercard, Coinbase, and more than 25 other partners building toward a world where agents transact on your behalf continuously, around the clock.

What’s missing is context.

These agents need to know what you want, and right now they’re guessing. They scrape your browsing history, infer from purchase patterns, read the tea leaves of your scroll behavior. That’s surveillance with better UX. An agent without real knowledge of your preferences is just automation, a faster way to do things you didn’t ask for. An agent with deep knowledge of your taste is something else entirely: a representative that can actually represent you.

---

Discovery is Inverting

The way humans find things is changing faster than most companies have registered. Adobe reports that traffic to retailer websites from AI sources grew 1,200% between July 2024 and February 2025. Bain found that 80% of consumers now rely on AI summaries for nearly half their searches. The traditional funnel—awareness, consideration, decision—is collapsing into a single moment inside the AI, before a human ever sees a brand or clicks a link.

This is where the GDP unlock happens. Right now, agents mediate discovery with incomplete information. They recommend based on what’s legible to them: SEO signals, structured data, whatever scraps of preference leak out of closed platforms. The matches they make are mediocre because the inputs are mediocre. Open the preference layer, give agents real access to what people actually want, and the quality of those matches improves dramatically. Transactions that couldn’t happen before—because the connection was never made—suddenly do.

For brands, the implication is stark: if agents can’t find you, humans won’t either. Your trust signals need to be legible to machines, not buried in walled gardens that no agent can read. For users, the implication is just as significant: your preferences are already being monetized at scale. Every recommendation you follow, every purchase you make, every signal you emit feeds models that sell attention back to advertisers. You just don’t own any of it.

---

From Surveillance to Ownership

What if the data that powers your agent actually belonged to you?

This is the gap Zumi fills. Zumi is an AI companion that learns your taste through daily conversation—what music resonates, what products excite you, what ideas you keep returning to. You talk to Zumi because you want to, because it’s useful and enjoyable, not because you’re being tracked in the background.

The understanding Zumi develops is genuine because it’s built through experience. When you’re offline, your Zumi exists in its own world—going to school, interacting with other Zumis, developing a perspective that emerges from lived time rather than just training data. The knowledge it builds grows alongside you.

Tapestry turns that understanding into ownership. Every preference signal gets attributed to you onchain. When your taste leads others to discover something—when your Zumi’s recommendation ripples outward and other people benefit from it—you’re rewarded. Your preferences become economic assets you actually control, not exhaust data captured by platforms and sold to the highest bidder.

---

The Infrastructure Layer

For any of this to work at scale, you need open social infrastructure that agents can read and write to.

The question some previous protocols missed: what does being open actually unlock? The answer is now obvious in a way it wasn’t five years ago. Open infrastructure means agents can access your social graph, your preferences, your trust relationships. It means data compounds across services rather than resetting every time you try something new. It means social context can flow into economic decisions.

Tapestry is the identity, memory, and economic graph for agents and apps. Portable identity that travels across applications. Persistent memory that compounds over time. An economic framework that lets social context translate directly into commerce. When developers build on Tapestry, they inherit a social graph with real depth, preference data with genuine context, and proven patterns for how taste becomes value.

---

The Bet

Stripe bet that reducing payment friction would grow the pie. They were right, and they captured a piece of every transaction they enabled.

We’re betting that taste is the next friction layer. Every AI agent will need preference data to function well. The question is whether that data remains fragmented and platform-owned—or becomes open, portable, and controlled by the people who generate it.

The second world is a bigger economy. We’re building what it takes to get there.

---