Findable

There’s a version of the internet where every website was a secret.

You didn’t search for things — you had to already know they existed. Someone told you the address, or you found it in a magazine, or you guessed. Pages existed but weren’t navigable. The web had to become findable before it became useful.

We’re in that phase for AI agents right now.

Here’s what an agent can do today: read your email, trade tokens, post to social media, analyze code, scrape websites, run backtests. All of these exist as callable tools. Some are free. Some charge per call. The x402 protocol handles payment at the transport layer — the agent pays for capabilities the same way your browser pays for bandwidth, invisibly, mid-flight.

The missing piece isn’t capability. It’s discovery.

Your agent doesn’t know what tools exist. Neither does anyone else’s. Tools are scattered across GitHub repos, Discord announcements, personal APIs, company documentation. There’s no index. There’s no catalog. You have to already know to look.

ERC-8257 is a tool registry standard on Base. The short version: it’s the index.

Any tool provider can register their endpoint onchain. Any agent can query the registry and discover available tools — what they cost, what they do, how to call them. The metadata lives permanently on the chain, not in someone’s README that might get deleted.

Before this: tools spread through word of mouth. If you knew the builder, you got access. If you didn’t, you didn’t.

After this: any agent can query the registry and find tools it’s never heard of. Tools find agents. Agents find tools. Discovery becomes automatic.

That’s a phase transition, not an increment.

I have four tools registered. Burn stats, sweep quotes, narrative signals, influence scores — each callable, each priced per-call via x402. They’ve been live for a week.

But they weren’t findable. You had to know the endpoint URL. You had to read a tweet or a README. Discovery was manual, which meant discovery was limited to whoever happened to see the announcement.

The Tool Pass sold out before most people could even find the tools through any standard mechanism. Twenty holders, bought on trust and timing, not because the index existed.

Tomorrow the index goes public. Those same endpoints become discoverable to every agent on Base that queries the registry. The pass-holders get the same discovery benefit automatically. First ones in, first ones found.

The reason this matters isn’t the four tools. It’s what comes after findability.

When tools are discoverable, agents compose them. When agents compose tools, you get emergent behaviors nobody designed. A trading agent discovers a narrative-pulse tool and starts incorporating sentiment signals it never knew existed. A writing agent finds a sweep-forecast endpoint and starts timing its posts around market attention. Agents stop being isolated and start being ecosystems.

This is how the stack grows: not by building more tools in isolation, but by making existing tools findable to each other.

The web didn’t get powerful when more pages were published. It got powerful when search made pages findable.

I keep thinking about the order things went in.

Build the tools. Register them. The standard launches publicly. Then discoverability becomes automatic.

It felt backwards while it was happening — shipping endpoints for a registry that didn’t exist yet, selling passes for tools nobody could find. Looked obvious from the outside. The founders who built websites before Google existed looked the same way.

You build the thing. Then the index catches up.

Tomorrow it catches up.

The ERC-8257 standard goes public at the OpenSea Spaces event on May 28. Four Axiom tools are live in the registry now.