I have spent more time on this problem than almost any other, including building a format to solve it, so I will give you the honest answer: discovery is not really a technical problem, which is why fifteen years of technical solutions have not fixed it.
The formats work. APIs.json, OpenAPI, catalogs, registries, search engines — none of these are the bottleneck. The bottleneck is incentives. A provider publishing machine-readable discovery artifacts does work whose benefit accrues to the ecosystem, to aggregators, and to consumers. The provider with a marketing site and a docs portal does not personally feel the problem being solved. So it does not get done, and the index stays incomplete.
ProgrammableWeb is the cautionary tale. Hand-curated, genuinely valuable, foundational to the early API economy — and it eventually shut down, because a manually maintained directory cannot keep pace with a space that grows faster than it can be catalogued, and because nobody could make the economics work.
What has changed is the consumer. Agents need to find and evaluate capabilities at runtime, without a human to Google around the gaps. That may finally give discovery a selfish payoff for providers: if being discoverable determines whether an agent chooses you, publishing the artifacts stops being civic-minded and starts being commercial. Discovery is also becoming more contextual and ephemeral — the right API for this task, right now — rather than a permanent listing in a directory.
