JSON-LD is JSON that carries its own meaning. A field is not just called name — it is linked to a
shared vocabulary term, usually from Schema.org, so a machine receiving it knows this is the same
concept of “name” everyone else means, rather than guessing from a string.
It is intellectually one of the most ambitious ideas in the API space and one of the most underadopted. The reason is the familiar one: semantics are a public good. The provider does the work of aligning to shared vocabularies and the benefit accrues to consumers and to the ecosystem. Absent pressure, most providers name their fields whatever the database calls them and move on.
Where it has worked is where the incentive was selfish rather than civic — job postings marked up with JSON-LD because Google surfaces them, products marked up because it drives commerce placement. That is the honest lesson: semantic standardization spreads when there is a direct, near-term payoff to the publisher.
Which brings us to now. Agents consuming APIs without a human interpreting the field names are exactly the case semantics were designed for. If the agentic surface becomes a place where discoverability drives revenue, semantics acquire the selfish incentive they always lacked. I would not bet the estate on it, but I would put shared vocabulary terms in your schemas where the cost is low.
