I think the answer to that lies in HTTP discovery protocols. I'm imagining self-organizing applications based on the need of the user, wiring together representations of disparate data sources into a cohesive interface.
You shouldn't have to be a dev to take advantage of the API universe. You'll just need tools that are able to reason about the data available.
That's the goal of the semantic web/linked data, and it already works for simple cases. But to take advantage of that you can't be stuck in the popular "REST-but-not-really" custom APIs; you need standard formats and protocols (e.g. RDF/a and SPARQL) that can be queried by software that knows nothing about that particular service.
You shouldn't have to be a dev to take advantage of the API universe. You'll just need tools that are able to reason about the data available.