The W3C JSON-LD Community Group
Go Back
JSON-LD CG
Minutes for 2025-05-21
- Present
- Ivan Herman, Gregg Kellogg, Leonard Rosenthol, Victor Lu, Roy Ruoxi, Gaowei Chang, Dmitri Zagidulin, Kaliya Young, Anatoly Scherbakov, Benjamin Young, Niklas Lindström, David I. Lehn
- Chair(s)
- Gregg Kellogg
- Scribe(s)
- Benjamin Young, Gregg Kellogg
- Agenda
- https://www.w3.org/events/meetings/c7ab3fcd-4dc4-4444-ae8a-c91ca2131113/20250521T120000/
- Topics
Topic: Announcements and Introductions
Benjamin Young is scribing.
Gregg Kellogg: Introductions?
Leonard Rosenthol: Leonard Rosenthol, Adobe
Victor Lu: I'm participating in SPDX which is using JSON-LD, so I'm interested in hearing more
Gregg Kellogg: Please join IRC if you can
Roy Ruoxi: I'm staff contact for Accessibility for A11Y groups
... and I'm responsible for some AI things in China
... so I reached out to the ANP group
... they are having some trouble getting into Zoom
... but once they are on, it will be good to see what they are sharing
Dmitri Zagidulin: I'm Dmitri Zagidulin. I do a lot with DIDs. Looking into Agentic AI. And am especially interested in their overlap.
Kaliya Young: I'm Kaliya Young and I have been in the community for some time, but am very interested in AI and this presentation.
Gregg Kellogg: Besides folks with technical issues, anyone else like to say hi?
Gregg Kellogg: https://github.com/json-ld/json-ld-star/pull/61
https://github.com/json-ld/json-ld-star/pull/61 -> Pull Request 61 Updated test results for JSON-LD-star (by gkellogg)
... k. moving on for the moment to other announcements
... I've done some updates to the JSON-LD-star test suite...though not the spec just yet
... there is a thumbnail sketch in the PR, though
... and I'd love feedback from niklasl and PA (who's not here)
... once that review happens, we can make those spec changes in advance of a future 1.2
... if you have comments on that, please add them to the PR
... anything else for announcements?
Topic: Agent Network Protocol
Gaowei Chang: I'm the creator of the ANP community
Gregg Kellogg: Would someone like to give an overview of the ANP work?
... and how the JSON-LD groups might help you?
Roy Ruoxi: Gaowei has slides
Gaowei Chang: ANP is the Agent Network Protocol
... this is our model
... it is an open protocol for agent communication
... we want to see an agent internet
... that supports discovery and collaboration
... so this includes identity, description, and discovery
... the identities are DIDs, the descriptions are structured data
... JSON-LD in ANP is used for the descriptions
... as well as the interfaces and goals of the agents
... we use terms from schema.org
... JSON-LD helps connect the public data of agents into our data network
... that makes it easy for AI to understand
Gaowei Chang: Many developers prefer "simple JSON" instead of taking the time to learn JSON-LD
Roy Ruoxi: Gaowei is there more to share about JSON-LD?
... or links to other material online?
Gregg Kellogg: Does anyone have questions or comments?
Victor Lu: Just to comment about JSON-LD. In SPDX, JSON-LD has been adopted
... we also hear the complaint from JSON developers
... is there anything to make it simpler?
Gregg Kellogg is scribing.
Benjamin Young: I'd like to hear more from the ANP community about the perceived pain of developers using JSON-LD.
... We need to understand where that pain is felt. When there are communities such as schema.org and you stay within those bounds, it's just like typing other JSON.
... Where it's been hard for developers is when they don't have an established base and want to extend one.
... In these cases, we can create examples to show simplification.
... Knowing the particular pain points can help in this.
Victor Lu: In the community I'm involved in, there's no need to promote JSON-LD because it's the only solution we think works well
... and we are defining new standards for software bill of materials
... so the comments often come from developers who are trying to build intermediate software files
... I'll also share a link to the SPDX Slack space
Ivan Herman: -> SPDX page https://spdx.dev
Gregg Kellogg: There are some great things on json-ld.org to help folks learn
... but there are a few things that often trip people up
... and often in an attempt to make things easier in some ways, we've made them harder in others
... like property maps which we added in 1.1
... they more naturally reflect the way people write JSON "in the wild"
... but the algorithms that it takes to parse them into JSON-LD may scare or confuse people
... so I'm wondering have you looked into creating best practices or templates?
... and are their things we could do such as adding examples to the playground?
ANP Specification: https://github.com/agent-network-protocol/AgentNetworkProtocol/blob/main/07-anp-agent-description-protocol-specification.md
Dmitri Zagidulin: Like many of you, I've heard similar complaints for years
... the main "lift" seems to be around data modeling
... not really about using existing contexts or libraries
... those are fairly push-button or follow-the-tutorial
... it's when you have to actually model new data
... how to create a context, where/how to host it, etc.
... much of that is pointlessly difficult
... really, I want a graphical drag-and-drop context builder
... more time on that would bring many benefits
Leonard Rosenthol: I've heard this conversation before as well
... it comes down to communities and use cases
... I've brought it into various places such as XMP
... it's also in Content Credentials
... we use it internally at Adobe
... and in all that time I've never heard push back about JSON-LD complexity
... maybe if you don't understand JSON beyond JavaScript object serialization, then sure...it might be confusing
... but if you're used to data structure design, etc., then I don't see the lift
... maybe that's the differentiator
Gregg Kellogg: Thank you. a good community to look at may be schema.org
... they have extensive examples that cross many domains
... and they have loads of tools to turn RDFS (or their variant of that) into contexts, etc.
... that may not always be adequate, but it could be a helpful start
Dmitri Zagidulin: I'd love to hear more about JSON-LD and AI
... besides dealing with other developers perceived pain about JSON-LD
... do you have any specific asks of this group?
Benjamin Young: Leonardr's comments about data modeling are right; it's about data modeling.
... Most developers don't do data modeling, they just use ad-hoc JSON in their applications.
... They typically aren't concerned with interoperability. Groups that have don't tend to have an issue with JSON-LD complexity.
... If you don't understand data modeling, it is more difficult :)
Slides from the ANP presentation: https://www.agent-network-protocol.com/blogs/posts/anp-w3c-webagents-presentation.html
Leonard Rosenthol: We've had great success with JSON-LD and have not had any issues with using it with LLMs
... under the hood, I have no idea if the LLMs are treating it as raw JSON or Linked Data
... but I will not there's no remote context retrieval
... we pull them all down first
... so JSON-LD really serves it's purpose
Gregg Kellogg: I do think we'll see more "cooked in" contexts
... by using the `@context` values as identifiers for context files that should always be pre-cached/stored and not retrieved
... but now I think we're drifting into specific use case modeling
... that may also be a need for ANP as well--to have their community store their contexts locally for use at runtime
... Dmitri, did you want to expand on what you said in Zoom chat?
Dmitri Zagidulin: When I say it's difficult to create lightly
... whenever I sit down to convert exiting JSON or create new JSON-LD contexts
... there's way more of the RDF ecosystem that I have to know about than I'd really care to
Gregg Kellogg: It doesn't sound like something we can easily address in a spec
... but we do have notes and explanations around vocabularies vs. contexts, etc.
Dmitri Zagidulin: Absolutely. For me it's about more tutorials and wysiwyg tools
... the another point of confusion: JSON Schema vs. JSON-LD contexts
... we know they're different, but developers get confused
... there are a couple of projects such as LinkML that lets you define a vocabulary in a document and it generates JSON-LD contexts and vocabulary documentation and JSON Schemas
... there's another one that lets you roundtrip between JSON Schema and JSON-LD contexts
... maybe explaining the overlap and lack of overlap
... I'll share links
Gregg Kellogg: We can and should put out some blog posts...we just need volunteers
... the best people to author these types of things are those that have felt the pain
... so Dmitri, you or others who've felt that pain are ideal for those contributions
LinkML (YAML based, compiles to JSON-LD contexts, JSON Schemas, protobuf definitions, etc etc): https://linkml.io/
Anatoly Scherbakov: I do agree that JSON-LD is still not yet a part of the mainstream JSON developer community
... in general, the movement of the industry seems to be moving "toward the code"--such as Infrastructure as Code, etc.
... I do understand there are many other use cases
... but generally the industry moves toward more code vs. less code
... maybe if more of the URLs in JSON-LD resolved to human readable content, that would be helpful
... I think LLMs can certainly help us write JSON-LD
... when you have a piece of code an LLM helped you write, you can visualize it, then you talk to the LLM and it helps you evolve the JSON-LD
... so it can help us write Linked Data
Gregg Kellogg: Sounds like vibe coding
Leonard Rosenthol: Jumping back to something earlier
... were you talking about CBOR-LDs predefined contexts? or something else?
Gregg Kellogg: I think CBOR-LD has recently moved beyond the registry thing to something else
... but someone else will need to speak to that
... My point is mainly that specs are beginning to reference point-in-time contexts with their defined URLs
... vs. context URLs that may change at any point--as schema.org's primary context URL does
... we are already discussing defining a more advanced context loader specification change
... that would allow you to more narrowly define what context loading scenarios you're OK with
... and it would help remove the ambiguity
Leonard Rosenthol: Thanks.
Roy Ruoxi: I have something from Gaowei
... because the ANP community are a big fan of Tim Berners-Lee's Semantic Web
... they have tried to do ANP with all the Semantic Web and Linked Data capabilities
... and now they have many industry demos from the ANP community
... and many companies are starting to use ANP with JSON-LD in their products
... but there are others who do think JSON-LD is complicated
Roy Ruoxi: https://www.w3.org/community/agentprotocol/
... in the meanwhile, Gaowei and ANP folks setup a community group in the W3C
... we'd love to hear more from communities to see if we still want to use JSON-LD in the protocol or something else
... so I wanted to mention that here if others wanted to be involved.
Benjamin Young: Reflecting on document loading and pinning down the document loader.
... That makes reading JSON-LD more like a spec. JSON-LD is always self-identifying and it needs to be clear about that.
... For example, the schema.org context is very "webby", but they also publish versioned URLs for different releases.
... They can be hard to find, but you can find it if you try.
... We've heard from communities that want guidance around that. It's not a pivot of the web, but a different way of looking at it.
... Provide more data to describe what you want, say using a hash. A document loader helps with that.
Ivan Herman: When I was young and foolish I was very involved in RDF
... so I was in that community when JSON-LD arrived
... when it was proposed, the claim was that it would simplify RDF to people who did not know RDF
... and we have gone through several iterations of things
... and JSON-LD is now very much the same as RDFa
... we tried to put a graph based model into something that is not graph-based
... it's true for RDF and JSON-LD
... so maybe that starting position was pushed to the edges...and we may want to revisit that
... but that presupposes that RDF is simple...but perhaps that's debatable
... when I look at a JSON-LD context file...I have to virtually run a parser in my head and think about the output in RDF
... there is a complexity here because we are trying to bridge two worlds
... it is interesting to hear that the ANP folks started with RDF and then used JSON-LD because it was there
... so maybe we could consider how to make RDF easier to write in JSON
... that is pretty philosophical
Victor Lu: We started with RDF also
... so that path was similar for us
... even so, the problem is the people who use SPDX complain that it's not as easy as JSON
Gregg Kellogg: It may be that we can't please everyone all the time
... I've written most of these specs...and I often get confused
... I think most folks in RDF land start with Turtle
... as that's the easiest
Ivan Herman: Absolutely
Gregg Kellogg: I then write frames to turn it into the JSON I want
... so maybe we just need more explanation around how to work with all the layers
... and we see loads of people attempting to create "simpler" JSON-LD
... so we end up with many more half-baked alternatives
Ivan Herman: And we dare not use the RDF term outload
Benjamin Young: Victor, I'd be curious about what is tripping people up. I sometimes hear from developers confusion about making things objects vs literals?
... It's either complex or properly modeled, depending on how you look at it.
Another tool Dmitri mentioned https://github.com/transmute-industries/verifiable-data/tree/main/packages/jsonld-schema
Victor Lu: I'm not an expert, but even experts are struggling with JSON-LD at times
Dmitri Zagidulin: Specifically, check out the 'json-ld to json schema' section https://github.com/transmute-industries/verifiable-data/tree/main/packages/jsonld-schema#json-ld-to-json-schema (and the reverse)
Gregg Kellogg: We'd love to keep these conversations going!
... and please plan to join us in Japan at TPAC
... we'll be back here in 2 weeks