Re: Technical Notes, public web agents

Hi Paola

Regarding the chronology and dependency analysis in the document, I would
clarify the sequence slightly.

The work actually started from the *implementation layer*. *AgentIDL was
first extracted from the system implementation* as an interface language
describing agent interactions and execution semantics.

The *Agent Ontology was defined afterwards* to formalize the entities and
relations already present in the interface layer (agents, capabilities,
delegation, contracts, etc.).

The alignment block I posted on Dec 2 mapping these classes to the *Ontic
Categories* framework was therefore an *exploratory semantic alignment
exercise*, intended to evaluate whether the framework could improve
cross-ontology interoperability.

During that alignment exercise, some adjustments were indeed made to the
ontology structure based on the Ontic Categories comparison. However, this
should be understood as *iterative ontology refinement*, rather than a
design dependency.

In other words, the development sequence was roughly:

implementation → AgentIDL → Agent Ontology → alignment with Ontic
Categories.

陳信屹 <tyson@slashlife.ai> 於 2026年3月7日週六 下午4:01寫道：

> Paola,
>
> Thank you for the detailed note and for sharing the Ontic Categories work.
>
> A couple of clarifications first:
>
>
>    1.
>
>    The agent ontology namespace is already published at
>
>    https://s-agent-comm.github.io/agent-ontology/latest/index.html
>
>    Current version is *v0.4.0*.
>    2.
>
>    The repository README already includes a reference to the *Unified
>    Ontic Ontology* (DOI: 10.6084/m9.figshare.30760739). I will add the
>    specific commit reference shortly for clarity.
>
> On the technical questions you raised:
>
>
>    -
>
>    *Proof / Verification artefacts* are indeed important for agent
>    systems (e.g., attestations, delegation proofs, execution records).
>    -
>
>    Your observation about the *KR Interface* gap is also accurate. In
>    practice, AgentIDL’s compiler and schema layer operate as the bridge
>    between conceptual models and executable agent interactions.
>
> We already have engineering implementations for these layers in the
> current system, so these categories likely deserve clearer treatment at the
> ontology level. However, this probably warrants a *separate thread* so
> the discussion can focus specifically on ontology structure.
>
> More broadly, AgentIDL is designed primarily as an *executable interface
> layer for agent interoperability*, rather than a top-level ontology. For
> that reason, AgentIDL currently *does not have a hard dependency on Agent
> Ontology*, and we want to preserve flexibility for mapping to multiple
> ontology frameworks in related vertical domains.
>
> That said, my current inclination is that we should eventually define a *minimal
> agent ontology* (which is 0.4.0) providing the basic constraints needed
> for interoperable systems. In practice, those constraints would likely
> surface directly at the *IDL interface layer*.
>
> Regarding publication and process: since AgentIDL may evolve within the CG
> ecosystem, it would make sense to discuss the *publication and review
> process within the CG* before any formal adoption.
>
> Looking forward to continuing the discussion.
>
> Best,
>
> Tyson
>
> Paola Di Maio <paola.dimaio@gmail.com> 於 2026年3月7日週六 下午3:17寫道：
>
>> Thanks Andrei and everyone
>>
>> I  ll continue to try to capture key aspects of the Agentic AI landscape,
>> although the scope of work for AI KR includes
>> capturing and representing also non agentic AI, So in essence we are
>> looking at
>>
>> (See the table below from the W3C AI KR CG)
>> [image: AI landscape W3C AI KR CG]
>> The key challenge for many of us is to form and maintain a coherent view
>> of the all the multiple versions of the truth
>> developing across layers *protocols, fragmented representations, efforts,
>> standards etc  without losing the plot
>>
>> We cannot have meaningful  interoperability let alone trust and security
>> in Agentic AI without an ontology (the plot)
>>
>> Especially when confounding factors come into play *accidental or
>> deliberately injected misrepresentations
>>
>> In this AI agents frenzy, it is easy to lose orientation
>>
>> I have started working towards this disambiguation on the AI KR CG and
>> shared some of the work
>> on this list in this short talk  *link below in case someone missed it
>>
>>
>> 9th Feb 2026  *can be replayed at x1.2 speed Paola Di Maio on Agentic
>> Ontology
>>
>> https://drive.google.com/file/d/1G6igFKX9C7B5FC1La_3gO7DjFJsNQ1Ft/view?usp=sharing
>>
>> I am sharing this work across a number of forums, although unfortunately
>> I am often met with  walls
>> of silence, trolling, plagiarism and misattribution *I mentioned this
>> concern on the AI KR public list
>> and chatted with Tyson about it in a zoom call
>>
>> Nice to see Tyson is moving on with the agent ontology which he
>> discussed on the AI KR list
>> https://lists.w3.org/Archives/Public/public-aikr/2025Dec/0011.html
>>
>> Now here  *I joined the S agent CG just now! bear with me while I catch up
>> https://lists.w3.org/Archives/Public/public-s-agent-comm/2026Feb/0000.html
>>
>> We should continue to work on that, making sure we do not lose the
>> continuity between efforts
>>
>> I shared with the AI KR GG list and then privately with Tyson specific
>> resources
>> * specifically the Ontic Categories
>> https://lists.w3.org/Archives/Public/public-s-agent-comm/2025Dec/0001.html
>> In the form of a diagram and a table of unified ontic categories,  I also
>> created a DOI
>> https://figshare.com/articles/online_resource/Ontic_Categories/30760739?file=60028508
>> as everyone is surely aware of  rampant IP theft all over the internet
>>
>>  The diagram in the screenshot below is still public in the AI KR CG wiki
>> , but I have shared privately the table of ontic categories pending
>> negotiation with others who may be interested to collaborate on applying
>> for funding, still unsure if this should be in the public domain, so happy
>> to discuss possibilities in a call
>>
>>
>> I also note that the IDL Ontology  now references the ontic categories I
>> presented in the AI KR CG. Am I right?
>>
>> I have mapped the chronology of the interactions here,
>> IDL Mapping to Ontic Categories
>> <https://docs.google.com/document/d/1Y2rAXfZGZn1bmZJ4FJvWa0zwguM64L0J/edit?usp=sharing&ouid=100993141196967133475&rtpof=true&sd=true>
>>
>> Tyson perhaps can you confirm the mappings?
>>
>>
>> I look forward to the collaboration and to get a better grasp of the AI
>> KR domain and how it is evolving
>>
>> Best regards
>>
>> Paola Di Maio
>> Chair, W3C AI KR CG
>>
>>
>> Agent IDL/Ontic Analysis doc
>> <https://docs.google.com/document/d/1Y2rAXfZGZn1bmZJ4FJvWa0zwguM64L0J/edit?usp=sharing&ouid=100993141196967133475&rtpof=true&sd=true>
>> *excerpt
>> 4.2 Technical Gaps Identified
>>
>> The Ontic Categories framework currently has no adequate category for
>> verifiable proofs, cryptographic attestations, or execution records (the
>> [Proof] annotation gap). Should a Trust/Verification sub-bubble be added?
>>
>> The "KR Interface" bubble (the category for things like the AgentIDL
>> compiler -- mechanisms that bridge ontological definition and executable
>> code) was identified as missing on November 28, 2025. AgentIDL's compiler
>> is a concrete instantiation of this gap. Should this now be formalised as a
>> new category in the Ontic Categories map?
>>
>> The contract:Contract class is simultaneously mapped as ArtifactSocial
>> and UFO:SocialRelator. These are ontologically distinct (artifact vs.
>> relation). Which reading should be canonical in the AgentIDL context?
>>
>> The ledger:Ledger class is mapped as ArtifactSocial but BFO suggests
>> InformationArtifact. Is there a sub-category distinction between
>> information-bearing artifacts and purely social artifacts needed in the
>> Ontic Categories?
>>
>> The s-agent-comm.github.io/agent-ontology namespace (which AgentIDL
>> imports) returns 404. Before any CG adoption or endorsement, this ontology
>> must be published and reviewed. Should publication of the agent-ontology be
>> a prerequisite for further CG engagement with AgentIDL?
>>
>> 4.3 Proposed Next Steps
>>
>> Based on the above, the following actions are proposed for discussion by
>> the CG:
>>
>> Tyson to confirm exact repository creation dates and commit history, and
>> to add explicit citation of the Unified Ontic Ontology (DOI:
>> 10.6084/m9.figshare.30760739) in the AgentIDL README and ontology headers.
>>
>> The CG to formalise the "KR Interface" category in the Ontic Categories
>> map, using AgentIDL's compiler as a concrete example/use case.
>>
>> The CG to consider whether a Trust/Verification sub-category should be
>> added to the Ontic Categories to cover ProofBinding and related
>> verification artefacts.
>>
>> Clarify the governance relationship between the s-agent-comm CG (i as a
>> separate W3C CG) and the AI KR CG to avoid parallel, uncoordinated
>> standards development.
>>
>> Review W3C CG IP policy with respect to public mailing list content used
>> as the basis for external specifications,
>>
>>
>>
>>
>> On Sat, Mar 7, 2026 at 9:27 AM 陳信屹 <tyson@slashlife.ai> wrote:
>>
>>> Hi Andrei, Paola
>>>
>>> Thanks for sharing, looking for the release of interoperability report.
>>>
>>> Here the related works in s-agent-comm CG:
>>>
>>> The agent ontology with the currently available documentation is here:
>>> https://github.com/s-agent-comm/agent-ontology
>>>
>>> The AgentIDL reference implementation is here:
>>> https://github.com/s-agent-comm/agent-idl
>>>
>>> While the ontology layer describes what an agent is (roles,
>>> capabilities, contracts, etc.), AgentIDL defines how an agent acts and
>>> speaks. It serves as a "Semantic API" that can be compared to: OpenAPI/IDL
>>> for the web, but operating at a semantic level. gRPC's .proto files, but
>>> with support for intents, trust, and grammar. Solidity's function
>>> signatures, but for agent behavioral protocols. Semantically, it combines:
>>> Ontology (meaning) + Grammar (syntax) + Protocol (execution) This creates
>>> an agent-level Application Behavior Interface (ABI).
>>>
>>> Andrei Ciortea <andrei.ciortea@inria.fr> 於 2026年3月7日週六 上午3:31寫道：
>>>
>>>> Dear Paola,
>>>>
>>>> Thank you for pushing this topic forward. Knowledge representation for
>>>> AI agents and multi-agent systems is clearly of interest to several active
>>>> participants in the WebAgents CG.
>>>>
>>>> The ongoing work on the interoperability report will hopefully help
>>>> outline requirements and needs for knowledge engineering efforts in this
>>>> space. The report will likely be a good place to connect and consolidate
>>>> related contributions.
>>>>
>>>> I also agree that it would be useful to strengthen interaction and
>>>> alignment not only within the WebAgents CG, but also with related groups
>>>> such as the AIKR CG and the S-Agent-Comm CG. I will bring up this point in
>>>> the next regular meeting.
>>>>
>>>> Regarding the initiative mentioned by Antoine in the last meeting —
>>>> this refers to the Hypermedia Multi-Agent Systems (hMAS) ontology, which is
>>>> currently organized into 3 modules:
>>>> - hmas-core: http://purl.org/hmas/core
>>>> - hmas-interaction: https://purl.org/hmas/interaction
>>>> - hmas-regulation: https://purl.org/hmas/regulation
>>>>
>>>> The GitHub repository with the currently available documentation is
>>>> here: http://github.com/hyperagents/hmas
>>>>
>>>> For additional context, this paper shows how we use the hMAS ontology
>>>> in one of our frameworks for Web-based MAS (presented in a past regular
>>>> meeting): https://link.springer.com/chapter/10.1007/978-3-032-01082-7_7
>>>>
>>>> Best wishes,
>>>> Andrei
>>>>
>>>>
>>>>
>>>> ------------------------------
>>>>
>>>> *De: *"Paola Di Maio" <paola.dimaio@gmail.com>
>>>> *À: *public-webagents@w3.org
>>>> *Cc: *"Autonomous Agents on the Web Community Group" <
>>>> public-webagents@w3.org>
>>>> *Envoyé: *Dimanche 1 Mars 2026 09:47:54
>>>> *Objet: *Technical Notes, public web agents
>>>>
>>>> Dear Andrei and everyone
>>>>
>>>> Thanks for listening to my presentation at the last meeting
>>>>
>>>> The bottom line from a AI KR perspective is that
>>>> there can be no security/safety/interoperability in AI until we have a
>>>> clear Knowledge representation/conceptual model /ontology for the domain
>>>>
>>>> I am a bit surprised that the information technology community has been
>>>> silent about knowledge representation in AI/Agents
>>>>
>>>> Just to say that it would be great to continue discussions via this
>>>> mailing list  in between meetings
>>>> as things continue to happen that need  our constant attention
>>>>
>>>> Please remind us where we are in this CG from time to time!
>>>>
>>>> I would also love to hear from other participants what they are working
>>>> on, especially on the agent interoperability front
>>>> *I have more to share on that topic if/when you d like to hear about it
>>>>
>>>> The latest  note from me is shared via some other CGs
>>>> https://lists.w3.org/Archives/Public/public-aikr/2026Feb/0015.html
>>>>
>>>> and the TN here
>>>> https://github.com/Starborn/webmcp/blob/main/TN4.md
>>>>
>>>>  can benefit from being discussed
>>>> The priority is to make sure that things do not happen above our head
>>>> without understanding what is going on
>>>> because there is not enough shared knowledge/understanding about what
>>>> is taking place
>>>>
>>>> This knowledge fragmentation is something that can be engineered to
>>>> create the vacuum where very undesirable things can happen
>>>> *that Agenti AI trolls take over the web :-)
>>>>
>>>> I would also like to continue contributing to the interoperability
>>>> report AND hear more about the hyperagent
>>>> *was it the hyperagent we were talking about that you mentioned?
>>>>
>>>> There is so much going on, it's nice to be reminded how the thread are
>>>> holding together
>>>>
>>>> We do important work but there is lack of interaction and things happen
>>>> very fast in the world of web ai agents etc
>>>>
>>>> Have a great weekend everyone
>>>>
>>>> Paola
>>>>
>>>> On Fri, Feb 27, 2026 at 6:58 PM Andrei Ciortea (W3C Calendar) <
>>>> noreply+calendar@w3.org> wrote:
>>>>
>>>>> View this event in your browser
>>>>> <https://www.w3.org/events/meetings/c5855bad-fd18-45ce-973f-e0f716dd6bed/20260227T160000/>
>>>>> W3C WebAgents CG: Biweekly Call (Fridays) Upcoming Canceled
>>>>>
>>>>> 27 February 2026, 16:00 -17:00 Europe/Zurich
>>>>>
>>>>> Event is recurring Every 4 weeks on Friday, starting from 30 January
>>>>> 2026, until 1 January 2027
>>>>> Autonomous Agents on the Web Community Group
>>>>> <https://www.w3.org/groups/cg/webagents/calendar/>
>>>>>
>>>>> Biweekly call of the W3C Autonomous Agents on the Web (WebAgents)
>>>>> Community Group.
>>>>>
>>>>> For more information about regular meetings, see the group's wiki
>>>>> <https://github.com/w3c-cg/webagents/wiki/Regular-CG-Meetings>.
>>>>>   Participants Organizers
>>>>>
>>>>>    - Andrei Ciortea
>>>>>    - Rem Collier
>>>>>    - Ege Korkan
>>>>>    - Antoine Zimmermann
>>>>>
>>>>> Groups
>>>>>
>>>>>    - Autonomous Agents on the Web Community Group
>>>>>    <https://www.w3.org/groups/cg/webagents/> (View Calendar
>>>>>    <https://www.w3.org/groups/cg/webagents/calendar/>)
>>>>>
>>>>> Report feedback and issues on GitHub <https://github.com/w3c/calendar>.
>>>>>
>>>>>
>>>>
>>>>