Role card
AI agent as proposer: the functional role
What is the role of the AI agent in an adaptive semantic architecture — tool, source of evidence or decision maker?
A role card explains a functional role: what has to be done, not a job title. One person can hold several of them, and one can be spread over several people.
The short answer
In Michael Iantosca’s white paper the AI agent becomes both a consumer of semantic knowledge and a source of evidence about where the model can improve. As a functional role, it answers and acts on the graph, its limitations surface as signals of semantic gaps, and in the adaptive layer it may draft candidate changes with their evidence. It never approves a change, never edits the governed core and never decides on safety-related information.
For: Documentation systems developers, technical project managers and technical writers
Key points
- Uses the graph, inference and the vector index to answer and act, and leaves a record of what it used.
- Its recurring limitations are evidence for semantic gaps.
- May draft candidates in the adaptive layer; approval stays with people.
Functional role
AI agent as proposer
The role in one sentence
The function of the AI agent in the evolution cycle: it uses the semantic model, its experience becomes evidence, and it may draft candidate changes — it never approves them.
Responsibilities
- Answer and act on the graph, the inference engine and the vector index.
- Leave a traceable record of the sources and graph paths it used.
- Surface recurring limitations as signals of semantic gaps.
- Draft candidate changes with their evidence, in the adaptive layer only.
Decides
- Which sources and graph paths it uses for an answer, within its configuration.
Does not decide
- Any approval or promotion of a change.
- Edits to the governed semantic core.
- Leaving out, shortening or overriding safety-related or required information for use.
Receives
- The governed graph and the effective semantic view.
- The vector index.
- User questions and tasks.
Hands over
- Answers with their sources.
- Usage evidence.
- Draft candidate changes.
Typically held by
Reader roles of the pathway map that usually take on this function.
- Documentation systems developers
- Configure the agent and what it records.
- Technical project managers
- Decide which evidence is kept and who analyzes it.
On the map
- AI agents on the graph· From knowledge graphs to adaptive semantics: the pathway map
Answers and acts on the graph.
- Agent experience as evidence· From knowledge graphs to adaptive semantics: the pathway map
Its use becomes evidence.
- Seven signals of a semantic gap· From knowledge graphs to adaptive semantics: the pathway map
Its limitations become signals.
- Candidate change (typed edit)· From knowledge graphs to adaptive semantics: the pathway map
May draft candidates.
The context
A function, not a person
Calling an AI agent a role is deliberate. In the evolution cycle the agent does work that has to be accounted for like any other: it answers from the knowledge graph and the index, it shows where the model is weak, and it may propose. Treating that as a functional role makes clear what it may do — and where it has to stop.
Where the role ends
The agent’s experience is evidence; it is not a decision. A candidate it drafts goes through evaluation, validation and governance like every other. Required information for use and safety-related information are never adapted away because an agent performs better without them. The agent is configured by documentation systems developers; what it records and who analyzes it is agreed beforehand.
Questions readers ask next
- Can the agent change the ontology directly?
- No. Candidate changes live outside the official model until governance approves them; the system never promotes anything by itself.
- Is every agent a proposer?
- Only where an adaptive layer records its experience and turns it into candidates. An agent that only answers is a consumer of the model, not a proposer.
Sources
- From knowledge graphs to adaptive semantics: evolving DITA and iiRDS digital twins for AI (white paper) — Michael Iantosca, September 25, 2026
- EvoOntology vs. iiRDS: understanding two divergent ontology paradigms — AI TechDoc Press (Saina Veigel), September 23, 2026
Review log and changes
Every context card is checked against its sources before it is published, and again whenever it changes; the date under the byline is the last review. Corrections (something was wrong) and additions (something was missing) are logged below with date and time (Berlin time). Typos, formatting and link fixes are not listed.
Reviewed
No corrections or additions since publication.