OneDrive, Copilot, and the Emerging Knowledge Infrastructure Stack

Microsoft’s evolving vision for OneDrive provides a useful commercial example of a broader transition in knowledge infrastructure. OneDrive is no longer being positioned only as a place to synchronize and share files. With Copilot, reusable skills, collaborative Markdown and HTML, deeper SharePoint integration, AI-aware provenance, and agents that can participate in file-centered workflows, it is beginning to function as an intelligent working surface over organizational knowledge.1

This direction aligns with the knowledge infrastructure architecture I have been developing, but it does not define that architecture. OneDrive and SharePoint belong within a larger system in which durable artifacts, semantic relationships, provenance, governance, machine intelligence, and human judgment remain distinct but connected layers.

The central architectural conclusion is:

OneDrive can be an important part of knowledge infrastructure, but OneDrive is not the knowledge infrastructure.

Architectural Mapping

                         REAL WORLD
                             │
                    observation / creation
                             ↓
              ┌──────────────────────────┐
              │  SOURCE / ARTIFACT LAYER │
              │                          │
              │ OneDrive • SharePoint    │
              │ GitHub • databases       │
              │ data platforms • sensors │
              │ external evidence        │
              └─────────────┬────────────┘
                            │
                      provenance
                     identity / time
                            ↓
              ┌──────────────────────────┐
              │ GOVERNED KNOWLEDGE LAYER │
              │                          │
              │ authoritative knowledge │
              │ versions • provenance   │
              │ status • classification │
              │ ownership • permissions │
              └─────────────┬────────────┘
                            │
                     Knowledge CI/CD
                            ↓
              ┌──────────────────────────┐
              │ SEMANTIC / RELATION LAYER│
              │                          │
              │ ontology • entities      │
              │ relationships • context │
              │ knowledge graph         │
              │ culture • place • trust │
              └─────────────┬────────────┘
                            │
              APIs • Search • Connectors • MCP
                            ↓
              ┌──────────────────────────┐
              │ INTELLIGENCE / AGENT LAYER│
              │                          │
              │ Copilot • agents • RAG   │
              │ reasoning • prediction  │
              │ reusable skills         │
              └─────────────┬────────────┘
                            │
                        synthesis
                            ↓
                       HUMAN JUDGMENT
                            │
                           action
                            ↓
                    outcomes / observations
                            │
                            └──────→ new evidence

1. OneDrive as an Artifact and Persistence Layer

A mature knowledge architecture should not replace one information silo with another. GitHub may contain version-controlled research, ontologies, schemas, validation rules, and reproducible analysis. Microsoft 365 may contain working documents, spreadsheets, presentations, meeting outputs, and operational procedures. Databases may remain authoritative for structured records, while GIS, transactional applications, sensors, and external evidence retain their own roles.

The integrating principle is not physical consolidation. It is governed discoverability across authoritative systems.2

Knowledge ≠ OneDrive
Knowledge ≠ SharePoint
Knowledge ≠ GitHub
Knowledge ≠ a graph database
Knowledge ≠ PostgreSQL

Knowledge infrastructure
      ↓
governs and relates knowledge
      ↓
across all of them

This is an operational consequence of the principle that the model is not the database. The conceptual model describes the entities, relationships, meanings, evidence, and conditions that matter. Storage technologies implement portions of that model, but no single repository should be mistaken for the model itself.3

2. Microsoft 365 as a Collaborative Knowledge Surface

Microsoft 365 is especially well suited to the human collaboration surface of the architecture:

  • collaborative working documents;
  • Office files and presentations;
  • spreadsheets and operational analysis;
  • meeting-related artifacts;
  • procedures and institutional guidance;
  • comments, review, and coauthoring;
  • identity, permissions, and sharing;
  • AI-assisted synthesis across related files.

GitHub and similar version-control systems serve a different but complementary role. They are particularly useful for ontology definitions, controlled vocabularies, Markdown research, schemas, SHACL shapes, architecture documentation, validation pipelines, and reproducible analytical work.

A knowledge graph serves yet another role. It represents entities, relationships, temporal state, claims, evidence, identity, and provenance across otherwise separate artifacts and systems.

The distinction prevents any one platform from being asked to perform every knowledge function.

3. The Semantic Layer as the Missing Middle

Microsoft 365 can identify files, folders, people, libraries, projects, and activity. A semantic knowledge layer must go further by representing what those things mean and how they relate.

An ontology provides the shared concepts and relationships of a domain. A knowledge graph connects actual entities and observations using that structure.4 For example:

Person
   contributed-to → Project

Project
   resulted-in → Intervention

Intervention
   occurred-in → Community

Intervention
   intended-to-improve → Outcome

Claim
   supported-by → Evidence

Evidence
   derived-from → Dataset

Claim
   appears-in → Document

The document does not carry all of the knowledge. Relationships among documents, observations, people, places, evidence, decisions, processes, and outcomes also carry knowledge.

This is why the semantic layer should remain independent of the file repository. OneDrive may contain the artifact in which a claim appears, while the knowledge graph represents the claim’s relationship to evidence, context, community, decision, and outcome.

4. Provenance as Connective Tissue

Technical lineage asks where data came from and how it moved. Knowledge provenance asks why a claim should be believed, what supports it, under what circumstances it was produced, and how it has been transformed.5

Microsoft’s emerging ability to distinguish AI-assisted artifacts and identify agent contributions is valuable, but it represents only part of the required provenance model. A wider knowledge lineage might be:

SOURCE
  ↓
OBSERVATION
  ↓
EVIDENCE
  ↓
CLAIM
  ↓
INTERPRETATION
  ↓
DECISION
  ↓
ACTION
  ↓
OUTCOME
  ↓
NEW OBSERVATION

Each transition may need to retain:

  • who or what produced it;
  • when and where it was produced;
  • the source and method used;
  • transformations applied;
  • confidence and uncertainty;
  • authority and review status;
  • version and temporal validity;
  • AI or agent involvement;
  • assumptions and limitations.

This is knowledge lineage, not merely document history.

5. Community Knowledge Extends Beyond Enterprise Content

The architecture must support forms of knowledge that may never originate in a formal document. Conventional information systems are relatively good at storing measurable characteristics but much weaker at representing culture, context, trust, institutional history, relationships, and community interpretation.

A preliminary formulation is:

Place + Culture + Context + Relationships + Trust → Community Knowledge

Another is:

Data → Evidence → Context → Culture → Relationships → Knowledge → Action

These are not necessarily linear progressions. They are recursive and relational. Action changes conditions. New experience produces new evidence. Relationships affect interpretation. Culture influences meaning throughout the cycle.6

The architecture must therefore be able to represent relationships such as:

Library
   trusted-by → Community

Resident
   participated-in → Listening Session

Listening Session
   informs → Priority

Priority
   conflicts-with → Agency Assumption

Community
   experienced → Previous Intervention

Previous Intervention
   influences → Trust

Trust
   affects → Participation

This moves the design beyond an enterprise document system toward a community knowledge fabric. Community voice cannot simply become another database field. If local knowledge is stripped of its cultural, historical, relational, and contextual conditions, the words may remain while much of their meaning disappears.6

6. Skills as Versioned Process Knowledge

Microsoft’s concept of reusable Copilot skills fits naturally within Knowledge CI/CD. It also introduces an important distinction:

Knowledge = what we currently understand
Skill     = how we repeatedly do something

The governed knowledge environment should therefore accommodate both knowledge objects and process objects.

Knowledge objects

  • evidence;
  • observations;
  • claims;
  • models;
  • interpretations;
  • decisions.

Process objects

  • procedures;
  • methods;
  • skills;
  • workflows;
  • validation routines;
  • agent instructions.

A skill can then be represented relationally:

Skill
  ├── uses → Knowledge
  ├── requires → Permission
  ├── performs → Process
  ├── produces → Artifact
  └── creates → Provenance

A reusable skill should itself be versioned, reviewed, governed, and connected to the knowledge on which it depends. In this sense, skills become executable forms of institutional process knowledge.

7. Copilot as a Replaceable Intelligence Consumer

Copilot is not the intelligence architecture. It is one consumer and producer within it.

                 GOVERNED KNOWLEDGE
                         │
       ┌─────────────────┼─────────────────┐
       │                 │                 │
    Graph             Search            Models
       │                 │                 │
       └─────────────────┼─────────────────┘
                         │
                   ACCESS LAYER
          Connectors • APIs • MCP • GraphRAG
                         │
       ┌─────────────────┼─────────────────┐
       ↓                 ↓                 ↓
    Copilot           Local AI          Agents
       ↓                 ↓                 ↓
       └─────────────────┼─────────────────┘
                         ↓
                       PEOPLE

This separation allows AI systems to change without requiring the organization to rebuild what it knows. Microsoft Copilot, locally hosted models, specialized agents, causal models, Bayesian systems, and future interfaces can all operate over the same governed knowledge foundation.2

The durable asset is the knowledge environment, not the current assistant.

8. A Unified Learning Loop

Several strands of the architecture can be understood as different resolutions of the same learning cycle.

One formulation extends DIKW into a cycle of judgment and institutional learning:

Data → Information → Knowledge → Understanding → Judgment → Action → Outcomes → New Data5

A representation-and-action formulation is:

Reality → Observation → Representation → Abstraction → Prediction → Planning → Action → New Observation7

These can be combined into a unified architecture:

REALITY
   ↓
OBSERVATION
   ↓
EVIDENCE
   ↓
REPRESENTATION
   ↓
SEMANTIC RELATIONSHIPS
   ↓
KNOWLEDGE
   ↓
UNDERSTANDING / MODELS
   ↓
HUMAN + MACHINE REASONING
   ↓
JUDGMENT
   ↓
ACTION
   ↓
OUTCOMES
   ↓
NEW OBSERVATIONS
   └────────────────↺

Running horizontally across the cycle are:

PROVENANCE
IDENTITY
TIME
AUTHORITY
PERMISSIONS
CULTURE
CONTEXT
VERSIONING
GOVERNANCE

Six Architectural Planes

The emerging framework can be summarized as six planes.

1. Observation plane

Sensors, datasets, documents, meetings, field observations, lived experience, and community knowledge.

2. Artifact and persistence plane

OneDrive, SharePoint, GitHub, databases, data lakes, GIS, operational systems, archives, and external evidence.

3. Semantic plane

Ontologies, entities, identity resolution, relationships, knowledge graphs, context, and temporal meaning.

4. Trust and governance plane

Provenance, authority, versioning, classification, identity, permissions, culture, stewardship, and uncertainty.

5. Intelligence plane

Search, GraphRAG, causal models, Bayesian networks, Copilot, specialized agents, predictive systems, and reusable skills.

6. Human-action plane

Interpretation, judgment, decisions, interventions, outcomes, reflection, and learning.

Knowledge CI/CD is not a seventh plane. It is the lifecycle that moves knowledge safely through the architecture:

create → validate → review → approve → publish
      → index → use → monitor → revise

Its governing boundary remains:

Machines validate structure. People validate meaning.

The Role of Human Judgment

The architecture should not culminate in AI. It should culminate in better human judgment.

                 HUMAN JUDGMENT
                        ↑
                 AI / AGENTS / SKILLS
                        ↑
                  KNOWLEDGE ACCESS
                        ↑
                GOVERNED KNOWLEDGE
                        ↑
               SEMANTIC RELATIONSHIPS
                        ↑
             ARTIFACTS + OBSERVATIONS
                        ↑
                      REALITY

The cycle then returns through action, outcome, observation, and learning.

The objective is not to automate wisdom. It is to improve the evidence, context, memory, and feedback available to people exercising judgment.5

Conclusion

Microsoft’s OneDrive direction is not the knowledge infrastructure architecture itself. It is evidence that a major platform is evolving in a direction that makes such an architecture increasingly practical.

OneDrive and SharePoint can provide a capable artifact, collaboration, identity, permission, and governance surface. Copilot and agents can consume and produce knowledge through that surface. Markdown-based skills can make institutional processes reusable. AI contribution tracking can strengthen artifact provenance.

The larger architecture must still preserve distinctions among artifacts, semantics, claims, provenance, context, culture, trust, reasoning, and judgment. It must connect Microsoft 365 to GitHub, databases, graphs, operational systems, external evidence, and local knowledge without allowing any one platform to become synonymous with organizational knowledge.

The durable proposition is therefore:

Knowledge infrastructure is the persistent semantic, technical, and governance environment through which observations become evidence, evidence becomes knowledge, knowledge informs judgment and action, and the results return as new institutional and community knowledge accessible to both people and computational agents.

Microsoft 365 can become a powerful participant in that environment. The architecture remains larger, more durable, and intentionally independent of the current AI interface.

References

  1. Microsoft, “OneDrive Just Got a Major Upgrade: New Copilot Capabilities, Simpler Sharing, and Smarter Photos,” Microsoft Community Hub, 2026. https://techcommunity.microsoft.com/blog/onedriveblog/onedrive-just-got-a-major-upgrade-new-copilot-capabilities-simpler-sharing-and-s/4565995 ↩

  2. Jason F. Kronemeyer, “Building Knowledge Infrastructure for an AI-Ready Organization,” working paper, 2026. Internal source: building-knowledge-infrastructure-for-an-ai-ready-organization.html. ↩ ↩2

  3. Jason F. Kronemeyer, “The Model Is Not the Database: Why Learning Systems Should Think in Graphs Before They Store Them,” 2026. https://www.jasonkronemeyer.com/knowledge-infrastructure/educational-technology/data-architecture/2026/11/12/the-model-is-not-the-database.html ↩

  4. Jason Kronemeyer, “Ontologies and Knowledge Graphs: Differences and Relationship,” research note, 2025. Internal source: Ontologies and Knowledge Graphs.docx. ↩

  5. Jason F. Kronemeyer, “DIKW and Learning Infrastructure,” research brief, 2026. Internal source: research-brief-dikw-learning-infrastructure.html. ↩ ↩2 ↩3

  6. Jason F. Kronemeyer, “Culture, Context, and Community Knowledge: A Missing Dimension in Digital Opportunity,” research note, October 23, 2026. Internal source: 2026-10-23-doin-culture-context.md. ↩ ↩2

  7. Jason Kronemeyer, “Beyond LLMs: World Models, Abstraction, and the Next Architecture of AI,” research note, October 30, 2026. Internal source: 2026-10-30-note-world-model-abstraction.md. ↩