Portable Intelligence

AI models are powerful collaborators. They are poor places to keep the durable version of your work.

Conversations disappear into account histories. Company systems become inaccessible when roles change. Prompt libraries become tied to one platform. Context fragments across tools.

Portable Intelligence is a framework for building a durable layer beneath those interfaces.

The repository—not the AI—is the durable layer.

The goal is not to become independent of AI. It is to become independent of any single AI model, employer, software platform or interface.

The problem

Modern knowledge work is increasingly fragmented across:

• AI conversations
• company document systems
• email and messaging
• notes applications
• prompt libraries
• individual files and presentations

Each tool can be useful. The problem begins when the tool becomes the only place where the knowledge exists.

When the interface changes, the context disappears. When someone leaves a role, the method often leaves with them. When a model is replaced, teams rebuild working context from scratch.

Portable Intelligence treats those tools as temporary interfaces to a more durable system.

The portable work layer

The framework introduces a layer between raw work and the tools used to produce outputs.

Capture
  ↓
Structure
  ↓
Repository
  ↓
AI models and human collaborators
  ↓
Outputs
  ↓
Learning
  ↓
Improve the repository

The repository is not necessarily GitHub. It could be another well-governed knowledge system. What matters is that it is canonical, portable, legible and reusable.

AI models can then read from it, work with it and contribute back to it. They become replaceable collaborators rather than containers for the work itself.

The five principles

1. Canonical

Maintain one recognised source of truth.

A project can have many outputs, conversations and working documents, but there should be one durable place where the current method, evidence and decisions are maintained.

Without a canonical source, teams spend time reconciling versions and reconstructing context.

2. Portable

Use durable formats and structures that can move between tools.

Markdown, plain text, CSV, JSON, standard code and well-labelled media are easier to reuse than knowledge locked inside one proprietary interface.

Portability does not mean everything must be public. It means the work can survive a change of tool.

3. Legible

Another human should understand the repository without needing the original AI conversation.

That requires clear naming, documentation, decision records and enough context to explain why something exists.

A folder of outputs is not a knowledge system. The reasoning connecting those outputs must also be visible.

4. Executable

The repository should produce useful work.

It may generate:

• websites
• articles
• presentations
• products
• workflows
• assessments
• research briefs

Executable knowledge is more valuable than archived knowledge because it can be activated repeatedly.

5. Living

Every project should improve the system beneath it.

New evidence should strengthen the evidence base. New decisions should improve the method. New outputs should reveal gaps in the repository.

The repository compounds because each use leaves it more capable than before.

Traditional workflow versus Portable Intelligence

Traditional workflow

Idea
  ↓
AI conversation
  ↓
Output copied elsewhere
  ↓
Context lost
  ↓
Start again

Portable Intelligence

Idea or signal
  ↓
Canonical repository
  ↓
ChatGPT / Claude / Copilot / Codex / future tools
  ↓
Website / article / product / workflow
  ↓
Learning returned to the repository

The output is no longer an endpoint. It becomes another source of learning for the system.

Two working examples

Editorial Intelligence Core

The Editorial Intelligence Core repository stores:

• frameworks
• principles
• workflows
• prompts
• article ideas
• experiments

The website is the publishing layer. AI tools help develop and activate the material, but the repository remains the canonical source.

That makes the methodology reusable across models and publication formats.

Thai learning project

A personal Thai learning repository contains:

• an interactive lesson engine
• song lessons
• guided stories
• writing exercises
• product decisions
• a build log

The project began inside an AI conversation. Its durable value now lives in the repository, where another model—or another developer—could continue the work.

The AI helped build it. The repository owns the continuity.

Model-agnostic by design

Portable Intelligence makes AI workflows model-agnostic by design.

The workflow, evidence, instructions, examples, validation history and human decision rules should belong to the organisation rather than to ChatGPT, Claude, Microsoft Copilot or any other interface. A model may execute part of the workflow, but it should not become the only place where the workflow can be understood or maintained.

This does not mean every model will produce identical results. Models have different capabilities, tool access and behaviour. Portability means the durable editorial knowledge can be moved, tested and adapted without reconstructing the thinking from zero.

Model-agnostic does not mean model-indifferent. The durable context layer should survive a change of model, while the execution layer still makes deliberate choices about which model is best for a task and how the surrounding workflow needs to adapt.

A more realistic architecture is:

Durable context
  ↓
Adaptable harness
  ↓
Selectable models

The model is replaceable. It is not irrelevant. The accumulated knowledge and judgement are not interchangeable at all.

That distinction becomes particularly important for learning editorial systems, where feedback from real work improves the instructions and checks used on the next project. The learning belongs in the canonical knowledge layer. The current AI agent is the interface through which that learning is applied.

The market is moving towards the same separation

This is no longer only a design preference inside Editorial Intelligence. Different parts of the enterprise AI market are increasingly separating durable organisational context from the models that use it.

Notion is giving organisations more control over which models agents can use while keeping workspace context and agent configuration intact. Glean puts enterprise search, permissions and organisational context beneath access to multiple models. Snowflake is developing governed model routing around the data layer. Palantir’s Ontology gives different agent and model systems a shared operational representation of the organisation. Open protocols such as MCP make it easier for compatible models and agents to reach tools and governed context without every connection being rebuilt from scratch.

These systems are not equivalent, and open interfaces do not eliminate vendor lock-in or security problems. But the direction strengthens the underlying Portable Intelligence thesis: organisational context can be treated as durable infrastructure while the intelligence working over it changes.

The model is not the memory develops the evidence and the limits of that argument in more detail.

Governance matters

Portable Intelligence is not an argument for moving employer or client information into personal repositories.

A durable system must distinguish clearly between:

• personal intellectual property
• public work
• client-owned work
• employer-owned work
• confidential information

Portability applies to methods, reusable structures, non-confidential learning and appropriately owned assets. Good governance is part of the framework, not a restriction added afterwards.

Adoption checklist

Before starting or restructuring a project, ask:

• Where is the canonical source of truth?
• Could another AI model continue this work?
• Could another human understand it without the original conversation?
• Are the important files stored in durable formats?
• Are decisions and rejected ideas documented?
• Does the repository produce outputs, or only store them?
• Does each new project improve the underlying system?
• Are ownership and confidentiality boundaries explicit?

Relationship to Editorial Intelligence

Editorial Intelligence explains how organisations turn distributed signals into evidence, narrative, action and learning.

Portable Intelligence defines the durable infrastructure beneath that cycle.

It answers a different question:

Where should knowledge live so that it remains usable across tools, teams, AI models and time?

Together, the two ideas create a stronger operating model. Editorial Intelligence develops and activates knowledge. Portable Intelligence ensures that knowledge can compound.

Framework status

Draft v0.1 — actively evolving through real projects.

The framework emerged through building rather than theory alone. Future versions will test it across organisational, consulting and personal knowledge systems.