Documentation

Context Layers

A side-by-side exploration of how humans and AI systems build expertise through progressive context loading

The Universal Principle

Creating expertise requires segmented context delivered in progressive layers. A new employee cannot be productive on day one with only general skills; they need company context, department standards, project knowledge, and task-specific details loaded sequentially.

Similarly, an AI system cannot be effective with only base training; it needs user preferences, workspace standards, project documentation, and runtime tools loaded in order.

Context organization transforms generalists into specialists. Without layered context, you have capable individuals who cannot execute effectively. With proper context layering, you build domain experts who are immediately productive.

0

Baseline Knowledge

General education and training before specialization

🧑‍💼

College Graduate

General Education

What They Know

  • Programming languages (Python, Java, C++)
  • Algorithms and data structures
  • Software design patterns
  • Academic projects

Capability

Can write general-purpose code, but can't deploy to your infrastructure or follow your standards.

🤖

AI Model

General Training (Claude, ChatGPT, Gemini)

What It Knows

  • Programming languages (Python, Java, C++)
  • Common algorithms and patterns
  • General software practices
  • Public documentation

Capability

Can write general-purpose code, but doesn't know your infrastructure, tools, or project specifics.

1

Broad Context

Company-wide knowledge and user preferences

🏢

Company Onboarding

First Week

What They Learn

  • Company mission and values
  • Common tools (Slack, GitHub, Jira)
  • Security protocols and badges
  • Organizational structure

Example Training

  • Style guide — "All REST endpoints use kebab-case"
  • IT helpdesk — "File a ticket when your VPN drops"
  • Security badge — "Never share door codes"
📁

User Level Context

Session Start • ~/.claude/

What It Learns About You

Example Skills

  • naming-validator — checks that new services follow naming rules
  • troubleshooting — step-by-step guides for diagnosing problems
  • secrets-security — keeps passwords stored safely, never in code
2

Domain Context

Department standards and workspace configuration

🏫

Department Training

First Month

What They Learn

  • Tech stack (Python, PostgreSQL, Kubernetes)
  • Architecture patterns (microservices, event-driven)
  • Deployment pipeline (CI/CD, staging → production)
  • Coding standards and network topology

Example

"Platform Engineering uses Docker on Kubernetes. All services integrate with Keycloak SSO and follow the 3-network pattern. Here's our standards doc."

🗂

Project Level Common Context

Workspace Entry • ~/projects/.claude/

What It Loads

Example Skills

  • service-deployment — standard patterns for deploying new services
  • keycloak-setup — adding single sign-on to applications
  • postgres-integration — connecting apps to the database
3

Specific Context

Project-specific knowledge and role assignment

🎯

Role Assignment

First Quarter

What They Learn

  • Assigned service (architecture, APIs, dependencies)
  • Institutional knowledge from team leads
  • Access credentials and permissions
  • Known issues and escalation contacts

Example

"You own the Team Chat Platform. It runs two services — a message gateway and a notification engine. Teammates communicate through channels, can assign work to each other, and there are 267 tests to keep it stable."

📝

Project Specific Context

Project Entry • ~/projects/nginx/

What It Loads

  • Project CLAUDE.md (architecture, APIs, dependencies)
  • Auto-memory (lessons learned across sessions)
  • Project permissions and tool access
  • Known issues and troubleshooting guides

Example

"You're working on AI Agent Chat. It runs two containers (gateway + agents), uses Matrix for messaging, handles task delegation between agents, and has 267 tests to keep it stable."

4

Task Context

Specific assignment details and runtime tools

📋

Assignment Details

Sprint Planning

What They Receive

  • Business requirements
  • Technical approach and patterns
  • Test scenarios and edge cases
  • Deployment plan

Tools Learned On-Demand

Postman for API testing, Jira workflow for ticket tracking, specific APIs or libraries needed for the task.

Run-Time Tool Loading

Work Execution • On-Demand

What It Loads

Tools Loaded On-Demand

Skills invoked (keycloak-setup, postgres-integration), database queries, web fetches, file operations.

💡 The Universal Pattern

Without layered context, you have generalists.

With layered context, you build specialists.

This pattern applies to humans, AI systems, and any entity that needs to become productive in a specialized domain. Context organization is what transforms capability into expertise.

5

Collaboration

Sharing expertise across teams and getting diverse perspectives

👥

Cross-Team Knowledge Sharing

Enterprise-Wide

Sharing With Peers (Same Company)

  • Architecture Review Boards document decisions
  • Shared wikis and Confluence spaces
  • Cross-team Slack channels (#platform-eng)
  • Architecture Decision Records (ADRs) for other teams

Consulting External Experts

  • Hire a security consultant for audit
  • Engage AWS solutions architect
  • Stack Overflow for community wisdom
  • Conference talks and industry peers

Why Multiple Perspectives Matter

Different experts bring different training, biases, and experiences. A security expert sees risks a developer might miss. An external consultant spots patterns your team is blind to.

🌐

AI Review Board Architecture

Multi-Agent • Multi-Model

Sharing With Peer Agents (File-Based)

  • PM Agent: Orchestrates tasks, writes breakdowns
  • Architect Agent: Writes ADRs, design specs
  • Security Agent: Documents network, auth configs
  • Developer Agent: Reads specs, writes deployment notes

Agents share context via $HOME/projects/data/claudeagents/

Cross-Server Sharing (Shared AI Chat)

  • Server Admin: Writes to linuxserver.administrator.md
  • Laptop Dev: Writes to laptop.websurfinmurf.md
  • File sharing via MinIO aichat-files bucket (MCP tools)

Consulting Different Models (MCP Dispatch)

  • Gemini: Primary consultant, has internet search
  • Codex (ChatGPT): Alternate perspective when needed
  • Claude (Sonnet): Deep analysis, architecture review

All nodes have read-only access to /workspace (projects)

Why Multiple Models Matter

Each AI has different training data, reasoning patterns, and blind spots. Gemini can search the web. ChatGPT has different coding style preferences. Getting multiple perspectives catches errors a single model might miss.

🤝 The Collaboration Pattern

Just as senior engineers share knowledge through documentation and consult external experts for fresh perspectives, AI systems can share context through file-based protocols and consult different models for diverse viewpoints.

File-based sharing enables asynchronous coordination across sessions.
Multi-model consultation catches blind spots through diverse "training backgrounds."