AI-First Context Infrastructure

This proposal establishes a unified AI-accessible layer ensuring all product development information (code, specs, designs, project management) is within AI's reach. This is the infrastructure layer that enables truly AI-assisted development.

Core Insight: AI effectiveness depends on the quality and completeness of accessible context. When AI can only see code but cannot access requirements, designs, and decision records, its output is necessarily limited. Building unified context infrastructure transforms AI into a true development partner.

Problem Statement

Current State: Context Silos

Evidence from Team Feedback

Pain Point	Impact on AI Workflow
Scattered Requirements	AI cannot understand full feature context, can only guess from code
Isolated Design Assets	AI-generated UI inconsistent with design specs, requires extensive manual correction
Lost Decision Records	AI doesn't know why certain design choices were made, may repeat mistakes
Unclear Project Status	AI cannot determine which features are in development vs completed
Cross-Team Information Gaps	AI cannot discover that other teams have already solved similar problems

Fundamental Challenge: Accessible ≠ Usable

Even after solving access issues, there's a deeper challenge:

Key Insight: Various platforms (Confluence, Figma, Jira, etc.) were not designed to serve as a Single Source of Truth, and humans currently lack the awareness to maintain such information properly. Therefore, even if we can access information through agents, we face three problems:
The information may be outdated
The information may be incomplete
High probability of not knowing where to look—unclear where in the platform to find the needed information

This means this proposal has two levels of objectives:

Level	Objective	Challenge
Level 1: Accessibility	Enable AI to access product development information	Technical integration and data governance
Level 2: Effectiveness	Ensure AI accesses truly meaningful information	Culture and process change (building maintenance awareness)

Level 1 Solution Paths:

Level 1 is not limited to "connecting AI directly to existing platforms." There are two complementary paths:

Path	Approach	Best For
Path A: Real-time Connection	Connect to existing platforms (Figma, Jira, etc.) via MCP Servers	Information that changes frequently, requires real-time queries
Path B: Centralized Migration	Create a new Git-based platform, migrate data from various platforms	Core knowledge, information requiring long-term maintenance

For Path B implementation details, refer to the Agent-Friendly Knowledge Base proposal, which details how to migrate from human-oriented wikis like Confluence to Git-based Markdown knowledge systems.

Level 2 Solution Strategies:

Centralize Core Knowledge: Migrate critical information from various platforms to Git-based systems, establishing clear maintenance responsibilities
Standardize Frontmatter: Use structured metadata (e.g., lastUpdated, status) to help AI assess information timeliness
Continuous Context Cleanup: Establish regular review mechanisms to keep knowledge base content updated (see Continuous Context Cleanup)
AI-Assisted Validation: Have AI proactively check document update status and consistency before using information

Proposed Solution: Hybrid AI Context Architecture

Target Architecture

Core Principles

Context Completeness: All information affecting development decisions should be AI-accessible
Hybrid Architecture: Core knowledge centralized (Git-based), external systems connected via MCP in real-time
Single Entry Point: Unified access through MCP Gateway, reducing AI's cognitive burden
Incremental Migration: Start from the most painful context silos, gradually expand coverage
Permission Alignment: AI access permissions align with user permissions, no additional exposure

Architecture Layers

Layer 1: Core Knowledge Base (Centralized)

Information requiring long-term maintenance, suitable for centralized management:

System	Content	Source Proposal
Knowledge Base	Domain knowledge, guidelines, best practices	Agent-Friendly Knowledge Base
Requirement Store	Product requirements, user stories, acceptance criteria	Global Requirement Store
Spec Platform	Feature specs, API contracts, UX specs	Internal Spec Platform
Ubiquitous Language Glossary	Term definitions, domain concepts	Ubiquitous Language
Architecture Decision Records	ADRs, technology selection rationale	To be established

Layer 2: Real-time Connections (MCP Servers)

Information that changes frequently, suitable for real-time queries:

MCP Server	Context Provided	Priority
Git MCP	Code, PRs, Issues, Commit history	P0 (Existing)
Figma MCP	Designs, Design Tokens, Component library	P1
Jira MCP	Task status, Sprint, Roadmap, Epics	P1
Confluence MCP	Existing docs (transitional use)	P2
Teams MCP	Discussion records (for search)	P3

Layer 3: MCP Gateway

Unified access entry point providing:

typescript

// MCP Gateway Concept
const mcpGateway = {
  // Unified query interface
  query: async ({ intent, scope, filters }) => {
    // Determine which resources to query based on intent
    // e.g., "find all related information for this feature"
    const results = await Promise.all([
      knowledgeBase.search(intent),
      requirementStore.search(intent),
      specPlatform.search(intent),
      figmaMcp.search(intent),
      jiraMcp.search(intent)
    ]);
    return aggregateResults(results);
  },

  // Context loading
  loadContext: async ({ feature, product }) => {
    return {
      requirements: await requirementStore.get(product, feature),
      specs: await specPlatform.get(product, feature),
      designs: await figmaMcp.get(product, feature),
      tasks: await jiraMcp.get(product, feature),
      relatedCode: await git.search(feature)
    };
  }
};

Information Flow Integration

Complete Context for Feature Development

Cross-System Traceability

Requirements              Specs                    Designs                Implementation
────────────────────────────────────────────────────────────────────────────────────────
REQ-001            →     func/device-mgmt    →   Figma/DeviceList   →   DeviceList.tsx
(User Story)             (Feature Spec)           (UI Design)            (Code)

                   →     api/device-mgmt     →                      →   deviceApi.ts
                         (API Contract)                                  (API Client)

                   →     ux/device-mgmt      →   Figma/DeviceCard   →   DeviceCard.tsx
                         (UX Spec)               (Component Design)      (Component)

Implementation Roadmap

Phase 0: Requirements Discovery (Q2 2026/Apr-Jun)

Goal: Understand current state, determine priorities

[ ] Inventory all product development information sources
[ ] Assess AI accessibility status for each source
[ ] Identify most painful context gaps
[ ] Design hybrid architecture details
[ ] Determine MCP Server development priorities

Deliverables:

Information source inventory with AI accessibility assessment report
Hybrid architecture design document
Implementation priority ranking

Phase 1: Core Knowledge Base Establishment (Q3 2026/Jul-Sep)

Goal: Build centralized core knowledge layer

Week	Work Items
1-2	Establish Git-based knowledge base repository
3-4	Migrate critical domain knowledge and guidelines
5-6	Build requirement store basic structure
7-8	Pilot one product's requirement migration
9-10	Spec platform MVP launch
11-12	Integration testing and adjustments

Deliverables:

Working knowledge base repository
One product's requirement migration complete
Spec platform MVP

Phase 2: Expansion & Integration (Q3-Q4 2026/Jul-Dec)

Goal: Expand knowledge base coverage, evaluate if real-time connection layer is needed

Primary Path: Centralize to Git

Prioritize migrating information from various platforms to Git-based knowledge base, accessed via AI's native Git MCP support:

Information Type	Source Platform	Migrate To
Design Specs	Figma	`knowledge-base/designs/` (Design Tokens, component specs)
Project Status	Jira	`knowledge-base/projects/` (Sprint summaries, Roadmap)
Technical Docs	Confluence	`knowledge-base/docs/` (migrate active content)

Backup Plan: MCP Server Development

If centralization cannot meet real-time requirements, then consider developing MCP Servers to directly connect to platforms:

MCP Server	Scope	Trigger Condition
Figma MCP	Design Token reading, component spec queries	Design changes too frequent to migrate in time
Jira MCP	Task status, Sprint info, Epic links	Need real-time Sprint status
MCP Gateway	Unified queries, context loading	When multiple MCP Servers exist

Deliverables:

Expanded knowledge base coverage
Migration process and tooling documentation
(If needed) Working MCP Servers

Phase 3: Full Integration & Rollout (Q4 2026/Oct-Dec)

Goal: Fully enable AI-First context access

[ ] All products complete knowledge base/requirement migration
[ ] MCP Gateway running stably
[ ] Developer training and promotion
[ ] Establish maintenance mechanisms and governance processes
[ ] Performance optimization and caching strategies

Deliverables:

Fully operational AI-First context infrastructure
Developer usage guide
Maintenance and governance documentation

Relationship with Existing Proposals

This proposal serves as an orchestration layer, integrating the following existing proposals:

Proposal	Role in This Architecture
Agent-Friendly Knowledge Base	Core Knowledge Layer - Domain knowledge and guidelines
Global Requirement Store	Core Knowledge Layer - Requirement management
Internal Spec Platform	Core Knowledge Layer - Spec hosting
Ubiquitous Language	Cross-Layer Shared - Terminology consistency

Success Metrics

Metric	Target	How to Measure
Context Completeness	>80% features have complete context	Audit accessible information per feature
AI Context Load Time	<5s	MCP Gateway latency monitoring
Developer Satisfaction	>4/5	Quarterly survey
Context Usage Rate	>70% of AI sessions use it	Session analysis
Cross-System Traceability	100% requirements traceable to code	Link verification

Risks and Mitigation

Risk	Mitigation
Migration Cost Too High	Use incremental migration, migrate active content first; leverage AI-assisted migration tools
Real-time Requirements Unmet	Evaluate actual requirements first; consider MCP Server development as backup if Git sync is insufficient
Permission Management Complexity	Leverage Git's existing permission system; add access control at knowledge base level
Information Freshness	Establish regular sync mechanisms; use frontmatter `lastUpdated` for AI to assess timeliness
Team Adoption Resistance	Start from pain points, demonstrate concrete value

G1: Single Source of Truth - Unified context infrastructure as canonical source
C1: Context Engineering Competency - Building effective AI context
C3: AI First - AI-native infrastructure design

Workflow Framework

Guiding Principles

Governance

Culture

Execution

AI-First Context Infrastructure

Problem Statement

Current State: Context Silos

Evidence from Team Feedback

Fundamental Challenge: Accessible ≠ Usable

Proposed Solution: Hybrid AI Context Architecture

Target Architecture

Core Principles

Architecture Layers

Layer 1: Core Knowledge Base (Centralized)

Layer 2: Real-time Connections (MCP Servers)

Layer 3: MCP Gateway

Information Flow Integration

Complete Context for Feature Development

Cross-System Traceability

Implementation Roadmap

Phase 0: Requirements Discovery (Q2 2026/Apr-Jun)

Phase 1: Core Knowledge Base Establishment (Q3 2026/Jul-Sep)

Phase 2: Expansion & Integration (Q3-Q4 2026/Jul-Dec)

Phase 3: Full Integration & Rollout (Q4 2026/Oct-Dec)

Relationship with Existing Proposals

Success Metrics

Risks and Mitigation

AI-First Context Infrastructure ​

Problem Statement ​

Current State: Context Silos ​

Evidence from Team Feedback ​

Fundamental Challenge: Accessible ≠ Usable ​

Proposed Solution: Hybrid AI Context Architecture ​

Target Architecture ​

Core Principles ​

Architecture Layers ​

Layer 1: Core Knowledge Base (Centralized) ​

Layer 2: Real-time Connections (MCP Servers) ​

Layer 3: MCP Gateway ​

Information Flow Integration ​

Complete Context for Feature Development ​

Cross-System Traceability ​

Implementation Roadmap ​

Phase 0: Requirements Discovery (Q2 2026/Apr-Jun) ​

Phase 1: Core Knowledge Base Establishment (Q3 2026/Jul-Sep) ​

Phase 2: Expansion & Integration (Q3-Q4 2026/Jul-Dec) ​

Phase 3: Full Integration & Rollout (Q4 2026/Oct-Dec) ​

Relationship with Existing Proposals ​

Success Metrics ​

Risks and Mitigation ​

Related Principles ​

AI-First Context Infrastructure

Problem Statement

Current State: Context Silos

Evidence from Team Feedback

Fundamental Challenge: Accessible ≠ Usable

Proposed Solution: Hybrid AI Context Architecture

Target Architecture

Core Principles

Architecture Layers

Layer 1: Core Knowledge Base (Centralized)

Layer 2: Real-time Connections (MCP Servers)

Layer 3: MCP Gateway

Information Flow Integration

Complete Context for Feature Development

Cross-System Traceability

Implementation Roadmap

Phase 0: Requirements Discovery (Q2 2026/Apr-Jun)

Phase 1: Core Knowledge Base Establishment (Q3 2026/Jul-Sep)

Phase 2: Expansion & Integration (Q3-Q4 2026/Jul-Dec)

Phase 3: Full Integration & Rollout (Q4 2026/Oct-Dec)

Relationship with Existing Proposals

Success Metrics

Risks and Mitigation

Related Principles