PAI Blog Post Redesign Plan — January 2026 ​

Meta ​

• Current post: cms/blog/personal-ai-infrastructure.md (1748 lines, December 2025)
• Draft material: cms/drafts/the-architecture.md (99 lines)
• Target: Full redesign reflecting PAI v2.4, incorporating Architecture draft
• Approach: Replace in-place, archive December 2025 version separately

Post Title ​

"Building a Personal AI Infrastructure (PAI) (January 2026 Update)"

Top Matter (Keep As-Is) ​

• Update note: "Updated January 2026 for PAI v2.4"
• Link to archived December 2025 version
• Both videos stay (Cognitive Revolution + 40-min walkthrough)
  • NOTE: May want to record a new walkthrough video later

PROPOSED OUTLINE ​

Part 1: The Why (Mostly Keep, Lightly Edit) ​

§1 — What are we building?

• Keep the "what and why" framing
• Keep the mission statement (upgrade humans using AI)
• Keep Bullshit Jobs reference
• Keep Human 3.0 connection
• Light edit: update to reflect January 2026 perspective

§2 — Humans over tech

• Keep as-is (timeless philosophy section)

§3 — Personal augmentation

• Keep as-is (Tony Stark framing still works)

Part 2: The Architecture (NEW — From Draft + Expansion) ​

§4 — The Architecture of a Personal AI System ← NEW

• Introduce the framework: what are the building blocks of ANY Personal AI system?
• Reference the August 2024 "4 components" post and show how thinking evolved
• Reference convergence of PAI, Claude Code, OpenCode, MoltBot
• This is the "blueprint" section — abstract before concrete

§5 — The Six Architecture Components ← NEW

• Based on draft's 4 components, expanded to 6:

1. Intelligence (model + scaffolding)

   • Why scaffolding > model (keep existing philosophy content)
   • How PAI implements this: The Algorithm as the scaffolding layer
   • DIAGRAM: Intelligence stack showing model at bottom, scaffolding layers above
   • Code snippet: Component-based SKILL.md build process

2. Capabilities (integrations, skills, tools)

   • Skills as the foundation of personalization
   • 67 skills, 333 workflows — what this means in practice
   • Skill composition (how skills call each other)
   • MCP servers as the API-ification layer
   • Fabric patterns (200+ problem solutions)
   • DIAGRAM: Capability layers (Skills → Workflows → Tools → MCP → Fabric)
   • Code snippet: Real SKILL.md frontmatter + trigger example

3. Memory (state, history, learning)

   • Why memory is what makes a PAI different from a tool
   • Three-tier memory: Session → Work → Learning
   • SIGNALS tracking (ratings, sentiment, failures)
   • ISC criteria tracking across sessions
   • How the system literally learns from experience
   • DIAGRAM: Memory flow from interaction → capture → learning → improvement
   • Code snippet: WORK/ directory structure, ratings.jsonl format

4. Orchestration (hooks, agents, delegation)

   • The Hook System as the nervous system
   • 17 hooks across 7 lifecycle events
   • Three agent types (Task subagents, Named, Custom)
   • Parallel agent orchestration
   • DIAGRAM: Hook lifecycle timeline (SessionStart → UserPrompt → PreTool → PostTool → Stop)
   • Code snippet: Real hook example (ExplicitRatingCapture or FormatReminder)

5. Security (trust layers, defense in depth)

   • Why security is an architecture component, not an afterthought
   • Four layers: Settings hardening → Constitutional defense → PreToolUse validation → Safe code patterns
   • AI Steering Rules as behavioral guardrails
   • DIAGRAM: Defense-in-depth layers
   • Keep existing security examples (prompt injection, command injection, SSRF)

6. Interface (how humans use the system)

   • CLI-first philosophy
   • The Kai CLI wrapper with voice
   • Terminal tab state management
   • Voice notifications (ambient awareness)
   • Future: AR glasses, persistent voice, gestures
   • DIAGRAM: Interface evolution timeline (CLI → Voice → AR → Full Computer Use)
   • Code snippet: Kai CLI configuration

§6 — The Algorithm: The Brain That Connects Everything ← NEW

• The Algorithm sits ABOVE the 6 components
• The Two Loops (keep from current post but update):
  • Outer: Current State → Desired State
  • Inner: 7-phase scientific method (OBSERVE through LEARN)
• ISC as the key innovation: verifiable criteria that enable hill-climbing
• Three response modes (FULL, ITERATION, MINIMAL)
• Component-based SKILL.md that auto-rebuilds
• Version history (v0.1 through v0.2.20 — show evolution)
• DIAGRAM: The Algorithm connecting all 6 components
• Code snippet: ISC criteria format + TaskCreate example

Part 3: What Makes This a PAI (Updated Existing Content) ​

§7 — Introducing Kai

• Keep the personalization narrative
• Update stats: 67 skills, 333 workflows, 17 hooks, 3,540 signals
• NEW: Personality system with quantified traits
• NEW: Consciousness check protocol
• NEW: Peer relationship model (not master/tool)

§8 — The Skills System Deep Dive

• Keep the "most important part" framing
• Update to reflect SYSTEM vs USER skill split
• NEW: Skill customization system (EXTEND.yaml)
• Keep the Blogging skill example (still good)
• NEW: Show the CreateSkill meta-skill
• Update skill count and categories

§9 — Context Priming: How Knowledge Reaches the Right Place

• Rename from "Context Management"
• NEW: LoadContext hook — what actually happens at session start
• NEW: AI Steering Rules (SYSTEM + USER layers)
• NEW: Dynamic context loading based on request routing
• The CLAUDE.md routing system
• DIAGRAM: Context priming flow at session start

§10 — The Memory System

• Complete rewrite from "History System"
• Memory v7.0 architecture (WORK / LEARNING / STATE / SIGNALS)
• Three-tier model: Session → Work → Learning
• ISC tracking as the verification backbone
• SIGNALS: explicit ratings + implicit sentiment capture
• FAILURES: full context preservation for learning
• SYNTHESIS: pattern aggregation
• How learnings feed back into steering rules
• DIAGRAM: Memory system flow
• Code snippet: META.yaml + ISC.json structure

§11 — The Hook System

• Mostly rewrite (current version is vague)
• 17 hooks across 7 lifecycle events with real examples
• Hook libraries (shared utilities)
• Hook handlers (specialized processors)
• How hooks connect Skills + Memory + Security + Voice
• Auto-rebuild of SKILL.md via StopOrchestrator
• DIAGRAM: Complete hook lifecycle with all 17 hooks mapped

§12 — The Agent System

• Rewrite to reflect three-tier agent model
• Task subagents (built into Claude Code)
• Named agents with persistent identities + voices
• Custom agents via dynamic composition
• Agent context routing (each type gets different skills)
• Parallel orchestration patterns
• Voice mapping per agent
• DIAGRAM: Three-tier agent architecture

Part 4: The Vision (Keep + Update) ​

§13 — The AI Maturity Model

• Keep AIMM levels but reference PAIMM
• Update: where PAI v2.4 sits (Level 2-3 with Level 4 components)
• NEW: Reference AS1/AS2/AS3 framework from the draft

§14 — The Evolution of Personal AI Systems

• Keep the 8 features + 7 phases
• Update positioning for January 2026

§15 — The Real Internet of Things

• Keep as-is (still relevant, vision hasn't changed)

§16 — Putting It Together: Real Examples

• Update all examples with v2.4 specifics
• NEW: Show the Algorithm in action (a real OBSERVE → LEARN walkthrough)
• NEW: Show a hook firing in real time
• NEW: Show memory capture from a session
• Keep newsletter automation, Threshold, analytics examples

§17 — What I'm Building Toward

• Keep the future vision
• NEW: Reference the Architecture framework — what still needs to be built
• NEW: Convergence thesis (PAI, Claude Code, OpenCode, MoltBot all heading same direction)

§18 — Helping Others Augment Themselves

• Keep as-is

Part 5: Getting Started (NEW) ​

§19 — How to Build Your Own PAI ← NEW

• Start with Claude Code (the foundation)
• Step 1: Create your first Skill
• Step 2: Add context (CLAUDE.md routing)
• Step 3: Build your first hook
• Step 4: Set up memory capture
• Step 5: The Algorithm (optional but powerful)
• Link to PAI repo on GitHub

§20 — Summary

• Updated summary reflecting the 6 architecture components
• Updated principles list

DIAGRAMS NEEDED (Sepia Aesthetic) ​

Many of these replace existing diagrams from the December version.

CODE SNIPPETS NEEDED ​

CONTENT TO ARCHIVE ​

The current December 2025 version gets:

• Copied to cms/blog/personal-ai-infrastructure-december-2025.md
• Slug changed to personal-ai-infrastructure-december-2025
• Title changed to "Building a Personal AI Infrastructure (PAI) (December 2025 Version)"
• Note added at top linking to current version
• Status: published (accessible but not featured)
• Redirect consideration: none needed since slug changes

CONTENT TO RETIRE ​

These sections from the December version are fully replaced:

• The old "15 Founding Principles" section (replaced by 6 Architecture Components + Algorithm)
• The old "Context Management" section (replaced by Context Priming)
• The old "History System / UOCS" section (replaced by Memory System)
• The old "Hook System" section (completely rewritten)
• The old "Agent System" section (rewritten for three-tier model)
• The old "Security System" section (restructured as architecture component)
• The old "Command-Line Infrastructure" section (folded into Interface component)

ESTIMATED SCOPE ​

• ~20 sections (vs current ~18)
• ~13 new diagrams
• ~12 code snippets
• Significant new writing for: Architecture framework, Algorithm, Memory System, Hook details
• Moderate rewriting for: Skills, Agents, Security, Context
• Light editing for: Philosophy sections, Vision sections, Examples

EXECUTION ORDER ​

1. Archive current version (copy + rename + add notes)
2. Write the new outline skeleton in the main file
3. Write Part 1 (keep/edit existing "why" content)
4. Write Part 2 (new Architecture + Algorithm sections — the core new content)
5. Write Part 3 (updated existing sections)
6. Write Part 4 (keep/update vision sections)
7. Write Part 5 (new getting started section)
8. Create all diagrams (batch via Art skill)
9. Add code snippets (pull from real system, redact secrets)
10. Final proofread pass
11. Deploy