🤖 🚀 ✨

The Evolution of
Subagent Expertise

How AI Agent Specialization
Transforms Complex Development

A Case Study with Optimizely CMS

The Strategy: Divide & Conquer

1

Identify Major Domains

Break the monolith into logical areas of expertise

2

Specialize Further

Within each domain, identify specific responsibilities

3

Create Expert Agents

Build specialized agents with deep domain knowledge

The Agent Creation System (ACS)

🤖

Meta-Agents

6 Agents
The Governance Layer
📁 .acs/agents/
📊
Solution Analysis
Understands project structure
🏗️
Agent Creation
Spawns specialized domain agents
🔧
Agent Maintenance
Keeps agents up to date
⬆️
Upgrade
Manages system evolution
📚
Docs
Documents everything
🎯
Feedback
Learns and improves
The Governance Layer
These Meta-Agents analyze your project structure and create specialized Domain Agents for each area of expertise.

Agent Creation Flow

From Request to Activation: How AI Agents Are Born

1
🗣️
Request
Input: Problem statement
"Create presentation agents"
~30 seconds
→ Problem defined
2
🔍
Analysis
Identifies: 3 domains
Design, Content, Technical
Analyzes: Overlaps & risks
~2 minutes
→ Specialization map
3
📐
Architecture
Defines: Responsibilities
Establishes: Guardrails ⛔
Maps: Handoff patterns
~3 minutes
→ Agent specs
4
📝
Documentation
Generates: 350-550 lines
YAML, patterns, examples
Includes: Anti-patterns
~5 minutes
→ 3 agent .md files
5
🔗
Integration
Updates: GOVERNANCE.md
3 routing rules added
Links: To ecosystem
~1 minute
→ Updated governance
6
✅
Activation
Status: Live & callable
Ready: Accept queries
Monitored: Performance
Instant
✨ Active agents
🎯 Real Example: Presentation Agents
User Input
"Create agents for this presentation using ACS"
Analysis Result
3 specializations identified
Documentation
Designer (350) + Content (450) + Technical (550 lines)
Governance
3 new routing rules
Total Time
~12 minutes end-to-end
Outcome
3 production-ready agents ✨

The Opportunity

Optimizely's product suite is incredibly powerful and complex.
Let's enhance how we work with these systems using specialized AI expertise.

🔌
Commerce Connect
🛍️
Configured Commerce
🧪
Web Experimentation
✨
Feature Experimentation
Today's Example
📝
CMS
🎨
CMP + DAM
💎
Opal
🔄
ODP/OCP
📈
Analytics

Your Presentation Creators

This presentation was designed, written, and built by specialized AI agents

🎨 📝 💻

3 Presentation Agents

Specialized domain experts for presentation creation
🎨
Designer
Visual hierarchy, brand consistency, layout optimization
"No scrollbars allowed"
📝
Content
Storytelling, messaging strategy, narrative structure
"Opportunity over problem"
💻
Technical
HTML/CSS/JS implementation, navigation, performance
"Sections array must sync"
🤝 Coordinated Creation
These agents collaborated using the same ACS governance model—each with clear responsibilities, deferred to each other's expertise.
✨ Meta-Moment
All 25 agents—these 3 presentation specialists plus 22 CMS domain experts—will be featured in the gallery ahead.

Step 1: Major Layers

Step 2: Subspecialties

25 Specialized Agents

Meet All 25 Agents

Smart Agent Selection

Smart Agent Selection

Smart Agent Selection

Why Subagent Expertise Works

🎯

Specialization

Each agent is an expert in their domain with deep, specific knowledge

🧠

Context Management

Focused knowledge means better token efficiency and relevant responses

🚧

Clear Boundaries

Agents know their limits and defer to experts when needed

🔀

Smart Routing

Governance system routes tasks to the right expert automatically

📋

Accountability

Every decision is attributed, enabling targeted feedback

🔄

Scalability

New domains? Just add a new specialized agent

The Result

From one overwhelming monolith to 22 cooperating experts, each with crystal-clear responsibilities, boundaries, and deep domain knowledge.

Better answers. Faster solutions. Scalable expertise.

Real-World Challenge

CMS 12 → CMS 13 Upgrade

39
Issues Identified
12
Packages Disabled
✅
Functional Build

Key Challenges

🏗️ Architectural Changes
Sites → Applications, .NET 10 migration
⚙️ API Surface Changes
Removed interfaces, renamed services
📦 Third-Party Lag
12 packages lacking CMS 13 versions
📋 Documentation Gaps
Preview version mismatches, runtime issues

AI-Assisted Migration

🚀 3 UPGRADE1213 Agents

🎯
Coordinator
Orchestrates phases, agents, verification, and progress tracking
📚
Docs Specialist
Living plan management, issue tracking, knowledge preservation
💻
Code Specialist
Code transformations, API migrations, build verification

4-Step Process

1
Pre-Planning
Analyze breaking changes
2
Migration
Apply upgrades & fixes
3
Resolution
Troubleshoot 39 issues
4
Documentation
Generate 65 docs

Documentation vs. Reality

AI Agents Discovered What Docs Couldn't Predict

38%
Documented
In breaking changes
13%
Partial Hints
Needed investigation
49%
AI Discoveries
Found by agents

Where AI Agents Added Value

📦 Third-Party Discovery
Identified 12 packages needing alternatives beyond Optimizely's official docs
⚡ Runtime Investigation
Discovered database conflicts and service registration issues through testing
🔬 Version Tracking
Caught preview version mismatches between docs and actual package releases
🔗 Cross-Package Analysis
Mapped authorization, constructor, and attribute interactions across packages

The Impact

❌ Traditional Approach

📅 Timeline: 3-4 weeks of manual troubleshooting
🔍 Discovery: Sequential issue hunting, reactive debugging
📝 Documentation: Scattered notes, tribal knowledge
🧠 Knowledge Loss: Context vanishes after project ends
⚠️ Blind Spots: 49% of issues discovered too late

✅ AI-Assisted Approach

⚡ Timeline: Days instead of weeks
🎯 Discovery: Systematic issue identification & gap analysis
📚 Documentation: 65 structured docs auto-generated
🔄 Persistent Knowledge: Captured for future upgrades
📊 Analytics: Quantified doc accuracy (38%/13%/49%)

📊 Final Metrics

29/39
Issues Resolved
65
Docs Generated
100%
Future-Ready