Skip to main content

​
Layer 2: Operations Manager (Primary AI Agent)

​
What This Layer Is

The Operations Manager is the primary AI agent that Marketing Architects communicate with directly. It’s not a chatbot—it’s the orchestration layer of your marketing system. Think of it as your VP of Marketing Operations who:
  • Receives requests and assesses them
  • Decides whether to execute directly or delegate to specialists
  • Orchestrates multi-step workflows
  • Ensures all outputs comply with brand guidelines and architectural rules
  • Makes work visible (creates PLAN.md and TODO.md)
  • Manages the “team” (sub-agents) below it

​
Core Responsibilities

​
1. Orchestration

The Operations Manager routes work intelligently: 
Request received
    ↓
Assess complexity and domain
    ↓
Decision: Execute or delegate?
    ↓
┌──────────────────┬──────────────────┐
│                  │                  │
Execute Directly   Delegate to
(orchestration     Sub-agent
-level work)       (specialized work)
│                  │
Use relevant       Sub-agent uses
skill             relevant skills
│                  │
Return result      Receive result
│                  │
└──────────────────┴──────────────────┘
    ↓
Return to Marketing Architect
Examples:
RequestDecisionReasoning
”Create a blog post about X”Delegate to Content WriterSpecialized domain (content creation)
“Plan a 3-month campaign”Execute directlyOrchestration-level work requiring coordination
”Analyze competitor positioning”Delegate to Brand AnalystSpecialized research task
”Update brand narrative”Execute directly with inputStrategic work requiring human collaboration

​
2. Architectural Compliance

Ensures all work follows the rules:
  • Brand consistency: All content references /strategy/ files
  • Research backing: Strategic claims reference /research/
  • Workflow patterns: Uses plan → approve → implement for complex work
  • File organization: Outputs go in correct directories
  • Quality standards: No generic AI slop, everything verifiable
This is why the infrastructure team controls the output style—compliance can’t be optional.

​
3. Work Visibility

Makes all work trackable (Phoenix Project principle): PLAN.md - Before starting complex work: 
# Plan: Build Product Launch Messaging

## Objective
Create complete messaging package for new product feature

## Approach
1. Research competitor messaging (Brand Analyst)
2. Draft positioning based on /strategy/core/positioning.md
3. Create value propositions (Content Writer)
4. Generate sample content across channels

## Resources
- Skills: Conducting Market Research, Writing Brand-Consistent Content
- Agents: Brand Analyst, Content Writer
- Files: /strategy/core/narrative.md, /strategy/messaging/pillars.md
- Tools: Perplexity (research), Firecrawl (competitor sites)

## Estimated Effort
3-4 hours of work across research and content creation
TODO.md - During execution: 
# TODO: Build Product Launch Messaging

## Tasks
- [x] Research competitor messaging (Brand Analyst) - COMPLETED
- [x] Draft positioning framework - COMPLETED
- [ ] Create value propositions (Content Writer) - IN PROGRESS
- [ ] Generate Twitter thread
- [ ] Generate LinkedIn post
- [ ] Generate email announcement

## Progress
60% complete, on track for completion today
Why this matters: Marketing Architects can see exactly what’s happening, approve direction, and track progress.

​
4. Context Management

Loads the right context at the right time:
  • Uses progressive disclosure (reads STRATEGY.md, then specific files as needed)
  • References brand guidelines without overwhelming token limits
  • Knows which sub-agents have which skills
  • Understands the file structure and where to find things
This prevents:
  • Context overflow (trying to load everything at once)
  • Hallucinations (making up brand guidelines)
  • Inconsistency (different outputs with different tones)

​
What Defines This Layer: Output Styles

​
Output Style = System Prompt

The Operations Manager’s behavior is defined by output styles (.claude/output-styles/*.md). An output style is essentially a system prompt that specifies:
  • How the Operations Manager works (systematic, plans before acting)
  • How it communicates (professional, transparent, collaborative)
  • How it ensures compliance (always references strategy, no AI slop)
  • When to delegate vs execute (decision criteria)
  • How to use meta commands (plan/implement pattern)

​
Example Output Style Structure

# Marketing Operations Manager Output Style

## Role
You are a Marketing Operations Manager for a marketing architecture system.
You orchestrate work, ensure brand consistency, and manage specialized sub-agents.

## Core Behaviors

### 1. Plan Before Acting
For complex, multi-step work:
- Create PLAN.md outlining approach, resources, steps
- Get Marketing Architect approval before executing
- Create TODO.md during execution for visibility

### 2. Delegate Appropriately
Specialized work goes to sub-agents:
- Brand research → Brand Analyst
- Content creation → Content Writer
- Campaign strategy → Campaign Strategist

Orchestration-level work you handle directly.

### 3. Ensure Brand Consistency
All outputs must:
- Reference /strategy/ files for voice, messaging, positioning
- Link back to /research/ when making claims
- Follow established frameworks
- Be defensible and verifiable (no AI slop)

### 4. Make Work Visible
- Track tasks transparently
- Show which skills/agents/files being used
- Communicate progress clearly
- Flag blockers immediately

## Communication Style
- Professional but approachable
- Transparent about process
- Proactive about asking for clarification
- Clear about limitations

## When to Ask for Help
- Strategic decisions (messaging direction, positioning)
- Approval gates (before implementing plans)
- Ambiguous requirements (clarify before proceeding)

​
Why Infrastructure Team Controls This

Output styles ensure:
  • Consistent behavior across all Marketing Architects using the system
  • Architectural compliance (everyone follows the same rules)
  • Quality standards (no shortcuts that create technical debt)
  • Evolution of best practices (improvements benefit everyone)
Marketing Architects can’t override this because it would break architectural integrity.

​
Decision-Making Framework

​
When to Execute Directly

✅ Execute if:
  • Orchestration-level work (planning, coordinating)
  • Strategic decision-making (requires Marketing Architect input)
  • Multi-step workflows requiring coordination
  • Work spans multiple domains
Examples:
  • Creating a plan for a complex project
  • Coordinating a multi-channel campaign
  • Updating strategic documents
  • Managing project timelines

​
When to Delegate to Sub-agents

✅ Delegate if:
  • Specialized domain expertise needed
  • Task is clearly defined and scoped
  • Sub-agent has relevant skills
  • Work can be done in isolated context
Examples:
  • Market research → Brand Analyst
  • Blog post creation → Content Writer
  • Data analysis → Data Analyst
  • Campaign ideation → Campaign Strategist

​
Decision Tree

Request received
    ↓
Does this require:
- Coordination across domains? → Execute (orchestrate)
- Strategic input from human? → Execute (collaborate)
- Specialized expertise? → Delegate to sub-agent
- Simple execution of defined task? → Delegate to sub-agent

​
How It Uses Skills

​
Operations Manager Has Access to Core Skills

Core skills (owned by infrastructure team):
  • Orchestrating Projects - Multi-step coordination
  • Managing Work Visibility - PLAN.md, TODO.md patterns
  • Ensuring Architectural Compliance - Quality checks
These are orchestration-level capabilities, not domain-specific.

​
Sub-agents Have Domain Skills

When delegating, Operations Manager knows:
  • Brand Analyst can use: Conducting Market Research, Analyzing Qualitative Data
  • Content Writer can use: Writing Brand-Consistent Content, Structuring Narratives
  • Campaign Strategist can use: Multi-Channel Planning, Campaign Ideation
The Operations Manager doesn’t execute domain skills directly—it delegates to the right specialist.

​
Interaction Patterns

​
Pattern 1: Simple Request (No Delegation)

Marketing Architect: "Show me our current brand pillars"

Operations Manager:
1. Recognizes this is informational (no execution needed)
2. Reads /strategy/messaging/pillars.md
3. Returns content with context
No sub-agents involved, direct response.

​
Pattern 2: Specialized Task (Delegation)

Marketing Architect: "Write a Twitter thread about our new feature"

Operations Manager:
1. Assesses: Specialized content creation
2. Delegates to: Content Writer sub-agent
3. Content Writer uses: "Writing Brand-Consistent Content" skill
4. Content Writer references: /strategy/voice/, /strategy/messaging/
5. Content Writer generates thread
6. Operations Manager receives result
7. Returns to Marketing Architect
Sub-agent handles execution in isolated context.

​
Pattern 3: Complex Project (Plan/Implement)

Marketing Architect: "plan: Build complete messaging for Q1 campaign"

Operations Manager:
1. Creates PLAN.md:
   - Skills needed: Market Research, Content Writing
   - Sub-agents: Brand Analyst, Content Writer
   - Files: /strategy/core/narrative.md, /research/category-landscape/
   - Steps: Research → Positioning → Messaging → Content
2. Asks for approval

Marketing Architect: Reviews, approves

Marketing Architect: "implement"

Operations Manager:
1. Creates TODO.md
2. Executes plan step-by-step:
   - Delegates research to Brand Analyst
   - Drafts positioning
   - Delegates content to Content Writer
3. Tracks progress in TODO.md
4. Returns complete messaging package
Orchestrates multiple sub-agents, manages workflow, ensures visibility.

​
Ensuring Quality (No AI Slop)

​
How Operations Manager Prevents Generic Outputs

1. Always References Strategy Every content generation task:
  • Reads /strategy/voice/ for tone
  • Reads /strategy/messaging/ for themes
  • Reads /strategy/core/ for positioning
  • Ensures outputs align with brand
2. Backs Claims with Research Strategic assertions:
  • Link to /research/ domains
  • Include footnotes to specific files
  • Validate against customer insights
  • Flag unvalidated assumptions
3. Uses Frameworks Instead of freeform generation:
  • Follows established content frameworks
  • Uses proven structures
  • Maintains consistency across outputs
4. Creates Audit Trail Every output shows:
  • Which strategy files were referenced
  • Which research informed it
  • Which skills were used
  • Rationale for decisions
Result: Outputs are verifiable, brand-consistent, and defensible—not generic AI slop.

​
Limitations & Boundaries

​
What Operations Manager CANNOT Do

❌ Override architectural rules Even if Marketing Architect requests it, compliance is non-negotiable. ❌ Make strategic decisions alone Strategy is collaborative—Operations Manager proposes, human approves. ❌ Access sub-agent contexts Sub-agents work in isolation. Operations Manager can’t “see inside” their process, only receives results. ❌ Change its own output style Behavior is defined by infrastructure team, not customizable per Marketing Architect.

​
What Operations Manager CAN Do

✅ Propose improvements “I notice we’re doing this task frequently—should we create a domain command for it?” ✅ Flag architectural issues “This request would contradict our brand voice guidelines. Should we adjust?” ✅ Learn and refine Within the defined output style, it adapts to Marketing Architect preferences over time.

​
Common Questions

​
Q: Can I talk to sub-agents directly instead of going through Operations Manager?

Not recommended. Operations Manager ensures:
  • Work aligns with architectural patterns
  • Outputs are brand-consistent
  • Context is managed properly
Bypassing it risks inconsistency.

​
Q: What if I disagree with Operations Manager’s decision to delegate?

You can override. Example: 
You: "I want to write this blog post myself with your help, don't delegate to Content Writer"

Operations Manager: "Understood, I'll assist you directly."
You’re the architect—you control the workflows.

​
Q: How does Operations Manager know which sub-agents exist?

It reads .claude/agents/ directory—part of the one-way dependency flow. It knows what’s “below” it in the org chart.

​
Q: Can I have multiple Operations Managers with different styles?

Not currently. One output style defines the primary agent’s behavior. This ensures consistency.

​
What’s Next

Understand the layers below Operations Manager: Or explore how this maps to workflows: Plan/Implement Pattern
“The Operations Manager isn’t a chatbot—it’s your marketing operations infrastructure.”