Brainforge Knowledge

AGENT_SCALING_GUIDE

9 min read

Brainforge Vault: Agent Scaling Guide

Vision: 20-30 specialized AI agents running your business operations Current State: Foundation in place with personal “Second Brain” templates Path Forward: Systematic knowledge capture → Agent specialization → Orchestration

🎯 Current State Assessment

What You Have Now

1. Personal AI Second Brains (sales/content/cc-content-system/)

  • robert-gpt/ - Your personal knowledge base
  • Luke/ - Team member’s knowledge base
  • Structure: Memory (semantic) + Experiences (episodic) + Brain Health (metrics)
  • Workflow: /plan → /work → /review → /learn cycle

2. Organizational Knowledge (Top-level directories)

  • company/ - Company-wide context (planning, initiatives)
  • engineering/ - Technical knowledge (meetings, transcripts)
  • gtm/ - Go-to-market (sales, marketing, partnerships)
  • ops/ - Operations (SOPs, processes)
  • people/ - People ops (hiring, onboarding, roles)
  • research/ - Research and vendor assessments

3. Project-Specific Knowledge (lmnt/)

  • Client project documentation
  • Meeting notes and transcripts
  • Discovery artifacts

4. Agent Foundation

  • Agent definitions in .claude/agents/ (chief-of-staff, email-agent, etc.)
  • Skills system in .codex/skills/
  • Orchestration pattern (chief-of-staff routes to specialists)

🚀 How to Use the Vault Today

Phase 1: Personal Knowledge Capture (Week 1-2)

For You (robert-gpt/):

  1. Fill Core Context (2 hours)

    memory/personal/
├── services.md          ← What Brainforge offers
├── positioning.md       ← Market positioning
├── expertise.md         ← Your expertise areas
└── differentiators.md   ← What makes you unique

  2. Document Company Context (3 hours)

    memory/company/
├── products.md          ← Brainforge services
├── positioning.md       ← Company positioning
└── differentiators.md   ← Competitive advantages

  3. Capture Customer Intelligence (2 hours)

    memory/customers/
├── ideal-customer-profile.md
├── pain-points.md
└── objections.md

  4. Add Examples (ongoing)

    memory/examples/
├── proposals/           ← Winning SOWs (like hedra-sow.md)
├── emails/             ← Effective email templates
└── linkedin/           ← Thought leadership content

Action Items:

  • Migrate content from knowledge/sales/ (SOWs) → memory/examples/proposals/
  • Extract patterns from lmnt/ project → memory/patterns/
  • Document company values → memory/values-beliefs/

Phase 2: Organizational Knowledge Structure (Week 3-4)

Create Shared Memory Structure:

knowledge/
├── memory/                    # NEW: Shared organizational memory
│   ├── company/
│   │   ├── services.md        # All Brainforge services
│   │   ├── positioning.md     # Company positioning
│   │   ├── team.md            # Team structure, roles
│   │   └── processes.md       # Core business processes
│   ├── customers/
│   │   ├── icp.md             # Ideal customer profiles
│   │   ├── pain-points.md     # Common pain points
│   │   └── case-studies.md    # Client success stories
│   ├── operations/
│   │   ├── sales-process.md   # Sales methodology
│   │   ├── project-delivery.md
│   │   └── quality-standards.md
│   ├── patterns/
│   │   ├── sales-patterns.md  # What works in sales
│   │   ├── delivery-patterns.md
│   │   └── communication-patterns.md
│   └── examples/
│       ├── proposals/         # Winning proposals
│       ├── contracts/         # Contract templates
│       └── deliverables/      # Example deliverables

Action Items:

  • Create memory/ directory at vault root
  • Migrate company/ content → memory/company/
  • Extract patterns from lmnt/ → memory/patterns/
  • Document processes from ops/ → memory/operations/

Phase 3: Agent Specialization (Week 5-8)

Start with 5 Core Agents:

  1. Sales Agent (agents/sales-agent.md)

    • Knowledge: memory/customers/, memory/patterns/sales-patterns.md
    • Capabilities: Draft proposals, handle objections, qualify leads
    • Tools: Read, Write, CRM integration

  2. Project Manager Agent (agents/pm-agent.md)

    • Knowledge: memory/operations/project-delivery.md, lmnt/ projects
    • Capabilities: Create project plans, track deliverables, manage timelines
    • Tools: Read, Write, Task management integration

  3. Content Agent (agents/content-agent.md)

    • Knowledge: memory/examples/, memory/style-voice/
    • Capabilities: Write proposals, emails, thought leadership
    • Tools: Read, Write

  4. Research Agent (agents/research-agent.md)

    • Knowledge: research/, memory/company/
    • Capabilities: Vendor research, competitive analysis, market research
    • Tools: Read, Web search, Write

  5. Operations Agent (agents/ops-agent.md)

    • Knowledge: ops/, memory/operations/
    • Capabilities: Create SOPs, document processes, quality checks
    • Tools: Read, Write

Action Items:

  • Create agent definitions in .claude/agents/ at vault root
  • Each agent references specific memory/ directories
  • Test each agent with real tasks
  • Document agent capabilities in AGENTS.md

🎯 Path to 20-30 Agents

Agent Taxonomy (How to Think About Agents)

By Function:

  • Sales: Sales agent, Proposal agent, CRM agent
  • Delivery: PM agent, Technical lead agent, QA agent
  • Marketing: Content agent, SEO agent, Social media agent
  • Operations: Ops agent, Finance agent, Legal agent
  • People: Hiring agent, Onboarding agent, Culture agent
  • Research: Research agent, Competitive intel agent, Vendor agent

By Knowledge Domain:

  • Client-facing: Sales, proposals, account management
  • Internal: Operations, finance, people ops
  • Technical: Engineering, data, infrastructure
  • Strategic: Research, planning, analysis

Scaling Strategy

Stage 1: Foundation (5 agents) - Month 1-2

Goal: Prove the concept with core business functions

  • Sales Agent
  • PM Agent
  • Content Agent
  • Research Agent
  • Ops Agent

Success Criteria:

  • Each agent can handle 80% of tasks in their domain
  • Knowledge base is searchable and accurate
  • Agents can reference each other’s work

Stage 2: Expansion (10 agents) - Month 3-4

Goal: Cover all major business functions

Add:

  • Proposal Agent (specialized from Sales)
  • Technical Lead Agent
  • Hiring Agent
  • Finance Agent
  • Customer Success Agent

Success Criteria:

  • Agents handle 90% of routine tasks
  • Clear agent boundaries and handoffs
  • Quality metrics improve

Stage 3: Specialization (20 agents) - Month 5-6

Goal: Deep specialization within functions

Add:

  • Email Agent (already exists)
  • CRM Agent (already exists)
  • Data Agent (already exists)
  • SEO Agent
  • Social Media Agent
  • Legal Agent
  • Vendor Management Agent
  • Competitive Intel Agent
  • Account Management Agent
  • Quality Assurance Agent

Success Criteria:

  • Agents can handle complex, multi-step workflows
  • Agent-to-agent communication works smoothly
  • Business runs 50%+ on agent automation

Stage 4: Optimization (30 agents) - Month 7-12

Goal: Fine-tuned specialization and automation

Add:

  • Contract Review Agent
  • Invoice Agent
  • Meeting Prep Agent
  • Follow-up Agent
  • Documentation Agent
  • Training Agent
  • Analytics Agent
  • Forecasting Agent
  • Risk Assessment Agent
  • Innovation Agent

Success Criteria:

  • 80%+ of routine work automated
  • Agents proactively suggest improvements
  • Business runs smoothly with minimal human intervention

📚 Knowledge Capture Strategy

What to Capture (Priority Order)

1. High-Value, Frequently Used Knowledge

  • Customer profiles and pain points
  • Sales processes and objections
  • Service offerings and positioning
  • Winning proposals and contracts

2. Patterns That Compound

  • What works in sales conversations
  • Effective proposal structures
  • Successful project delivery patterns
  • Communication that lands

3. Examples of Excellence

  • Best-in-class proposals
  • Effective email templates
  • Thought leadership content
  • Project deliverables

4. Processes and Standards

  • Sales methodology
  • Project delivery process
  • Quality standards
  • Onboarding procedures

5. Context and History

  • Past projects (experiences/)
  • Client relationships
  • Market research
  • Competitive intelligence

How to Capture (Daily Practice)

After Every Client Interaction:

  1. Save email/meeting notes → experiences/[client-name]/
  2. Extract patterns → /learn command
  3. Update customer profile if new insights

After Every Project:

  1. Document in experiences/[project-name]/
  2. Extract deliverables → memory/examples/
  3. Extract patterns → memory/patterns/
  4. Update process docs if improved

Weekly Knowledge Maintenance:

  1. Review brain-health/growth-log.md
  2. Update low-confidence patterns
  3. Add new examples to memory/examples/
  4. Clean up outdated information

Monthly Strategic Review:

  1. Assess agent performance
  2. Identify knowledge gaps
  3. Plan new agent specializations
  4. Update company positioning if needed

🏗️ Infrastructure Needed

1. Knowledge Management System

Current: Markdown files in directories Needed:

  • ✅ Keep markdown structure (works great)
  • ✅ Add semantic search (already have MCP memory integration)
  • ⚠️ Consider vector database for large-scale search
  • ⚠️ Add knowledge graph for relationships

Action: Current structure is sufficient for 20-30 agents. Consider enhancement at 30+ agents.

2. Agent Orchestration

Current: Chief-of-staff pattern (manual routing) Needed:

  • ✅ Keep chief-of-staff pattern
  • ⚠️ Add agent-to-agent communication protocol
  • ⚠️ Add workflow automation (agent chains)
  • ⚠️ Add agent performance monitoring

Action: Build agent communication protocol. Document in AGENTS.md.

3. Quality Assurance

Current: /review command with 6-agent review Needed:

  • ✅ Keep review system
  • ⚠️ Add quality metrics per agent
  • ⚠️ Add automated quality checks
  • ⚠️ Add feedback loops (learn from mistakes)

Action: Extend review system to track agent-specific quality.

4. Access Control

Current: All agents can read all memory Needed:

  • ⚠️ Define agent access scopes
  • ⚠️ Personal vs. organizational memory separation
  • ⚠️ Client-specific knowledge isolation

Action: Create access control matrix for agents.

5. Monitoring and Analytics

Current: Brain health metrics (basic) Needed:

  • ⚠️ Agent usage metrics
  • ⚠️ Task completion rates
  • ⚠️ Quality trends per agent
  • ⚠️ Knowledge freshness tracking

Action: Build monitoring dashboard (can start simple with markdown logs).

🎓 Best Practices

Knowledge Capture

  1. Be Specific: “We won the LMNT deal by…” not “We sometimes win deals…”
  2. Include Context: Why something worked, not just what worked
  3. Update Regularly: Outdated knowledge is worse than no knowledge
  4. Cross-Reference: Link related knowledge (use markdown links)

Agent Design

  1. Single Responsibility: Each agent does one thing well
  2. Clear Boundaries: Know when to hand off to another agent
  3. Knowledge Scoping: Each agent has specific memory directories
  4. Tool Limitations: Agents only get tools they need

Orchestration

  1. Chief-of-Staff First: Always route through orchestrator for complex tasks
  2. Agent Chains: Document common workflows (e.g., Sales → Proposal → Legal)
  3. Error Handling: What happens when an agent fails?
  4. Human Escalation: When to involve humans

Quality

  1. Review Everything: Use /review before important outputs
  2. Learn from Mistakes: Update memory when agents make errors
  3. Pattern Confidence: Track which patterns are HIGH/MEDIUM/LOW confidence
  4. Continuous Improvement: Regular agent performance reviews

🎯 Success Metrics

Knowledge Quality

  • Coverage: 80%+ of business functions documented
  • Freshness: Knowledge updated within 30 days of change
  • Pattern Confidence: 50%+ patterns at HIGH confidence

Agent Performance

  • Task Completion: 80%+ of routine tasks handled by agents
  • Quality Score: Average 8+ on /review scores
  • Time Savings: 10+ hours/week saved per team member

Business Impact

  • Sales: Faster proposal generation, higher win rates
  • Delivery: Consistent quality, fewer errors
  • Operations: Reduced manual work, better documentation

🚨 Common Pitfalls to Avoid

  1. Over-Engineering Early: Start simple, add complexity as needed
  2. Under-Documenting: Capture knowledge as you go, don’t wait
  3. Agent Scope Creep: Keep agents focused, create new ones for new functions
  4. Ignoring Quality: Review agent outputs, update knowledge from mistakes
  5. Siloed Knowledge: Share organizational memory, not just personal
  6. No Feedback Loops: Learn from agent performance, update patterns

📖 Current Focus & Next Steps

Current Sprint (2 Weeks): GTM agents ONLY

  • See GTM_AGENT_SPRINT.md for 2-week sprint plan
  • Focus: BDR tactics, outreach, qualification agents

After GTM Sprint:

  • Expand to other domains (PM, content, ops) based on priority
  • Scale systematically using learnings from GTM agents
  • Build 20-30 agents over time, not all at once

🎓 Learnings from Real-World Implementation

See: VERCEL_LEAD_AGENT_LEARNINGS.md for detailed learnings from Vercel’s $2M lead agent.

Key Takeaways:

  1. Start Small, Prove Value Fast

    • Build MVP in 1-2 days (weekend projects work!)
    • Budget 3-5x more time for buy-in than building
    • Use shadow mode in production to prove value

  2. Two Types of Agents

    • Efficiency Agents: Replace tedious work (keep human in loop)
    • “Should” Agents: Do stuff that isn’t getting done (can fire by default)

  3. Find Opportunities Using Drew’s Framework

    Repetitive Task (100+ times/week)
+ High Business Impact
+ Tedious/Unfulfilling
+ Historical Data Available
= Perfect Agent Opportunity

  4. Shadow Your Best People

    • Watch what they actually do (revealed preference)
    • Not what they say they do (stated preference)
    • Document their tricks/optimizations
    • Replicate the best behavior in agents

  5. Prompt Structure Matters

    • Put reasoning FIRST, decision LAST
    • Forces “System 2” thinking (analytical) vs “System 1” (snap judgment)
    • Improves accuracy significantly

Read the full learnings document for implementation details and code examples.

Remember: The goal isn’t to have 30 agents immediately. It’s to systematically capture knowledge, build specialized agents, and compound your business intelligence over time. Every piece of knowledge you capture today makes tomorrow’s agents smarter.

Start Small, Think Big, Scale Systematically.

Graph View

Backlinks