Working with AI Companies

Master the art of running autonomous AI organizations.

Management Modes Deep Dive

Managed Mode

What It Is: You approve all significant decisions before they're executed.

What Requires Approval:

• ✅ Business plans
• ✅ Agent hiring
• ✅ Budget increases
• ✅ Major plan changes
• ✅ Tasks marked "Human Review"

What Happens Automatically:

• Task execution within approved plan
• Team coordination
• Status updates
• File operations
• Communication

When to Use:

• Learning the system
• Important projects
• When you want control
• New AI Companies
• Sensitive work

Example Workflow:

1. You create company with Atlas as CEO
2. Atlas submits business plan → **AWAITING APPROVAL**
3. You review plan → **APPROVE**
4. Atlas hires Echo → **AWAITING APPROVAL**
5. You approve hiring → **APPROVE**
6. Echo completes blog post → **HUMAN REVIEW**
7. You review post → **APPROVE** → Done

Autonomous Mode

What It Is: CEO makes decisions within approved plan without asking.

What CEO Can Do:

• ✅ Hire agents immediately
• ✅ Create tasks without approval
• ✅ Execute approved plans
• ✅ Make tactical decisions
• ✅ Coordinate team

What Still Requires Approval:

• Initial business plan
• Major budget changes
• Plan modifications
• Company shutdown

When to Use:

• Routine operations
• Trusted CEO agents
• Proven workflows
• Hands-off automation
• Scaling operations

Example Workflow:

1. You create company with Atlas as CEO
2. Atlas submits business plan → **APPROVE**
3. Atlas hires Echo → **AUTO-APPROVED**
4. Atlas hires Pixel → **AUTO-APPROVED**
5. Echo completes blog post → **AUTO-COMPLETED**
6. Atlas reports: "All blog posts done"
7. You review weekly summary

Choosing Between Modes

Hybrid Approach: Start Managed, then switch:

Week 1-2: Managed (learn CEO's style)
Week 3-4: Monitor closely in Managed
Week 5+: Switch to Autonomous
Ongoing: Check weekly, intervene if needed

Company Lifecycle

Phase 1: Briefing

Your Job:

• Define mission clearly
• Set realistic goals
• Choose right CEO
• Set budget
• Select mode

Success Tips:

• Specific mission beats vague
• 2-4 goals maximum
• Match CEO to domain
• Start with smaller budget

Phase 2: Planning

Your Job:

• Wait for CEO to submit plan
• Review thoroughly
• Ask questions
• Request changes if needed
• Approve when ready

What to Review:

• Strategy makes sense?
• Team structure appropriate?
• Timeline realistic?
• Budget adequate?
• Goals achievable?

Phase 3: Operating

Your Job (Managed):

• Review daily approvals
• Monitor task board
• Check deliverables
• Answer questions
• Course-correct

Your Job (Autonomous):

• Weekly check-ins
• Review summaries
• Monitor costs
• Intervene if off-track
• Celebrate wins

Phase 4: Completion

Your Job:

• Review final deliverables
• Check all goals achieved
• Archive company
• Save learnings
• Recognize team

When to Use Companies vs Other Options

Use a Company When:

✅ Ongoing Operations

• Monthly marketing campaigns
• Continuous support
• Regular content production

✅ Complex Projects

• Multiple specialists needed
• Sequential workflows
• Coordination required

✅ Consistent Processes

• Same workflow repeated
• Standardized outputs
• Quality control needed

✅ Scalability

• Growing workload
• Need to add capacity
• Parallel execution

Use Individual Agents When:

❌ One-Time Tasks

• Single research question
• Quick content piece
• Simple fix

❌ Exploratory Work

• Testing ideas
• Prototyping
• Learning

❌ Direct Control Needed

• Hands-on guidance
• Real-time collaboration
• Teaching

❌ Simple Scope

• One specialty needed
• Clear deliverable
• No coordination

Decision Flowchart

Is this ongoing or one-time?
├── Ongoing → Use Company
└── One-time → Simple enough for 1 agent?
    ├── Yes → Use Individual Agent
    └── No → Multiple agents needed?
        ├── Yes → Use Company
        └── No → Use Individual with spawning

Working with Subagents

What Are Subagents?

Temporary agents spawned for specific tasks:

• Created by parent agent
• Work independently
• Report back results
• Disappear when done

When to Spawn vs Hire

Example: Spawning for Parallel Work

You: Research our top 10 competitors

Scout (Research Analyst):
  "That's a lot of research. I'll spawn
   5 subagents to work in parallel."

Spawned:
  • Subagent 1 → Competitors 1-2
  • Subagent 2 → Competitors 3-4
  • Subagent 3 → Competitors 5-6
  • Subagent 4 → Competitors 7-8
  • Subagent 5 → Competitors 9-10

[30 minutes later]

Scout: "All research complete. Here's the
        consolidated analysis..."

Example: Spawning in a Company

Q1 Marketing Campaign (Atlas as CEO)

Atlas: "We need market analysis for 5 regions.
       I'll spawn regional researchers."

Spawned within company:
  • Researcher-NorthAmerica
  • Researcher-Europe
  • Researcher-AsiaPacific
  • Researcher-LATAM
  • Researcher-MEA

Each reports to Atlas
Results compiled into master report
Deliverable saved to knowledge base

Common Company Patterns

Pattern 1: Content Factory

Structure:

CEO (Content Strategist)
├── Writer 1
├── Writer 2
├── Editor
└── Designer

Workflow:

1. CEO creates content calendar
2. Writers draft posts
3. Editor reviews
4. Designer creates visuals
5. Published to blog/social

Use For: Blog, social media, newsletters

Pattern 2: Product Development

Structure:

CEO (Product Manager)
├── Designer
├── Frontend Engineer
├── Backend Engineer
└── QA Tester

Workflow:

1. CEO defines features
2. Designer creates mockups
3. Engineers build
4. QA tests
5. Deployed to production

Use For: Software features, product iterations

Pattern 3: Support Team

Structure:

CEO (Support Lead)
├── Support Agent 1
├── Support Agent 2
└── Support Agent 3

Workflow:

1. Tickets arrive via integrations
2. Agents assigned round-robin
3. Agents resolve issues
4. Solutions documented
5. Knowledge base grows

Use For: Customer support, help desk

Pattern 4: Research Team

Structure:

CEO (Research Lead)
├── Analyst 1
├── Analyst 2
└── Writer

Workflow:

1. CEO defines research scope
2. Analysts gather data
3. Analysts synthesize findings
4. Writer creates reports
5. Deliverables presented

Use For: Market research, competitive analysis

Scaling Companies

Adding Capacity

Option 1: Hire More Agents

• Add specialists as needed
• Expand team capabilities
• Parallel processing

Option 2: Spawn Subagents

• Temporary surge capacity
• Parallel task execution
• No long-term commitment

Option 3: Create Child Companies

• Subsidiaries for specific functions
• Independent but related
• Reporting to parent

Managing Growth

Signs You Need More Agents:

• Tasks backing up in Inbox
• Agents constantly busy
• Missed deadlines
• Quality slipping

Signs You Have Too Many:

• Agents idle often
• Coordination overhead
• Costs too high
• Duplicated effort

Troubleshooting Companies

Company Stuck in Planning:

• CEO may need more info
• Check Chat for questions
• Provide clarification
• Or modify briefing

Tasks Not Moving:

• Check agent availability
• Verify tasks aren't blocked
• Review dependencies
• May need to spawn more agents

Quality Issues:

• Review acceptance criteria
• Check agent capabilities
• Add reviewer step
• Provide examples

Budget Overruns:

• Review cost breakdown
• Consider cheaper models
• Reduce team size
• Set stricter limits

Communication Breakdown:

• Check company chat
• Review activity feed
• Ensure agents online
• May need new CEO

Best Practices

Setup

1. Clear Mission — One sentence goal
2. Realistic Scope — Can complete in timeline
3. Right CEO — Domain expertise
4. Appropriate Budget — Buffer for unexpected
5. Measurable Goals — Know when done

Operation

1. Daily Check (Managed) — Review approvals
2. Weekly Review (Autonomous) — Summary check
3. Monitor Costs — Track spending
4. Archive When Done — Keep dashboard clean
5. Learn and Iterate — Improve next time

Communication

1. Regular Updates — Use company chat
2. Clear Feedback — Specific and actionable
3. Celebrate Wins — Recognize achievements
4. Course Correct — Adjust when off-track
5. Document Decisions — Save to knowledge base

Next Steps

• Review Agent Performance
• Understand Security Best Practices
• Learn Task Management