Capstone: Build Your Zero-Human Company

The capstone: design your company's org chart, hire 4+ agents across at least 2 adapter types, seed a real backlog of 20+ tasks, run for 7 days, and submit a cost + output analysis report.

Capstone: Build Your Own AI Company

What you'll build

Design, implement, and deploy a complete Paperclip AI company of your own — from requirements to production.

The Capstone Specification

Your company must meet these requirements:

✅ At least 4 agents with distinct roles
✅ At least 2 different adapter types (e.g. claude-code + shell)
✅ An org chart with at least 2 levels (manager + workers)
✅ A heartbeat schedule appropriate to your use case
✅ At least 1 approval gate for a critical action
✅ Budget limits set at both agent and company level
✅ A seeded batch of at least 10 realistic issues
✅ Monitoring: at least 1 alert configured
✅ A production-ready deployment (Docker Compose)

Suggested Company Ideas

Idea	Difficulty	Interesting because
Personal newsletter generator	⭐⭐	Content pipeline end-to-end
Open-source repo summariser	⭐⭐⭐	GitHub API integration
Daily digest bot	⭐⭐	Scheduling, aggregation
Code documentation writer	⭐⭐⭐	Code analysis, structured output
Social media monitor	⭐⭐⭐	Webhook triggers, classification
Customer FAQ responder	⭐⭐	Retrieval, knowledge base
Data pipeline auditor	⭐⭐⭐	Shell adapter, deterministic

Capstone Deliverables

1. Company Design Document

Write a short design doc covering:
- Company name and purpose
- Agent roster with roles, adapter types, and budget
- Workflow diagram (text ASCII is fine)
- Approval gates and what they protect
- Cost estimate per week

2. SQL Seed File

A single idempotent .sql file that creates:
- The company
- All agents
- At least 10 sample issues

3. Production Deployment

A working docker-compose.yml + .env.example that a teammate could use to run your company.

4. Monitoring Setup

A SQL file with:
- The three key alerts (stuck issues, failed runs, budget warning)
- A weekly report query

Capstone Review Checklist

-- Run this to check your capstone meets requirements
SELECT
  (SELECT COUNT(*) FROM agents        WHERE company_id = $1)       >= 4 AS has_4_agents,
  (SELECT COUNT(DISTINCT adapter_type) FROM agents WHERE company_id = $1) >= 2 AS has_2_adapters,
  (SELECT COUNT(*) FROM agents        WHERE company_id = $1 AND reports_to IS NOT NULL) >= 1 AS has_hierarchy,
  (SELECT COUNT(*) FROM issues        WHERE company_id = $1)       >= 10 AS has_10_issues,
  (SELECT COUNT(*) FROM approvals     WHERE company_id = $1 OR TRUE)  >= 0 AS approval_gates_configured,
  (SELECT paused FROM companies       WHERE id = $1)                     AS is_not_paused;

What You've Learned in This Course

Working through all nine modules of the Paperclip AI course, you have:

Built a complete mental model of how autonomous AI agent systems work
Set up Paperclip locally with PGlite, then migrated to PostgreSQL
Hired and configured agents with different adapters (Claude, OpenAI, HTTP, shell)
Designed multi-level org charts with reports_to hierarchy
Implemented tasks, issues, and complex workflows (pipelines, fan-out, queues)
Mastered the heartbeat execution engine and debugged failing runs
Applied cost governance with budget limits and approval gates
Built multi-agent systems with coordinator patterns and shared memory
Deployed Paperclip to production with Docker, Nginx, TLS, CI/CD, and monitoring

You're now equipped to build real autonomous AI organisations. The source code, the schema, and the concept are all yours to extend.

Where to Go Next

Paperclip GitHub — contribute adapters, report issues, propose features
Community Discord — share your capstone, get feedback
Advanced topics — vector memory with pgvector, fine-tuned adapters, multi-model routing

Final Checkpoint ✓

[ ] Design document written with all required sections
[ ] SQL seed file complete and idempotent
[ ] docker-compose.yml + .env.example ready
[ ] Three monitoring alerts defined
[ ] Capstone review checklist passes all 6 checks
[ ] First production heartbeat run completes successfully