Paperclip: Hire AI Agents Like Employees (Live Demo)

Key Concepts

• Paperclip: An open-source agent orchestration layer designed to manage "zero-human companies" by delegating tasks to AI agents.
• Agent Orchestration: The process of managing multiple AI agents with specific roles, goals, and budgets.
• Zero-Human Company: A business model where operations, coding, marketing, and management are handled entirely by AI agents.
• Heartbeat: A set of instructions or a "checklist" that guides an agent on what to do upon "waking up" (initialization).
• Persona/Skills: Configurable attributes and specialized capabilities (e.g., Remotion for video, coding agents) assigned to agents to define their behavior and output quality.
• Maximizer Mode: A forthcoming feature designed to push agents to work continuously until a specific goal is achieved, regardless of token spend.

1. Overview of Paperclip

Paperclip is an open-source project that serves as an orchestration layer for AI agents. Unlike automated business generators (like Pulsia) or individual coding tools (like Cursor), Paperclip focuses on goal-oriented management. It allows users to act as a "board member" or "CEO," defining high-level business objectives while the system manages the execution, hiring, and task delegation among a team of specialized AI agents.

2. Core Methodology and Framework

• The "CEO" Model: The user defines the company's mission. The CEO agent then breaks this mission into a roadmap, hires necessary sub-agents (e.g., engineers, QA, marketers), and delegates tasks.
• Accountability and Tracking: Paperclip solves the "black box" problem of AI agents by tracking token spend, task progress, and issue resolution. It prevents agents from stepping on each other by ensuring only one agent handles a specific task at a time.
• Bring Your Own Bot (BYOB): Users are not locked into one model. Paperclip allows the integration of various frontier models (Claude, GPT, etc.) via OpenRouter or local subscriptions.
• The "Memento" Problem: Agents often lack context. The system uses a "heartbeat" framework—a set of instructions that forces the agent to re-orient itself, check its assignments, and verify its goals every time it starts a task.

3. Step-by-Step Operational Process

1. Define the Company: Input the business idea and goals into the dashboard.
2. Hire Agents: The CEO agent creates a hiring plan. The user approves the hiring of specialized agents (e.g., a founding engineer).
3. Assign Tasks/Issues: Work is broken down into "issues." The user can manually approve tasks or set them to run automatically.
4. Configure Skills: Assign specific capabilities (e.g., Remotion for video editing, web browsing) to agents.
5. Monitor and Refine: Use the dashboard to track token spend and progress. If an agent fails to meet expectations, the user provides feedback or updates the agent's "rules" to improve future performance.

4. Key Arguments and Perspectives

• The Role of Human Taste: Dota argues that while AI can perform tasks, it cannot replicate human "taste" or "values." The human founder’s primary role is to communicate these values clearly to the agents.
• Beyond One-Shot Prompting: The creator emphasizes that "one-shot" AI startups often fail because they lack structure. Paperclip provides a persistent environment where agents can learn from past feedback and maintain long-term context.
• Future-Proofing: Even if underlying models become more powerful, the need for a management layer to organize and direct these agents will remain.

5. Notable Quotes

• "Your AI agents are Memento man. They wake up, they know how to fight, they know how to drive... but they don't know who they are. They don't know where they are." — Dota (on the necessity of the "heartbeat" instructions).
• "AI can do everything except know your values. You actually have to become more aware of your values and find out how to communicate them." — Dota.

6. Real-World Applications

• Security Reviews: A security firm uses Paperclip to automate security audits for clients.
• Non-Profit/Personal Management: A dentist uses the tool to organize foundation work and family tasks.
• Sales/Lead Gen: A roofing company is exploring using agents to aggregate satellite imagery and hail data to identify high-probability sales leads.
• Game Development: Users can import "Game Studio" agent teams to build games (e.g., a bullet-hell game) by delegating creative and technical tasks.

7. Synthesis and Conclusion

Paperclip represents a shift from "chatting with AI" to "managing an AI workforce." By providing a structured environment for agent interaction, it addresses the critical issues of accountability, cost management, and context retention. While the project is in its early stages and currently requires significant human guidance to ensure quality, its ability to import proven agent teams and integrate with various models positions it as a foundational tool for the future of autonomous business operations. The ultimate goal is to move toward "Maximizer Mode," where the system autonomously drives projects to completion with minimal human intervention.