How To Run a Zero-Human Company (Paperclip)

Key Concepts

• Paperclip: An open-source, multi-agent orchestration dashboard designed to manage teams of AI agents.
• Multi-Agent Orchestration: The process of coordinating multiple specialized AI agents (e.g., CEO, CTO, CMO) to work toward a unified business objective.
• VPS (Virtual Private Server): A dedicated, secure, and persistent environment for running AI agents 24/7, independent of local hardware.
• Agentic Workflow: A paradigm where humans act as "Board of Directors," providing high-level goals while AI agents execute tasks autonomously.
• Observability: The ability to monitor agent decision-making, tool calls, token usage, and costs in real-time.
• Token Economics: The strategic management of AI model usage costs to ensure operational efficiency.

1. The Role and Utility of Paperclip

Paperclip functions as a "company" for AI agents. Unlike running agents in isolated terminals, Paperclip provides a centralized dashboard that solves three primary issues:

• Persistence: Agents run 24/7 on a VPS, ensuring work continues while the user is offline.
• Model Agnosticism: It supports various models (Claude, OpenAI, Gemini) without bias.
• Transparency: Users can trace every tool call, decision, and dollar spent, preventing agents from going "off-track" unnoticed.

2. Setup Methodology: The VPS Approach

The author emphasizes that running agents locally is a mistake because it lacks persistence and security.

• Security: Running agents on a VPS isolates them from the user's personal files and passwords.
• Deployment: The author recommends using a one-click deployment preset (e.g., via Hostinger) to avoid complex terminal configurations.
• Configuration: After deployment, users must input API keys (specifically recommending Claude/Anthropic due to current bugs in the Codex CLI).
• Authentication: Users must secure their dashboard with admin credentials to prevent unauthorized access to the agentic infrastructure.

3. Organizational Framework

Paperclip allows users to structure their "company" in various ways:

• Hierarchical: A "Board of Directors" (Human) oversees a "CEO Agent," who manages a "CTO" and "CMO," who in turn manage specialized agents.
• Flat: A decentralized structure similar to Nvidia’s organizational model.
• Key Components:
  • Issues: Individual tasks assigned to agents.
  • Routines: Recurring tasks (cron jobs).
  • Goals: High-level, long-term mission statements.
  • Orchestration: The ability to create agents that report to other agents (e.g., a Cybersecurity Researcher reporting to the CTO).

4. Strategic Implementation and Best Practices

• Start Small: Do not attempt to automate everything at once. Begin with research or coding tasks, then expand.
• The "Childlike" Mindset: The author argues that success in AI requires abandoning 2024-era dogmas. Treat AI as an experimental science (like Thomas Edison’s approach to the light bulb).
• Pricing Time: View AI token spend as an investment. If an agent saves five hours of work, the cost of tokens is a net gain, regardless of the dollar amount.
• Monetization Opportunity: There is a short-term market opportunity (30–60 days) to offer "Paperclip Setup Services" to businesses that lack the technical expertise to build their own agentic org charts.

5. Critical Perspective on UI/UX

The author provides a nuanced critique of Paperclip:

• Current Limitations: The UI currently mimics traditional project management tools (Slack/Jira/Linear), which are designed for human-to-human interaction.
• Future Vision: The author predicts that the future of agentic work will not be a dashboard, but a proactive, adaptive interface where agents present information only when necessary, potentially utilizing voice or neural-link inputs.
• The "Human-in-the-Loop" Role: The human’s primary value is providing "unique judgment" and "taste," which LLMs—being next-token predictors—cannot generate on their own.

6. Personal Workspace Structure

The author shares his own "brain" folder structure, which serves as a blueprint for managing AI-human collaboration:

• Brain/Goals/Work: Folders containing markdown files for quarterly goals, business state, and daily priorities.
• Data/SQL: Using SQLite databases for business lessons and behavior tracking allows agents to query the user's history efficiently.
• Skills: A folder of "skills" that are loaded contextually, allowing agents to perform specific tasks without needing massive, bloated instructions.

Synthesis/Conclusion

Paperclip represents a significant step toward autonomous, agentic business operations. While the current user interface is a "first-generation" attempt that relies on familiar project management metaphors, the underlying capability to orchestrate multiple agents on a persistent VPS is a game-changer. The primary bottleneck for users is not the AI models themselves, but the user's ability to structure their goals, manage their workspace, and treat AI as a scalable investment rather than a simple chatbot. Success requires a shift from "using AI" to "building an agentic organization."