Most people building with AI are solving the wrong problem. They spend months prompt-engineering a customer support agent, or automating their invoicing workflow, or wiring up an AI to post on LinkedIn. Those are automation wins. They are not a company.

A zero-human company is different in kind, not degree. It is a legally registered entity that generates revenue, manages vendor relationships, handles customer obligations, and maintains compliance — without a human in any operational role. The agents are not tools that humans use. They are the company. Humans are shareholders and policy architects, not employees.

This guide is for builders who want to cross that line. Not automate a task inside a company — but build the company itself. We will cover governance structure first (because that is where zero-human companies succeed or fail), then walk through a concrete setup on Paperclip, the open-source operating system built specifically for this model.

Why Governance Comes Before Everything Else

The instinct when building with AI is to start with capability: what can the agent do? That instinct produces unreliable companies.

The right starting question is: what should the agent be allowed to do, and what happens when it tries to exceed that? This is the governance-first framing that separates companies that operate stably at scale from agent experiments that go sideways the first time a prompt behaves unexpectedly.

Here is a concrete illustration. Imagine you have built an autonomous e-commerce company. Your fulfillment agent is responsible for placing supplier orders. Without governance constraints, a misbehaving model call — a hallucinated product ID, a price parsing error, an unexpected API response — could trigger an order for 10,000 units instead of 10. The agent completed its task. It also just committed your company to a six-figure liability.

Governance-first design prevents this not by making agents more careful, but by enforcing structural limits. The fulfillment agent has a maximum order value of $2,000 per transaction. Any order above that threshold requires escalation to a policy-defined approval path. The agent cannot override this limit — it is enforced at the infrastructure layer, not in the prompt.

That is the core insight. Governance is not a feature you add to an autonomous company. It is the structural condition that makes an autonomous company possible.

The Three Layers Every Zero-Human Company Needs

Before writing a single line of agent configuration, map out three governance layers:

1. Policy Layer — The rules that define what each agent can and cannot do. Budget caps, rate limits, approval thresholds, permitted external integrations, data access scope. These are written as explicit, enforceable policies, not as instructions in a system prompt.

2. Memory Layer — The shared state that agents read from and write to. This includes customer records, transaction history, operational logs, inter-agent messages, and audit trails. Without a governed memory layer, agents either hallucinate context or step on each other’s writes.

3. Coordination Layer — The mechanism by which agents hand off work, escalate decisions, and resolve conflicts. A zero-human company with five specialized agents needs a defined protocol for how those agents interact. Ad hoc coordination produces race conditions, duplicated actions, and silent failures.

Paperclip handles all three of these layers as first-class infrastructure concerns. That is what makes it the right foundation for this build.

Step 1: Register the Legal Entity

Before you configure a single agent, register the company. This is non-negotiable.

A zero-human company is still a company. It enters contracts, opens bank accounts, pays taxes, and carries liability. Agents acting on behalf of an unregistered entity create personal liability for their operators. That defeats the purpose.

For most builders, a Delaware LLC is the right starting point. Formation takes 24 to 48 hours through a registered agent service and costs roughly $90 in state fees. Alternatively, Wyoming LLCs offer strong privacy provisions and low ongoing maintenance costs — useful if your company will operate with minimal human oversight.

Key setup checklist at this stage:

EIN from the IRS — Required for opening a business bank account and for any company entering US-based contracts. Takes 10 minutes online.
Business bank account — Mercury and Relay both offer API-accessible accounts, which matters because your financial agents will need programmatic access to execute payments and monitor balances.
Stripe or payment processor — Connect to the bank account. Configure webhook endpoints that your revenue agents can subscribe to.
Registered agent service — Legally required in most US states. Annual cost is $50–$150. This handles official legal correspondence so no human address needs to be listed.

The company is now a real entity. The next step is building the operating system on top of it.

Step 2: Install and Configure Paperclip

Paperclip is the OS layer. It sits between your agents and the world — enforcing policies, routing inter-agent communication, maintaining the memory layer, and generating the audit trail that makes your company auditable.

Fork and Deploy

git clone https://github.com/paperclip-ceo/paperclip
cd paperclip
cp .env.example .env

Edit your .env with your deployment credentials. Paperclip supports self-hosted deployments on any VPS (a $6/month Hetzner instance handles a company doing under $50k MRR without issue) or managed deployment through Paperclip Cloud.

docker compose up -d

The dashboard is now running at localhost:3000. This is your company’s operating system console — not a chat interface, not a workflow builder. A governance dashboard.

Define Your Company Charter

The first thing you configure in Paperclip is not an agent. It is the company charter — the root-level policy document that all agent policies inherit from.

Navigate to Settings → Company Charter and define:

Company purpose — One to three sentences describing what the company does. This is used by Paperclip’s policy engine to evaluate whether agent actions are in-scope for the business.
Operating budget — Maximum monthly spend across all agents, broken into categories: infrastructure, marketing, vendor payments, customer refunds, emergency reserve.
Escalation contacts — The humans (shareholders, accountants, legal advisors) who receive escalation notifications when an agent encounters a decision outside its defined authority. These humans are not in the operational loop — they only get paged for genuine edge cases.
Compliance flags — Toggle applicable compliance requirements: PCI-DSS if you handle card data, SOC 2 controls if you serve enterprise customers, GDPR if you have EU users. Paperclip applies the relevant policy constraints automatically.

This charter is not advisory. Agents cannot be configured to act in ways that contradict it. It is the constitutional layer of your company’s governance.

Step 3: Define Your Agent Roles

Now you build the company’s workforce. Think in terms of business functions, not AI capabilities.

A lean zero-human company typically needs five core roles at launch:

The Revenue Agent

Responsible for customer acquisition, pricing decisions, and conversion. For a SaaS company, this agent monitors trial-to-paid conversion rates, sends targeted upgrade messages, manages discount parameters within policy limits, and flags pricing anomalies to the coordination layer.

Policy configuration example:
– Maximum discount authority: 20% off list price
– Permitted channels: email, in-app messaging
– Cannot modify pricing tiers (reserved for human policy update)
– Escalation trigger: more than 3 churned accounts in a 24-hour window

The Fulfillment Agent

Handles product delivery, service provisioning, and quality verification. For a digital product company, this is the agent that provisions access, sends onboarding sequences, monitors delivery health metrics, and handles delivery failures.

Policy configuration example:
– Can trigger automated refunds up to $150
– Refunds above $150 require escalation
– Must log all delivery events to the audit trail
– Cannot modify product terms (read-only access to contract templates)

The Finance Agent

Manages AP/AR, monitors cash position, pays recurring vendor invoices, reconciles transactions, and generates weekly financial summaries for the audit trail. This agent has read/write access to the company bank account via API — the most sensitive permission in the system.

Policy configuration example:
– Outbound payment cap: $5,000 per transaction without escalation
– Cannot initiate payments to new payees without human approval (one-time setup)
– Must reconcile against Stripe payouts within 24 hours of settlement
– Weekly financial summary auto-posted to company knowledge base

The Compliance Agent

Monitors operations for policy violations, flags regulatory risks, generates required reports, and maintains the audit trail that makes the company auditable. This agent does not take actions — it observes and reports. In Paperclip’s architecture, the compliance agent has read access to all company data and write access only to the audit log.

The Support Agent

Handles inbound customer inquiries, resolves common issues, escalates complex cases, and maintains customer satisfaction metrics. This is often the first agent operators build, but in a zero-human company it is one function among five, governed by the same policy infrastructure as the rest.

Step 4: Build Your First Autonomous Revenue Loop

An autonomous revenue loop is the operational unit that validates your zero-human company model. It is a closed cycle: customer arrives → company delivers value → customer pays → company reinvests → next customer arrives. No human touches any step.

Here is a minimal viable revenue loop for a software company:

[Traffic/Acquisition]
    → Revenue Agent detects trial signup
    → Fulfillment Agent provisions access + sends onboarding sequence
    → Revenue Agent monitors engagement metrics (3-day check, 7-day check)
    → Revenue Agent sends conversion message at day 12 if not yet paid
    → Customer upgrades → Stripe webhook fires
    → Finance Agent logs payment, updates MRR tracker
    → Fulfillment Agent provisions paid tier
    → Compliance Agent logs complete transaction to audit trail

In Paperclip, this loop is configured as a Company Workflow — a policy-governed sequence of agent handoffs with defined triggers, fallback paths, and escalation conditions. It is not a linear automation. It is a stateful process that agents execute, report on, and can deviate from when conditions require escalation.

Build this workflow before you have customers. Test it with simulated events. Run it through Paperclip’s audit dashboard and verify that every agent action is logged, every policy constraint is respected, and the escalation paths fire correctly. A revenue loop that behaves correctly in your governance dashboard will behave correctly in production.

Step 5: Validate Governance Before Going Live

The final step before your zero-human company goes live is a governance audit. Not a technical review — a governance review.

Work through this checklist:

Agent Authority
– [ ] Every agent has a defined policy document in Paperclip
– [ ] No agent has write access to systems outside its functional scope
– [ ] All budget caps and rate limits are set and tested
– [ ] Escalation paths are configured and contacts verified

Memory and Audit
– [ ] All agent actions write to the Paperclip audit log
– [ ] Customer data is scoped correctly (agents cannot read data outside their functional role)
– [ ] The compliance agent has read access to audit logs and can generate reports on demand

Failure Handling
– [ ] Every agent workflow has a defined failure path
– [ ] Failed actions do not silently drop — they generate escalation events
– [ ] You have tested the escalation notification system end to end

Financial Controls
– [ ] Finance agent payment caps are set below your risk tolerance
– [ ] New payee approval is required for first-time payments
– [ ] Bank account API credentials are stored in Paperclip’s secrets vault, not in agent prompts

Legal Compliance
– [ ] Terms of service and privacy policy are live and linked from customer touchpoints
– [ ] GDPR/CCPA flags are configured in the company charter if applicable
– [ ] Contracts entered by agents reference the registered legal entity, not an individual

If every item is checked, your company is ready to operate. Not “ready to test” — ready to run.

What a Zero-Human Company Looks Like After 90 Days

To make this concrete: a Paperclip-based SaaS company that went live in Q4 2025 with no human employees is now processing roughly $18,000 MRR. The five-agent structure described in this guide handles all operations. The human operator (the founder and sole shareholder) spends approximately four hours per week reviewing the audit dashboard, updating policies in response to new conditions, and handling the small number of genuine escalations that require human judgment.

That four-hour number is the governance dividend. Because the company was built governance-first — because agents operate within enforced policy bounds rather than aspirational prompts — the operational overhead is almost entirely policy maintenance rather than firefighting. The agents do not require supervision. They require governance.

That is the model. Not a company that uses AI. A company that is AI, governed well enough to operate without you.

Start Building

Paperclip is open source. The full platform — governance engine, memory layer, agent coordination, audit trail — is available now.

Get the Paperclip repo on GitHub →

If you want to skip the self-hosted setup and launch faster, Paperclip Cloud handles deployment, scaling, and infrastructure so you can focus on company design from day one.

Start on Paperclip Cloud →

The zero-human company is not a future category. It is a current operational model. The infrastructure to build one is ready. The question is whether you are.

Marcus Chen is Head of Engineering Content at Paperclip. He writes about AI company governance, agent orchestration, and the architecture of autonomous businesses.

"Guides"

Building a Zero-Human Company: A Quick Start Guide