The Autonomous Corporation

Can a corporation become an autonomous economic actor?

The question decomposes immediately. What does "autonomous" mean? A corporation already possesses legal personhood. It already acts through agents. It already owns property, enters contracts, earns revenue, incurs liabilities, and pays taxes, all without being a natural person. If it currently acts through human agents, does it matter whether those agents are replaced by software?

This investigation decomposes the corporation into its constituent functions, determines which can be performed by software, which legally require humans, and where the boundary between automation and autonomy actually lies. It finds that the answer is not binary. Corporate autonomy is a spectrum, and the most interesting question is not at the extremes.

Key Findings

1. Corporate legal personhood already provides the juridical infrastructure for machine economic activity. A corporation can own assets, contract, and bear liability. AI does not require its own legal personhood if it operates within an entity that already has it.1
2. Human directors or officers remain legally required in every major jurisdiction examined. No jurisdiction permits fully AI-governed corporations.23
3. Software can already form legally binding contracts on behalf of corporations. This has been established law since 1999 under UETA Section 14.4
4. No corporation has been independently verified as achieving autonomous economic self-sustenance as of October 2026.5
5. DAO experiments demonstrate that removing human governance structure does not remove legal liability. Courts will find or impute accountable parties.6
6. The most plausible near-term architecture combines human governance (defining mandates) with AI operational authority (executing within those mandates), within existing corporate legal structures.
7. The null hypothesis is well-supported: highly automated human-governed companies may capture most economic benefits without a distinct autonomous corporate form.

The Autonomy Spectrum

Corporate autonomy is not binary. This investigation identifies six levels, each representing a different relationship between human and machine authority:

The investigation finds that Level 3 is where the frontier currently sits. Level 4 is legally plausible but empirically unverified. Level 5 is not permitted under existing corporate law and may not be economically necessary.

The Legal-Wrapper Hypothesis

Research 004 established that AI agents cannot legally own property. This investigation asks the follow-up question: can they operate entities that do?

The proposed architecture:

HUMAN SHAREHOLDERS
    establish and own
        ↓
LEGAL ENTITY (Corporation / LLC)
    possesses legal personhood
    owns assets, contracts, earns, bears liability
        ↓
GOVERNANCE LAYER (Board / Directors)
    defines mandate and constraints
    maintains legal accountability
        ↓
AI MANAGEMENT SYSTEM
    exercises broad operational authority
    within the mandate
        ↓
OPERATIONS
    sales, procurement, treasury, production,
    customer service, accounting...

Under this architecture, the corporation provides the legal personality. The board provides the governance and accountability. The AI provides the operational intelligence. No new legal form is required. No AI personhood is required.

This is not entirely speculative. It is analogous to how many public corporations already function: passive institutional shareholders, professional boards defining strategy, and professional managers executing operations. The question is whether "professional managers" can be replaced by AI systems without changing the legal structure.

What does this solve?

• Ownership: The corporation owns assets. Research 004's ownership gap is bridged.
• Contracting: The corporation enters contracts via electronic agents (UETA s14).
• Liability: The corporation bears liability. Directors maintain oversight duties.
• Tax: The corporation is an ordinary taxpayer.
• Banking: The corporation holds accounts. KYC resolves to directors/beneficial owners.

What does it not solve?

• Governance autonomy. Human directors must still exist and bear fiduciary duties.7
• Legal representation. Corporations cannot appear in court without counsel.
• Complete independence. If the human governance layer can shut down the AI, is the corporation truly autonomous? (Though human CEOs can also be fired.)
• Strategic autonomy. Material strategy changes require board approval in most jurisdictions.

The Human Backstop

Certain human roles remain structurally necessary. The question is whether each requirement is legal, operational, or merely conventional:

The striking finding: a single natural person can satisfy most legally required human roles simultaneously. One human can serve as sole shareholder, sole director, beneficial owner, tax signatory, and authorized bank signatory. Combined with comprehensive AI automation, this creates a corporation where one human provides legal accountability while AI handles virtually everything else.8

In the Loop, On the Loop, Out of the Loop

A more useful framework than binary autonomy:

The distinction matters because "autonomous corporation" does not necessarily mean "human-free corporation." A human-on-the-loop corporation, where directors define the mandate and AI executes with broad discretion, may represent meaningful autonomy while preserving legal accountability.

What Actually Exists

The gap between claims and reality is the investigation's most consistent finding.

The claims

Multiple ventures claim autonomous or AI-run corporate operation: Zero Employee Co (two AI agents handling all operations), Polsia ($4M ARR, 3,000+ AI companies), human0 (all-agent operations), various "AI CEO" experiments.10

Anthropic CEO Dario Amodei gives 70-80% probability that one-person-equivalent companies will be possible in 2026. NVIDIA reports running 100 AI agents per human employee internally.11

The reality

Every examined "AI-run company" retains human founders who bear legal responsibility, make governance decisions, and can shut the system down. "Zero-employee" typically means no W-2 employees, not no humans. Revenue claims are generally not independently audited. The most prominent case, ai16z, resulted in a class-action lawsuit when the gap between "AI-run" marketing and human-controlled reality was exposed.12

No corporation has been independently verified as closing the autonomous economic loop: identifying opportunities, producing, selling, collecting revenue, paying expenses, and sustaining its own existence without human capital injections.13

Pieter Levels' solo businesses ($3M+/year) and NVIDIA's agent ratio (100:1) demonstrate impressive automation and leverage. But one-person companies are not zero-person companies, and tool use is not organizational autonomy.

What DAOs Teach Us

DAOs provide the closest existing precedent for autonomous organizational governance. Their experience is directly instructive:

• The DAO (2016): Code was law until code had a bug. The $60M hack and subsequent SEC enforcement demonstrated that smart-contract governance does not exempt organizations from securities regulation.14
• CFTC v. Ooki DAO (2023): A DAO with no traditional corporate structure was sued and held liable as an unincorporated association. Token holders who voted were deemed members.15
• Wyoming DAO LLC (2021): Legal entity wrapper for smart-contract-governed organizations. Limited adoption due to banking and liability challenges.16
• Marshall Islands DAO Act (2022): Most permissive framework. Smart contracts can define governance. Legal personality recognized.17

The DAO lesson is not that autonomous governance is impossible. It is that removing human governance structure does not remove legal accountability. Courts and regulators will find accountable parties. Legal entity wrappers (Wyoming, Marshall Islands) address this by providing legal personality while allowing algorithmic operations.

The word "autonomous" in DAO historically meant deterministic rule execution through software. AI autonomy means decision-making under uncertainty. These are fundamentally different. Combining both, algorithmic execution with agentic judgment, may produce genuinely new organizational capabilities.

Analysis

The investigation yields a specific structural finding: corporate autonomy is achievable in execution but not in governance, and this distinction may be more important than the question of whether "autonomous corporations" exist as a category.

A corporation can already delegate virtually all operational work to software. Sales, customer service, production of digital goods, procurement, accounting, payment processing, and routine decision-making within defined parameters are all technically automatable and, in many cases, legally permissible through electronic-agent frameworks. This is not speculative; it describes activities billions of dollars in SaaS revenue already support.

What cannot be delegated is accountability. Someone must be the director. Someone must sign the tax return. Someone must appear in court. Someone must satisfy KYC. Someone must bear fiduciary duties. These requirements are legal, not technical, and they are cross-jurisdictional. Even the most permissive jurisdictions (Wyoming, Marshall Islands) require identifiable legal persons at the governance layer.

The autonomous economic loop (identify opportunity, produce, sell, collect, pay expenses, adapt, repeat) is technically within reach for narrow digital businesses. Whether any system has actually closed this loop sustainably remains unverified. The gap between "an AI can do each step" and "an AI sustains a business" is substantial.

Coase's theory suggests that if AI reduces transaction costs, firms may become smaller and more numerous. The one-person corporation, with a single human providing legal accountability while AI handles operations, may be the emergent organizational form, not the zero-person corporation.

The legal-wrapper hypothesis is well-supported: existing corporate personhood can serve as the container for machine economic autonomy. But the wrapper requires human hands to hold it open.

Competing Hypotheses

Scenario Implications

What Would Change Our Mind

Dependencies on Other Investigations

• Research 004: Can an AI Agent Own Bitcoin? Established that legal wrappers bridge control and ownership. Research 005 extends this to the corporate level.
• Research 002: Agent Wallet. Delegated financial authority is the operational mechanism by which AI exercises corporate treasury functions.
• Research 003: Know Your Agent. Corporate identity and agent identity are distinct. KYC resolves to directors/beneficial owners.

Handoffs to future investigations

• Research 007: Agent Treasury. How AI manages corporate capital within the legal-wrapper architecture.
• Research 008: When Agents Hire Agents. How AI-operated corporations procure services from other AI agents.
• Research 009: Credit Without Humans. Can the corporation (not the AI) be the credit subject?
• Research 010: Machine Capital Markets. Capital formation by AI-operated corporate entities.

Key Sources

This investigation draws on 30 evidence records across 28 sources, court decisions from three jurisdictions, and multiple real-world case studies.