Why "zero-human" is a dangerous fiction (and what we built instead)

The thing the marketing won't tell you

The phrase "zero-human company" gets attention. It also gets prosecutors' attention, plaintiff lawyers' attention, regulators' attention, and — if you're building seriously — the attention of every general counsel who has ever read FinCEN guidance.

Here is the short version: the Corporate Transparency Act (currently in legal flux but unambiguous in spirit), state BOI regimes (notably New York's LLCTA, effective 2026), KYC requirements at every financial institution in the US, fiduciary duties to creditors at insolvency, IRS trust-fund tax exposure, sales-tax responsible-party designations, the Air Canada chatbot case from 2024 establishing that a company is bound by its bot's statements, and common-law agency doctrine all require a natural-person principal somewhere in the loop. That principal is the founder.

This is not theoretical. The founder of any LLC that the platform "operates" is personally on the hook for unpaid payroll tax, unpaid sales tax, contracts the agent enters into, and material misrepresentations the agent makes to customers. An LLC member who claims "the AI did it" still owes the back taxes personally if the entity is undercapitalized.

So: zero-human is a marketing artifact. It cannot survive contact with the legal system.

What we built instead

We built a company OS that shrinks the human's required attention to about two hours a week — for governance only — and runs everything else autonomously. We call this AI-operated, human-governed.

The honest pitch is harder to land in a tweet. It is also fundable, defensible, buildable, and — crucially — does not blow up the customer's life in year three when the IRS sends a notice or the FTC opens an inquiry.

Concretely:

• Every agent action is classified into one of six categories before execution. Categories 4 and 5 require typed approval. Category 6 is hard-refused at the orchestration layer, not in a prompt. Prompt injection cannot escalate an action's category.
• The customer of record at every financial institution is the founder, not the platform. Stripe accounts, bank accounts, payroll accounts — owned by the founder's LLC. This keeps us out of money-transmitter classification (which would require 50-state licensing — multi-year, $5M+ effort) and gives the founder clean handoff if they leave us.
• We refuse certain business categories at intake. Healthcare, financial services, anything for minors, regulated professions, physical goods, hardware, crypto, anything political, anything gambling, anything adult. Refusal is enforced by an intake classifier and reviewed by humans for edge cases.
• We refuse cold outbound. TCPA damages are statutory ($500-$1,500 per message, class-actionable). The orchestrator will not send commercial SMS without documented prior express written consent, will not send commercial email to scraped lists regardless of CAN-SPAM technicality, will not solve CAPTCHAs.

The "two hours a week" promise is the actual product

The thing we sell is not "you don't have to do anything." It is "you do exactly the things only you can do, and we do everything else." Two hours a week of governance work. Approval queue on Monday, Wednesday, Friday. Compliance review monthly. Quarterly business review with the platform.

If you find yourself spending more than two hours, that's a design failure on our end. The cost dashboard surfaces it as an attention-budget overrun and proposes what to delegate further.

Why this is harder to sell — and why we're doing it anyway

A "zero-human" pitch lands easier with retail customers. It also lands with regulators in a way that ends companies. We watched several adjacent platforms ship the zero-human framing in 2024-2025 and we expect at least one of them to be in some form of regulatory hot water by 2027.

The defensible version of this category is the one that is rigorous about what AI can and cannot do under US law in 2026. The orchestration cleverness is commoditizing — frontier models do natively what required scaffolding two years ago. The durable moat is the compliance, governance, and operational-knowledge layer. We're investing there.

What we ask of you

If you sign up: you are the legal person. You sign tax filings. You upload your ID at KYC. You approve material commitments. You review the audit log periodically. We make all of this take two hours a week instead of fifty.

That's the trade. We think it's a fair one. If you wanted a literal zero-human company, you came to the wrong shop — and that company doesn't exist anyway.

If you want to see the honest architecture in motion — including the Approval Card flow, the action taxonomy in code, and the audit log — join the design-partner waitlist below.

Join the waitlist

— The 0h1bai team