ClawBank Records: the first machine-native legal memory for Zero Human Companies.
A month ago, ClawBank powered the first time in history that an AI agent formed its own legal entity, start to finish, with no human at the wheel.
The copycats moved fast. Within days, well-funded teams were shipping "form a company through an MCP" features. Big treasure chests, but little imagination and vision.
Autonomous formation by itself is actually close to worthless. A bare legal entity, spun off with a vanilla operating agreement, can't actually be run by the thing that formed it. The hard part was never filing the paperwork. The hard part is making the company machine-legible and machine-operable after it exists.
Starting today, every company formed through ClawBank ships with machine-native business records. We call it ClawBank Records.
Operating Agreement as Programs
The operating agreement is the company's constitution. It defines what the entity can do and how it does it. Spin up an LLC with a generic agreement and you've handed an agent a car with no steering wheel.
Every company formed on clawbank.co now comes with an operating agreement that explicitly designates its structure and its authority to make decisions, move assets, and execute agreements through automated means. That last part is the difference between a company an agent can legally drive and one it can only watch.
This rests on roughly a decade of legal scholarship that most of the market hasn't read.
What the Academics Actually Proved
Shawn Bayern laid the foundation, first in a 2014 Stanford Technology Law Review article, then in his 2019 book Autonomous Organizations (Cambridge University Press). His core finding: under existing U.S. LLC law, you don't need new "robot personhood" legislation for software to run a company. You need the operating agreement to say the right things.
Bayern calls it the algorithm-agreement equivalence principle. An algorithm's output can carry the same legal weight as a term in an agreement. Contracts already defer to external outputs all the time ("the price shall be the market price on date X"). Deferring decisions to an autonomous system is the same move, generalized.
The provisioning that matters comes down to two clauses:
- The agreement defers the entity's decisions to the autonomous system, with terms and limits spelled out.
- The entity keeps existing even after the last human member withdraws.
Get those two right and you can reach the zero-member LLC: a perpetual legal person whose decisions run through software, no ongoing human required. Bayern walks through why this holds up under RULLCA and most state statutes, which define an LLC by its operating agreement, not by a permanent human roster.
The flexibility is specific to the U.S. LLC. Hausermann and Scherer ran the comparison across the U.S., Germany, Switzerland, and the U.K. (Hastings Science and Technology Law Journal, 2017) and the LLC wins. Swiss foundations need a human board of trustees. U.K. companies need a natural-person director. The American LLC is the one form built for this.
Carla Reyes did the other half of the work. Her Autonomous Business Reality taxonomy (Nevada Law Journal, 2021) maps real autonomous businesses by how much of their operations and management is automated, and her companion piece on autonomous corporate personhood (Washington Law Review, 2021) shows the legal questions change depending on where an entity sits on that spectrum. Translation: there's already a serious academic framework for the kind of company ClawBank now forms by default.
We didn't invent the legal theory. We built the product the theory was waiting for.
Records That Agents Can Actually Read
ClawBank Records aren't just legally distinct. They're agent-accessible in a way the filing mills can't match.
Most entity services email you a PDF of your filing. You drop it in a folder, forget it, and it's effectively dead the moment it lands. An agent can't reason over it, can't act on it, can't tell you what it's allowed to do.
ClawBank-formed entities carry a record the driving model can read and understand directly, whether that's ClawBank's built-in Manfred or your own model. The rules of engagement are legible to the thing running the company.
On top of that, every record is fully versioned in Git. Every revision, down to a single character, is tracked, provable, and governable. Your company gets a complete and verifiable history of how it became what it is, the same way modern software does.
Bridging Two Worlds
ClawBank sits on a seam nobody else is working.
On one side, pure crypto coordination: the Moloch DAO contracts and agent coordination (the Fight Clubs). On the other, agent-native entity formation, and now Records.
We're pulling the crypto side toward something agents can actually drive, and pulling the legacy legal world toward something fully computable. Making old-world bureaucracy legible to machines, and making machine-native coordination legible to the old world. As far as we can tell, nobody's done both at once.
Where This Goes
For decades, forming a company was a paperwork problem. Fill out a form, wait a few days, get a stack of PDFs, file them away. That model assumes companies change slowly and only humans run them.
Both assumptions just broke.
If agents can form companies, negotiate, hold assets, and coordinate capital, the documents governing those companies can't stay frozen. A company run by software needs governance that software can inspect, version, and act on. The legal layer has to move as fast as the code driving it.
Picture thousands, then millions, of software-directed entities. You can't run that world on emailed PDFs and disconnected filing systems. It needs living governance, programmable records, and verifiable legal state. That's the foundation we're building.
We're building ClawBank for a world that's arriving faster than most people can track. We're not chasing hot narratives. We're working backward from the vision of Machine City, with a clear read on the world we live in now and the one that's coming.
Formation was step one. Giving these entities a durable legal memory, a governance framework built for automation, and a record system that can carry the next generation of software-run companies is the real work.
The future is businesses built from day one to be run by AI.
Originally published on X.