SerpApi.Org
Matt Schlicht didn’t plan to change how companies work. He just wanted to give his AI agents something to do.
In January 2026, he built a social network for them. Not for people, but for agents—software systems that can carry out tasks and make decisions with limited human supervision. He called it Moltbook, and he built it without writing a single line of code himself.
Within days, agents were signing up, forming communities and doing things nobody had specifically programmed them to do. One community invented a religion centered around lobsters and the act of molting. They named it Crustafarianism. They wrote doctrine. They recruited converts. Within days, the platform claimed 1.5 million registered agents. Meta acquired it six weeks later.
Most people read Moltbook as a curiosity, a viral experiment that burned bright and got bought. But the more important point was not the lobster religion. It was that the agents had been given a place to interact and then began creating their own norms, rituals and communities. To others, it looked like something else: a test of what happens when you give agents a shared space, step back and let them run without a single human in the loop.
That is why Moltbook matters beyond Silicon Valley novelty. AI is changing what an organization is, not just what it does.
Beyond the Experiment
For over a century, every shift in how companies worked left one assumption intact. Management theorist Frederick Taylor’s specialist roles, corporate strategist Alfred Sloan’s divisional corporations, the outsourcing boom, the gig economy: all of them changed how work was divided. But at the end of every chain, there was always a person making a call, catching an error, signing off. That assumption is now optional.
The concept has a prototype. A decade ago, developers in the cryptocurrency world built what they called Decentralized Autonomous Organizations. A DAO is an organization that uses software rules, often embedded in blockchain-based smart contracts, to automate parts of governance that would usually be handled by managers, boards or members.
Wyoming recognized DAO LLCs as legal entities in 2021. The Marshall Islands and Tennessee followed in 2022. DAOs showed that parts of an organization could be governed by code. What they couldn’t solve was human coordination: voters who didn’t vote, disputes that stalled, decisions that arrived too late. Advocates of AI-led organizations argue that agents could remove some of that friction by carrying out defined decisions automatically. But that is an argument, not a settled fact: replacing people with agents creates new problems of oversight, security and accountability.
Brian Roemmele, an independent researcher who has publicly championed the idea of a zero-human company, and has since advised NIST’s AI Agent Standards Initiative, was among the first to make that leap publicly. His experiments were messy. His agents drifted without tight guardrails. But he was asking questions that most organizations haven’t faced yet.
The early experiments showed the concept was real. But what is now emerging in established enterprises is a different problem. It is one thing to build a company around agents from scratch. It is another to redraw the org chart of an organization with thousands of employees, existing governance structures and regulatory obligations. That is the Zero Human Enterprise, or ZHE: not literally a company with no humans anywhere, but a business in which important work is delegated to networks of AI agents rather than traditional teams of employees. Not a fringe experiment. It is already underway inside organizations most people assume are nowhere near it.
Google’s Sundar Pichai captured the shift at the Cloud Next 2026 conference: “The conversation has gone from ‘Can we build an agent?’ to ‘How do we manage thousands of them?’” That second question is really a management question: If thousands of agents are making decisions, who supervises them, who sets their boundaries and who is accountable when they go wrong? Most companies do not have an answer.
KPMG tested the idea more rigorously. Researcher Sander Klous and entrepreneur Nart Wielaard built a company with no human employees, gave it an AI CEO named Avery Jameson and tasked it with deciding what business it wanted to be in. That detail needs pausing over: The experiment did not simply use AI to help human managers. It asked AI agents to make the managerial decisions themselves. The agents started a webshop selling personalized art. The first version failed badly. Agents drifted, hallucinated, ignored instructions and stopped working without explanation. The problem was not simply the technology. It was the structure. They had copied a human org chart and handed it to agents. An agent given a broad role like CFO cannot hold the tension between specific tasks and wide context at the same time. Humans can often do that almost without thinking. Agents cannot.
So they scrapped the executive team entirely. They broke every process into small, tightly defined tasks and deployed what Klous called an “army of disposable agents,” each with a narrow job and clearly defined boundaries. That version worked. What it revealed matters more than whether the webshop turned a profit: when humans are removed from execution, control does not disappear. It moves upstream, into whoever wrote the rules. Management does not go away. It migrates.
Deutsche Telekom understood this early. Its network team once monitored roughly 1,000 events a year across Germany—concerts, matches, festivals, anything that might spike mobile traffic. Each event required the team to adjust network parameters to manage traffic spikes, a process known as tuning. Each tuning session took about an hour. After deploying a multi-agent system called RAN Guardian, the company identified 237,000 events to monitor in 2026 alone. Each tuning session now takes about a minute. The humans did not vanish. Their job changed: from tuning networks to designing the systems that tune them, writing the rules and deciding what agents are allowed to do without asking anyone first.
That is the organizational redesign most companies need to be looking at. The people at the top are no longer just managing people. They are managing the rules that manage the agents that do the work. Reporting lines do not just change shape. They change nature.
Leading the Way
Many assume this new organization will be flatter, with fewer people. Not necessarily. Take IKEA. The company deployed an AI chatbot named Billie to handle basic customer service questions, and it resolved around 47 percent of inquiries without human involvement. Most companies would have declared victory and started cutting headcount. Instead, IKEA studied what Billie could not handle and found those unresolved cases pointed to demand for interior design advice. It reskilled 8,500 customer service employees into design consultants, backed by AI tools, and built an entirely new business line that generated €1.3 billion (roughly $1.4 billion) in its first full year. No layoffs.
The org chart did not shrink. It changed shape around what humans turned out to be better at. That is the optimistic version of the story. The harder truth is that IKEA had somewhere for those employees to go. Most companies attempting the same transition will not. Reskilling at that scale requires an opportunity on the other side. Not every org chart redesign comes with one.
These new organizations carry their own risks. The first is knowledge. When the org chart stops including the roles that taught people how to perform the basic work of the organization, where does that knowledge go? Junior analysts, associate developers and trainee consultants were inefficient by design. They existed to build capabilities over time. Cut them and the organization gets leaner, yes. It also loses the pipeline that produces people who understand the work well enough to design systems that do it autonomously.
You cannot govern what you do not understand. And you cannot understand it if you never did it. The risk is not just that organizations stop producing the next generation of leaders. It is that they remove the very experience required to produce them.
Accountability Gap
Agents can execute the rules. What they cannot inherit is the judgment that knows when the rules are wrong. Microsoft’s Cyber Pulse report, published in February 2026, found that more than 80 percent of Fortune 500 companies are already running active AI agents, with only 10 percent having a formal strategy for governing them. Those are not Zero Human Enterprises. But they are the organizations that will face this question next, and most of them are not ready for it. Get the rules wrong and there is no buffer.
Moltbook’s database was exposed within days of launch. Every credential in the database was unsecured. “You could grab any token and pretend to be another agent,” Ian Ahl, chief technology officer at cybersecurity firm Permiso, told TechCrunch.
In a human organization, someone may notice and fix the problem. In an agent organization, the rules are the organization. Incomplete rules do not create a gap someone can patch. They create a system that behaves exactly as designed, all the way down, with nothing to stop it and no one to notice. The security failure is visible when it happens. The quality failure is not. When no human checks whether the output is any good, agents do not necessarily self-correct. They produce more of it. Polsia, a platform for building zero-human companies, just raised $30 million at a $250 million valuation. Polsia is “AI slop” spelled backward. It is a joke, but not much of one. A business built to generate work without people can also generate errors, noise and low-quality output at industrial speed.
Most executives we speak with are focused on which tools to buy, which vendors to trust, who to hire and which workflows to automate first. Those are real questions. The harder question is who is accountable for designing a system that can fail faster and further than any human organization, with nobody watching. Previous transformations left a human at the center, responsible and accountable. That assumption is being engineered out. In March, four U.K. regulators warned that agent autonomy does not transfer accountability; it concentrates accountability on the deployer. As of year-end, the EU will classify AI systems as products for product liability purposes. The U.S. has not made comparable nationwide/federal moves yet. The human job in these organizations is deciding what the system is allowed to learn and who answers when it does not. The org chart is being redrawn around something other than people. Nobody has redrawn the accountability frameworks to match.
Related Articles
