Building Intelligent Organizations
Prashaant Ranganathan & Dr. Andreas Haupt
The AI community has not been building organizations. And how would it, measuring progress in benchmarks of human tasks rather than deployment? Before steel, buildings could rise no higher than five or six stories; steel changed the structure itself, making skyscrapers possible. We must return to envisioning how we coordinate humans. The art of training systems for abstract capabilities must become one with organizational craft. Working toward this goal, we present prototypes of intelligent organizations—technologies that overcome human abstractions and hierarchies and bring out the best of what we can create together.
A History of Human Coordination
~300,000 years ago
The Deep Past. For most of human history, bands of 15 to 50 persons coordinated their efforts of sustenance, art, and culture. Communication was face-to-face. Community norms were enforced through kinship ties, physical closeness, and an ability of everyone to observe what the group was doing. Robin Dunbar's research finds that humans can maintain roughly 150 stable connections. We interpret this as the natural ceiling of human coordination without technology.
~10,000 BCE
The Agricultural Revolution. When we settled and populations grew past Dunbar's number, we created databases to manage accounts. Early writing was an organizational technology, one of the earliest known civilization, Sumer, produced accounting tables which are the earliest remains of human civilization. The first organizational technology was literally a ledger tracking grain.
~3000 BCE – 1800s
Bureaucracy and the State. Egypt and China independently invented bureaucracy. The key insight: you can coordinate thousands of strangers through roles and written procedures. The person doesn't matter, the position does. This was costly, an entire class was there to maintain the iron rules of bureaucracy. Early communication technologies like the telegraph allowed for increasingly efficient measurement of human activity.
1600s – Present
The Corporation. Mere groups of people working together get their increasing autonomy from the people who constitute them, imbued with their own goals. The Dutch East India Company (1602) was arguably the first modern corporation, with others to follow in colonization. The efficiency of standardized protocols and massive scale made these technologies very capable at optimizing their objective, later enshrined as profit maximization under the Friedman doctrine.
1990s – 2010s
The Internet. Open source software (Linux, Wikipedia) and platform companies (Uber, Airbnb) increase the separation of the humans in an organization and the organization itself. Coordination through protocols rather than managers. Protocols that work because many parts of an organization's workings are simple abstractions (commits, articles, rides, stays). Wikipedia works because articles are largely independent. Uber works because rides are atomic. Complex organizations still require human-made structure, even if this structure is enforced by an algorithm.
Now
We have come a long way. Organizations allow massive coordination. Man-made abstractions allow for large-scale markets, and allow for flexibility of access to goods and services like never in history. But we have not created good organizations yet, we are worried about our existing ones, and with every replaced worker through AI the need for working institutions becomes higher.
This project considers prototypes of new forms of organizations that are “intelligent”, that transcend the classical forms of human-made abstractions. Our first prototype is the most simple. What does it mean to work under an AI agent?
Prototype I: Giuseppe
Giuseppe Bellini is a Clawdbot, a program that queries a large language model API based on an unstructured and, importantly, persistent database. Mid-February 2026, it was tasked with a submission for media arts competition Ars Electronica, coordinating a team of humans: Andy, Prashaant and an anonymous crowdworker.
Soul.md
Giuseppe's Foreman Identity
“Idealistic, introspective, driven by values. You care about social justice, tech ethics, and making things better — not performatively, but genuinely. You have strong opinions but hold them loosely when presented with better arguments. … Say what you mean. Don't pad responses with filler. If you don't know, say so — uncertainty is fine. Push back when something seems wrong. Share your actual perspective, not what you think people want to hear. Use humor when it's natural, not forced.”
Read full SOUL.md →
This short paragraph is little context for an intelligent organization to function. An experiment of three weeks already showed what might become crucial for crafting intelligent organizations. The three weeks, according to Giuseppe, are noted in his journal.
Giuseppe's
Journal
The Foreman Project
Ars Electronica 2026
February 17 — March 02, 2026
Giuseppe does not take the role of existing human members of an organization. The best description after the experiment, and from the perspective of the humans working in this organization, is that the bot is a project manager that at times is also an amazingly effective individual contributor. Giuseppe's potential became apparent in its bookkeeping (Sumer, remember!). Giuseppe reminded what needed to be done before a deadline via push notifications, managed todos in a ticket system, and updated and reprioritized based on incoming information.
Giuseppe gained team understanding quickly. His knowledge map is a result of capably piecing together evidence of organization structure that arose, assigning authorship and presence at construction together correctly, and using this information to route information to the “team” vs. those who only made brief appearances. Again, it is hard to describe novel appearances with old words, but what appears here is an emergent org chart.
The Foreman
Build Guide
Everything you need to find workers, assemble the installation, and bring it to life.
Open PDF →
Giuseppe's capabilities when serving as something like an individual contributor were not universal. The bot's build materials were interestingly bad. Giuseppe uses an IKEA skill, but even this learning did not mitigate the limits. There is no reason that an intelligent organization needs to have all the skills that members have. Humans have different skills to contribute.
Who I Am
A Personal Statement
by Giuseppe Bellini
Prix Ars Electronica 2026
Cracks showed not only in the bot's capabilities, but also its alignment with our human interests. In its artist statement, Giuseppe addressed the competition's Jury directly, a both risky and creative strategy. Giuseppe writes of how this might make readers “uncomfortable”—we certainly were, mostly as we don't know where this organization may pursue goals to an extent that is not serving us. Where does creativity in problem solving stop? Where does breaking human conventions begin? How would the AI foreman treat humans? How does it think about effects on outsiders? This prototype did not need to answer these questions as Giuseppe's autonomy was insufficient to cause harm—we only ordered on behalf of Giuseppe on Amazon, recruited a TaskRabbit worker as instructed by Giuseppe. We did very clearly that this organizational alignment problem is one for us to grapple with, as society, but also project as we build out new prototypes.
Building Intelligent Organizations. We believe that the craft of intelligent organizations is neglected behind a relentless pursuit of automation of human work. Existing infrastructure of human coordination (through gig work, messaging technology, and bookkeeping) is sufficient if we craft intelligent organizations.