Building a Company as an Intelligence Requires a "Brain"
May 21, 2026 · View on GitHub
From hierarchy to intelligence, what a company needs most is not a larger model, but a brain that prevents amnesia and enables continuous growth.
In From Hierarchy to Intelligence, co-published with Sequoia, Jack Dorsey posits a proposition worth taking seriously: the truly competitive companies of the future will not merely be "companies that use AI", but rather "a company built as an intelligence" (operating as an agent).
This statement is profound not because it sounds novel, but because it exposes a long-ignored reality: a company's operational efficiency ultimately depends on how information flows, how understanding is formed, and how experience is accumulated. In the past, this work was primarily carried out by managers within a hierarchical organization; in the future, an increasing share of this work will be handled by the system itself.
The true value of Jack's article lies in how clearly it frames the problem. Hierarchical organizations exist largely because humans must rely on cascading layers of reporting, communication, and coordination to keep a complex organization running. However, when AI can continuously maintain a company's "world model", the organization has the opportunity to shift from "relying on hierarchical messaging" to "relying on systemic understanding".
The question is: what does the system rely on to understand a company?
The answer is not a larger model, but better memory.
When many people discuss AI, their immediate focus is on reasoning, generation, and automation. But for an organization that truly wants to become "a company built as an intelligence", reasoning is not the first thing it lacks. What is genuinely missing first is a memory system capable of consolidating experiences, as well as continuously organizing and updating them.
Without such memory, even the smartest models will act like they are starting their first day on the job every single morning. They can answer questions, but they struggle to truly "live" within the company. They can analyze immediate information, but they cannot inherit the organization's judgments, relationships, and evolutionary processes from the past months or years. More realistically, without long-term memory, the massive amounts of information generated by different teams, agents, and systems within the company will ultimately scatter into new, isolated silos.
This is exactly where Anda Brain brings value.
In the human brain, the brain is responsible for turning experiences into memories, consolidating short-term events into long-term cognition. It is not the most conspicuous part of the brain, but without it, even the most powerful brain suffers from amnesia. A person can continue to think but cannot truly learn from yesterday. A company is the same: it can continuously deploy new AIs, but it cannot consolidate countless interactions, decisions, and feedback into a continuously growing organizational understanding.
What Anda Brain does is provide precisely such a "brain" for AI and enterprise systems.
It is not reinventing a document repository, nor is it just stacking another vector database. It transforms the fragmented information generated in the company's daily operations into a continuously growing cognitive graph. What a customer has said, how a project is progressing, which team depends on another, which judgments have been overturned by new facts, and what seemingly scattered changes are converging into a larger trend—these will no longer be mere fragments scattered across meeting notes, chat logs, ticketing systems, and analytics dashboards. Instead, they will be organized into a connectable, traceable, and evolvable knowledge graph.
This may sound like a technical issue, but it is fundamentally a management issue.
Why do traditional organizations need so many layers of reporting, syncing, and coordination? Because information is naturally dispersed, and human memory and attention are inherently limited. Many management actions are essentially functioning as manual information routers. Whoever knows more is responsible for summarizing; whoever sits higher up is responsible for transmitting. Once an organization grows large, this mechanism inevitably becomes slow, distorted, and backlogged.
What Jack aims to drive in From Hierarchy to Intelligence is replacing this inefficient information flow with a system, equipping the company itself with the capacity for continuous perception and understanding. Anda Brain aligns perfectly with this direction because it solves not "how to store more information", but "how to ensure information is digested".
Storing and digesting may seem similar, but they are entirely different.
Storing is simply writing things down. Digesting means merging duplicate information, retiring outdated judgments, highlighting conflicting cognitions, and distilling scattered events into higher-level patterns. Just as human sleep is not merely rest—the brain actively organizes the day's experiences during sleep—the key value of Brain is that it is not a passive warehouse, but an active system that continuously organizes memories in the background.
To make information "digestible", however, the choice of data structure is paramount. In a vector database, two pieces of information are merely two isolated dots; you cannot label "this is outdated", nor can you express "there is a causal relationship between these two events". Markdown files become harder to maintain as they grow longer. Key-value stores only keep the latest value, erasing historical trajectories entirely. Only a network structure like a knowledge graph naturally supports relationship tracking, contradiction detection, timeline evolution, and topic merging—the exact fundamental operations required for "digestion". Brain is built fundamentally on knowledge graphs and, drawing inspiration from the human brain's sleep mechanism, automatically executes the organization, consolidation, and evolution of memory in the background.
This means that a company no longer merely "accumulates more and more data", but begins to accumulate increasingly mature cognition.
What is truly scarce for an enterprise has never been just the data itself, but the relationships and histories between data points. A dashboard can tell you that conversion rates dropped this week, but it won't automatically tell you that this might be connected to a process change launched three weeks ago, a recent spike in a specific type of customer complaint, friction in sales scripts reported by the frontline, and a behavioral shift in a high-value customer segment. Single signals are not scarce; what is scarce is the ability to connect these signals into actionable understanding.
To use a more accessible analogy: a company's existing ERP, CRM, ticketing systems, and data warehouses are essentially recording "what happened"; whereas Brain helps the company formulate "what this means".
This is also why it is particularly suited for the next-generation organizational model Jack describes.
In From Hierarchy to Intelligence, Jack identifies three core roles: Individual Contributors (ICs) working deeply within various layers of the system; Directly Responsible Individuals (DRIs) solving specific problems cross-functionally; and player-coaches balancing hands-on execution with mentoring. These three roles share a common prerequisite: the system must proactively present a contextual understanding that spans across departments, time, customers, and events to every person and every agent. There should be no need to rely on others to sync backgrounds, no need to spend hours asking for context, and no need to act as an information relay station. When a new AI agent is plugged in, it shouldn't need to "retrain" on the whole company; when a new DRI takes over, they shouldn't have to spend weeks digesting old materials.
What Anda Brain provides is exactly this shared cognitive foundation.
Furthermore, this foundation carries a highly practical advantage: it does not lock a company's long-term memory into any single AI model. You might use one model today and switch to a more powerful one tomorrow, but the customer insights, organizational experience, decision trajectories, and historical lessons that truly belong to the company shouldn't be wiped out in the process. Models will iterate, and tools will change, but the cognitive assets a company has accumulated over the years should be continuously preserved and enriched. The true moat of the future is not just "whether you can call the most powerful model", but "whether you possess your own continuously growing organizational memory".
The company structure outlined in Jack's article consists of four layers: capabilities (atomic abilities), interfaces (delivery touchpoints), the intelligence layer (combinational abilities and proactive decision-making), and the world model (the foundation for the company’s understanding of itself and its customers). Anda Brain corresponds precisely to this world model layer—or more accurately, the "company world model" component: enabling the organization to maintain a continuously updated understanding of its own operations, historical experiences, and cross-departmental dynamics. It might not be the most visible layer at center stage, but it is likely the layer that first determines the system's absolute upper limit.
Without it, a company can certainly deploy numerous AIs and automate many tasks. But that is more like strapping a new engine onto an old organization. The true transformation happens when the company, as a whole entity, begins to possess continuous memory, evolvable understanding, and a shared world model. At that point, AI is no longer just "helping employees do tasks", but actively helping the company form cognition.
For companies like Block, this is particularly worth exploring. Jack notes in his article that Block possesses extremely rich customer transaction signals, alongside massive digital footprints naturally generated by a remote-first organization. For a company like this, data is the least scarce resource; what is most worth building is the memory infrastructure capable of transforming that data into organizational cognition. It is not about adding another assistant, but adding a brain; not about making the system better at answering, but making the system truly begin to remember and grow.
If From Hierarchy to Intelligence explains why companies should transition from hierarchies to intelligence, then what Anda Brain answers is the most critical question for the very next step:
When a company truly begins to operate like an intelligent agent, where should its memory reside, and how should it grow?
Anda Brain is an open-source AI memory engine that provides evolvable long-term memory capabilities for AI agents and enterprise systems based on knowledge graphs.
Extended Reading: