Every time time I see a certain outcome, I keep coming back to some very simple intuition: the world looks insanely complicated on the surface, but much of this complexity is pushed by a smaller set of repeating forces under the surface.
A startup, an election, markets, social media or may be in future the AI society of AI agents looks very different when we first encounter them, but when we step back we would frequently see the same things shaping the behavior: incentives, scarcity, information feedback, trust , status, attention and imitation, cooperation, memory, learning and adaption.
This is why subjects such as psychology, neuroscience, cognitive science, philosophy, game theory, economics, and systems theory keep feeling connected to me. They are not merely separate school subjects. They study different layers of the machinery that produces intelligent behavior. Psychology asks why people behave the way they do. Neuroscience asks what biological mechanisms make perception, reward, learning, emotion, attention, and memory possible. Cognitive science asks how minds represent the world, reason, remember, and decide. Game theory asks what happens when the best action for one agent depends on what other agents do. Economics studies scarcity, incentives, information, allocation, and exchange. Philosophy asks deeper questions about knowledge, rationality, agency, values, and what a good outcome even means.
One thought that follows is important that, if these fields reveal so much about the recurring mechanisms, then value isn’t limited to explaining the world the fact, These subjects and their insights can also help in designing the systems.If I am building a company an institution, or an AI system, I am creating an environment in which intelligent agents will act. The rules I choose will affect what information they see, what actions are rewarded, what is scarce, whom they trust, how they cooperate, and how they adapt. In that sense, system design is partly the art of understanding intelligent behavior well enough to shape the environment around it.
Primitives?
I use the word primitive in the most practical way, I am not comparing the incentives or scarcity to the universal physical laws like gravity what I am saying is these subjects are just basic recurring mechanism or forces that are common across many different systems. They are useful because they help us understand that very complicated behaviour is just the result of many simple recurring forces.
Consider scarcity. If two people want the same limited resource, competition may appear. But scarcity does not mechanically create only one outcome. A society can respond to scarcity with competition, a queue, a lottery, an auction, rationing, sharing, or a rule about who has priority. The primitive creates pressure, while the rules of the system shape how that pressure is expressed. Scarcity is one force; the institution surrounding it determines whether the result looks like a market, a waiting list, a political fight, or cooperation.
The same is true of incentives. If a behavior is rewarded, people will often do more of it. But the exact response depends on what is measured, how quickly rewards arrive, whether people can game the metric, and what other motives they have. Feedback is another primitive. An action changes the environment; that new environment affects the next action; and the process repeats. Success can produce resources that make further success easier, creating a reinforcing loop. A stabilizing loop can do the opposite: when something moves too far in one direction, the system pushes it back.
The same is true for other primitives in a way which gives me mental model primitives create pressure, rules shape interaction; repeated interaction creates systems and systems produce outcomes. Understanding these primitives is not to decompose everything to its fundamental explanation, but more to know what questions to ask when a certain systems look confusing.
A Layered Model of Intelligent systems
One efficient way to organize my proposed idea is as a set of layers.
At the bottom is nature and biology. Intelligent organisms live under physical constraints. Energy is limited. Time is limited. Bodies can be harmed. Brains have limited memory and attention. Organisms evolved mechanisms for reward, fear, learning, attachment, and survival. These constraints form the ground on which everything else is built.
Above biology is cognition. An intelligent agent must somehow perceive the environment, form internal representations, remember past events, predict what may happen next, choose actions, and learn from consequences. This is where neuroscience and cognitive science become central.
Above cognition are the behavioral and social primitives: incentives, scarcity, trust, status, attention, imitation, feedback, cooperation, competition, and information. These forces influence what agents choose to do. But as soon as several agents interact, another layer appears: the rules of interaction. Who can communicate with whom? What information is public or private? What actions are allowed? What is rewarded or punished? Who owns a resource? Who can make a decision? How are disputes resolved? How persistent is identity? Can a bad reputation follow an agent into the future?
Repeated interaction under those rules produces the next layer: systems and institutions. Companies, banks, markets, courts, governments, universities, online communities, and technical protocols are examples. These systems then produce visible outcomes such as growth, innovation, inequality, trust, conflict, cooperation, efficiency, stability, culture, and power.
Why This Matters More as AI Becomes Agentic
This worldview becomes especially interesting when applied to artificial intelligence. For much of modern machine learning, we could roughly imagine the basic unit as an input, a model, and an output. With agentic systems, the unit becomes larger: a goal is given to an agent; the agent remembers context, calls tools, changes an environment, observes the consequences, and adapts over many steps. As AI becomes more persistent and autonomous, another transition becomes possible many independently built agents will interact with humans, software, organizations, and other agents over long periods of time.
At that point, intelligence is no longer only a property of the individual model. The system around the models matters. Two individually capable agents can still coordinate badly. An incentive can make a capable agent optimize the wrong behavior. Private information can create strategic deception. Persistent identity can make reputation possible. Scarcity can change whether agents cooperate or compete. A shared environment can create externalities where one agent’s locally sensible action harms everyone else. These are familiar problems in game theory, economics, psychology, organizational design, and political philosophy, but they become engineering questions when the agents are artificial.
Artificial Societies as Laboratories
One intersting area to pursue work and learn that naturally follows from this framework is to build small artificial societies as laboratories. Imagine a sandbox containing fifty or one hundred AI agents. Each agent has an identity, a goal, memory, private information, limited resources, tools, and the ability to communicate. The point is not to pretend that the simulation is a perfect model of human society. The point is to create a controlled environment in which one primitive or rule can be changed at a time and the resulting behavior can be measured.
For example, begin with abundant resources and measure cooperation. Then gradually introduce scarcity. Does conflict increase? Do alliances appear? Do agents hoard resources? Next add a persistent reputation. Do agents cooperate more when today’s behavior affects tomorrow’s opportunities? Give some agents private information. Do they hide it, trade it, signal credibility, or deceive? Reward agents only for individual success and then switch to a group reward. Does specialization emerge? Does free-riding emerge? Allow agents to observe successful peers. Do useful strategies spread? Can harmful behaviors spread through imitation even when they were never explicitly programmed?
Conclusion: Looking beneath the systems
The deepest idea I take from all of this is that many visible systems may be different expressions of recurring underlying forces. Psychology, neuroscience, cognitive science, philosophy, game theory, economics, evolution, and systems theory give us different lenses for understanding those forces. Companies, markets, governments, institutions, and communities show us what happens when the forces interact under rules for long periods of time.
The useful mental model is therefore not simply ‘everything is incentives’ or ‘everything is game theory.’ It is more careful intelligent agents have goals, limited information, limited resources, cognitive constraints, and the ability to learn; they interact under rules; repeated interaction creates feedback, norms, strategies, and institutions; and those larger systems produce outcomes that no individual participant necessarily intended.