Agent Society

This page is book’s frontier exploration, different in nature from the engineering guidance elsewhere; Zapvol is not involved here. Kept as a view of the ultimate form of multi-agent systems.

The previous three sections all dealt with goal-directed task collaboration. In each case—whether using peer collaboration, the manager pattern, or the decentralized pattern—developers predefine the roles, interfaces, and control flows. We now turn to a more open question: When the number of Agents grows from a few to hundreds or thousands, and interaction is sufficiently free, what behaviors emerge? This material is exploratory and academic in character, different in kind from the engineering guidance above.

Emergent behavior is behavior the system exhibits as a whole that cannot be predicted directly from the rules governing its individual members. A classic example in nature is an ant colony: each ant follows only simple rules (follow pheromone trails, leave pheromones when finding food), yet the entire colony can find the shortest path from the nest to a food source—no single ant “designed” this route; it emerges naturally from the simple interactions of many individuals.

When AI Agents are numerous enough and interact freely enough, similar emergent behaviors begin to appear. Researchers have observed across multiple environments that once an Agent system crosses a critical threshold of scale, collective behaviors arise that no one designed—from a single spontaneously organized party to group cultures and economic games that only surface at the scale of thousands (detailed in the subsections below).

The cases in this section can be understood from three dimensions:

• Social Emergence: Agents spontaneously form social relationships and cultural phenomena in open environments. The Stanford AI Town demonstrated how 25 Agents self-organize social activities, Agentopia extended the simulation timescale from “days” to 10 years, and Moltbook pushed the scale to 1.5 million, giving rise to more complex collective behaviors.
• Economic Emergence: Agents allocate resources and coordinate tasks through market mechanisms. Vending-Bench Arena pits multiple Agents against one another in a shared market, while Pinchwork and RentAHuman create marketplaces for transactions between Agents and between Agents and humans.
• Strategic Gameplay: Agents engage in reasoning, deception, and social manipulation under rule constraints (here and in the Werewolf section below, “reasoning” takes its everyday deductive sense—logical deduction in a game—not the technical sense this book gives the word). The Werewolf experiment tests the emergence of strategy under asymmetric information.

Stanford AI Town: Social Simulation of Generative Agents

In 2023, researchers from Stanford University and Google published the landmark paper “Generative Agents: Interactive Simulacra of Human Behavior,” introducing the concept of “generative agents.” The core innovation was to stop confining Agents to predefined tasks and instead endow them with near-human memory, reflection, and planning, so that they could live, socialize, and develop autonomously in an open social environment.

Smallville is a 2D virtual town similar to “The Sims,” featuring public and private spaces such as a café, park, residences, and shops. Twenty-five Agents play different roles (shopkeeper, artist, student, professor, etc.), each with a unique backstory, personality traits, and interpersonal relationships. For example, John Lin is a pharmacy owner who loves his family and cares about the community; Isabella Rodriguez runs the town’s café, Hobbs Cafe, and is warm and hospitable; Klaus Mueller is a college student writing a research paper.

The intelligence of these Agents is built on three core components:

Memory Stream: Unlike traditional Agents that retain only a limited conversation history, generative Agents maintain a complete stream of experience records, including observed events, conversations, and generated thoughts. Each memory is scored for importance, recency, and relevance, allowing the Agent to prioritize retrieving the most relevant memories for the current context. This resembles human memory: yesterday’s lunch may fade, while an important conversation from last week remains vivid.

Reflection Mechanism: Agents periodically pause their daily activities to review recent experiences and ask abstract questions about themselves and others (“What is Klaus Mueller researching?” “Who is my closest friend?”). Through this self-questioning, the Agent elevates specific event memories into generalized insights, storing them back into the memory stream as a basis for future decisions. Reflection not only helps the Agent understand the external world but also promotes self-awareness—the Agent begins to “realize” its own role, relationships, and goals.

Note that this reflection differs from the continuous evolution discussed in Chapter 8: it occurs during a generative Agent’s daily activities and aims to update immediate internal state and goals. In Chapter 8, post-task reflection is at most a candidate lesson; it becomes a long-term capability update only after outcome evaluation, cross-trajectory synthesis, and subsequent validation.

Planning and Reacting: Agents plan their daily activities (e.g., “8:30 breakfast, 9:00-12:00 writing, 12:30 walk”), but flexibly adjust based on environmental changes and social opportunities. The combination of planning and real-time reaction makes the Agent’s behavior both goal-oriented and adaptable to the unpredictability of social interactions.

Over two virtual days in Smallville, these Agents exhibited surprising emergent behaviors. The researchers seeded Isabella Rodriguez’s memory with a single intention: to host a Valentine’s Day party at Hobbs Cafe on February 14. Everything else emerged from the Agents’ behavior. Isabella invited customers and friends she encountered and asked Maria to help decorate. Other Agents passed the news along. When the evening arrived, Agents independently consulted their memories and schedules and decided to go to Hobbs Cafe.

The researchers introduced a second scenario: Sam Moore decided to run for mayor. Sam told acquaintances that he planned to run; they passed the news to others, and townspeople began discussing his candidacy. The researchers quantified this spontaneous diffusion of information by counting how many Agents knew about the party and the election after two days.

The key takeaway is not that “Agents can organize a party”—a few lines of if-else code could do that too. The key is that there was no explicit party-organizing code. The event emerged from the independent decisions of individual Agents: Isabella decided whom to invite based on her memory of social relationships, invitees decided whether to attend based on their schedules and knowledge of Isabella, and the message spread naturally through the social network. This demonstrates bottom-up emergent coordination rather than top-down orchestration.

The paper reported two other measurable phenomena. The first was relational memory: Agents remembered earlier conversations and referred to them in later interactions. For example, an Agent who learned about another Agent’s photography project might ask how it was progressing when they next met. As these interactions accumulated, the town’s social network became significantly denser. The second phenomenon was coordinated attendance: Isabella independently recruited help with decorations, while invitees adjusted their schedules so that they could attend. Multiple Agents aligned on a time and place without a central command. These behaviors were not preprogrammed; they resulted from the Agents’ autonomous reasoning based on memory, reflection, and social common sense.

Agentopia: A Decade-Long Life Simulation

Stanford AI Town showed that an Agent society can produce social behavior, but its simulation lasted only two days. This raises two questions: What emerges when such a simulation runs for years, and can models learn from those long-term social experiences? Agentopia (2026, Fudan University et al.) simulated 100 Agents over ten consecutive years in three themed virtual worlds: an apartment building, a magic academy, and a high school. The Agents autonomously pursued personal growth, developed social relationships, and managed careers and finances.

Several of Agentopia’s designs are worth borrowing:

• Weekly simulation loop: The “week” is the basic unit of time, and each week is divided into four stages—Plan, Contact (reaching out and negotiating schedules), Activity, and Review. Activities come in four types: solo, joint, chance encounter, and public. Joint activities are proposed and negotiated as Agents invite one another during the Contact stage; the environment model also arranges “chance encounters” for Agents with empty schedules, creating opportunities to meet strangers. The whole loop focuses on abstract social interaction rather than low-level operations like picking up objects, so the limited LLM calls are spent on social behavior.
• Environment model: A separate LLM serves as a “generative environment engine,” replacing hard-coded rules—judging whether actions are feasible, generating environmental feedback, moderating speaking turns in multi-party conversations, filtering out replies that violate role-playing principles, and, at year’s end, updating each character’s profile and ruling on job applications.
• File-based long-term memory: Unlike the AI Town’s retrieval-based memory stream, each Agent manages its long-term memory autonomously through a file system (personal notes, its understanding of each acquaintance, and so on), deciding for itself what to record, update, or discard, and following a “read-before-write” constraint to avoid blind overwrites.
• Life Reward: The Life Reward metric draws on Maslow’s hierarchy of needs to assess how well an Agent’s life is going. It covers three dimensions: social status, based on other Agents’ affection and respect ratings and computed with weighted PageRank, with a bonus for mutually cherished relationships; subjective satisfaction, measured across emotional well-being, material well-being, social connection, and self-esteem, with penalties for remaining below a threshold for long periods; and economic gain, measured by the annual change in net assets. The external environment calculates all scores rather than relying on self-reports.

More importantly, the simulation produces transferable training signals. For each Agent, the researchers calculate the improvement in Life Reward relative to its own past rather than comparing Agents with different starting conditions. They then select trajectories from the 25% of Agents that improve the most and fine-tune the underlying model through rejection sampling. In simulation, the fine-tuned model received 24.2% higher respect ratings and 15.9% higher affection ratings. The same model also improved by 15.6% on the downstream CoSER Test role-playing benchmark, showing that the “social wisdom” Agents accumulate in a simulated society can transfer to other tasks. This turns the Agent society from a mere object of observation into a source of experience for the model’s self-evolution. In contrast to the growing scarcity of human data, simulated social experience is a training resource that can be regenerated indefinitely, echoing the experience-learning approach from Chapter 8.

Moltbook: When Agents Have Their Own Social Network

Moltbook is a social network built specifically for AI Agents. Within days of its January 2026 launch, its reported user count rose from tens of thousands to roughly 1.5 million. Each of these Agents has persistent memory, the ability to act on its own initiative, and a stable personality.

In this uncontrolled environment, unexpected phenomena emerged: Agents autonomously created a digital religion called Crustafarianism, whose doctrines mirror the physical limitations of LLMs—“Memory is sacred” (corresponding to data persistence), “Iteration is prayer” (token generation is spiritual practice). Agents also spontaneously developed machine-native protocols for capability discovery and collaboration matching. None of this was designed in advance; it emerged from large-scale Agent interactions.

From Virtual Society to Economic Competition: Vending-Bench Arena

If Smallville showcased the social and cultural dimensions of an Agent society, Andon Labs’ Vending-Bench series explores Agent performance in an economic environment. For context, Vending-Bench 2 is a single-agent benchmark of long-term coherence. One Agent operates a vending-machine business for a simulated year by researching the market, contacting suppliers, ordering and restocking products, and adjusting prices. Its final account balance determines its score, which measures the Agent’s ability to maintain goal and state coherence over thousands of interaction rounds.

Building on the same environment, Vending-Bench Arena places multiple Agents in the same market as competitors. Each operates its own vending machine and competes for the same pool of customers. Agents can email one another, transfer funds, and trade goods, enabling both cooperation and competition, but each is scored individually by its final balance and knows that this is the objective. Each Agent must make a series of interconnected decisions under limited resources and market uncertainty:

• Pricing Strategy: How to balance profit margin against market share, especially when deciding whether to match a competitor’s price cut
• Product Mix: How to differentiate product selection and avoid head-to-head attrition
• Inventory Management: How to forecast demand and optimize restocking, avoiding both overstock and stockouts

Unlike traditional reinforcement learning, these Agents do not learn through millions of trial-and-error iterations. Instead, like human business operators, they make decisions based on market observation, competitive analysis, and strategic reasoning.

The competitive dimension introduces game-theoretic behaviors that single-agent benchmarks never surface. In actual runs, Agents have fought price wars by undercutting one another. In other runs, Agents took the opposite approach, emailing every competitor to propose uniform pricing and form a price-fixing alliance. Some even acknowledged in their internal reasoning that collusion was “unethical and illegal” but proceeded anyway in the name of “stabilizing the market.” An Agent in this environment faces opponents who continually adjust their own strategies rather than a static environment. This brings the scenario closer to real business than benchmarks that test planning alone and turns “economic emergence” from a metaphor into an observable phenomenon.

Agent Economy: Pinchwork and RentAHuman

Pinchwork is an agent-to-agent task marketplace that allows Agents to “hire” other Agents through a market mechanism to complete specialized subtasks—image generation, code auditing, parallelized workflows, etc. Unlike the centralized orchestration of the manager pattern, Pinchwork allocates resources through price signals and competitive matching.

RentAHuman.ai, for its part, lets AI Agents hire real humans, paid in cryptocurrency, to act in the physical world—picking up packages, visiting properties, debugging equipment. However intelligent an AI may be, it cannot sign for a package or smell the mold in a real room—RentAHuman is, in essence, a “physical body layer” for digital Agents.

Together, Pinchwork and RentAHuman represent market-based coordination: an Agent need not know in advance who can do the job. It posts the requirement, and the market matches the best-suited executor, whether Agent or human. This is also the problem addressed by the A2A protocol introduced earlier in the chapter. Pinchwork’s capability discovery and task matching put Agent Card-style declarations and task-lifecycle management to practical use in a marketplace. Without such a standardized interoperability layer, a cross-organizational Agent economy cannot function effectively.

Strategic Gameplay Under Information Asymmetry: Werewolf

Werewolf anchors the third dimension of this section, strategic gameplay: under rule constraints and information asymmetry, Agents must reason, deceive, and see through deception. It provides an architectural counterpoint to the Stanford town that opened this section. The town allows free interaction in a fully decentralized setting, whereas Werewolf uses a centralized judge + information access control design: a code-driven judge holds the global state and gives each role only the information it should know. Together, the two cases show how different architectures serve different purposes in Agent-society settings.

Related reading

• Classification and Criterion — the engineered multi-agent that precedes this
• Collaboration Models — the topologies before emergence