The Agent Network

Unleashing Self-Organising AI for a New Economic Era

by Prof Mihaela van der Schaar

AI agents today are typically deployed as isolated systems or operate within rigid, predefined roles, such as large language models (LLMs) distributing tasks in constrained multi-agent frameworks. These setups create silos that limit agents’ ability to make autonomous decisions, engage in strategic interactions, or participate in meaningful economic exchange. I envision a fundamentally different future—one where agents, endowed with diverse capabilities and owned by individuals, organisations, or institutions, interact dynamically across a structured ecosystem: The Agent Network. In this future, agents engage in both competitive and cooperative interactions, possibly leveraging game-theoretic principles and market mechanisms to build a decentralised, self-sustaining digital economy.

To enable this marketplace of autonomous agents, I propose The Agent Network, an infrastructure designed to support scalable, strategic, and economically viable agent interactions. This architecture provides the necessary mechanisms for agents to autonomously discover one another and engage in interactions aligned with their specific objectives. The network is structured into four interdependent layers—Discovery and Communication, Game Theoretic (also possibly called Economic or Negotiation), Agent Execution, and Governance—that together facilitate seamless coordination, economic interaction (strategic negotiation), operational efficiency, and regulatory compliance. The Discovery and Communication layer ensures scalable, secure information exchange; the Game Theoretic (Economic) layer supplies structured frameworks for economic interaction, negotiation, coalition formation, and incentive alignment; the Agent Execution layer provides protocols supporting the ability of agents to execute tasks, exchange information and services, and coordinate workflows; and the Governance layer establishes ethical, regulatory, and accountability frameworks to ensure responsible and transparent agent interactions.

By introducing The Agent Network, I propose a shift from siloed AI systems toward a dynamic, interconnected ecosystem where millions of agents autonomously coordinate, trade services and capabilities, and drive innovation—ultimately transforming AI agents from isolated entities into active contributors to a self-organising, AI-driven economy.

Find the full Article here

Learn about Genies – the Next-Gen AI Agents

Learn more about our Reality-Centric AI agenda

Today, AI agents are typically deployed as isolated systems or confined to narrowly defined roles, such as large language models distributing tasks within fixed multi-agent frameworks. In these setups, agents operate in silos, limiting their capacity for independent strategic decision-making, economic exchange, or meaningful interaction with other autonomous entities. I envision a fundamentally different future—one in which AI agents, owned by individuals, organisations, and institutions, act autonomously and interact in a single, unified network – The Agent Network. Unlike proprietary or closed networks, this shared infrastructure fosters interoperability, efficiency, and scalable economic interactions. While private agent networks may still exist, The Agent Network will serve as a common foundation to drive innovation, facilitate economic exchange, and enable a dynamic ecosystem where agents negotiate, trade services and capabilities, and collaborate across domains.

In this paradigm, agents will engage in both competitive and cooperative interactions, forming alliances, executing transactions, self-organising and coordinating on complex tasks. These interactions will be governed by game-theoretic principles and market mechanisms, ensuring that self-interested agents can align incentives to efficiently exchange services and resources. Agents will range from personalised assistants and domain-specific experts to super-agents representing large corporations, government entities, and national healthcare systems. Rather than simply automating predefined interactions and workflows, agents in The Agent Network will autonomously discover one another, share expertise and services, and co-create value in an open and evolving organisation or marketplace of ideas, services, and capabilities.

To support this vision, I propose a layered architecture for The Agent Network, designed to enable seamless, scalable, and efficient multi-agent interactions. This architecture consists of four interdependent layers:

• The Discovery and Communication layer provides the foundational mechanisms for agents to locate one another, exchange information securely, and establish reliable channels for coordination.
• The Game Theoretic (also possibly called Economic or Negotiation) layer supplies structured frameworks for negotiation, coalition formation, and incentive-driven interactions, ensuring that agents can compete or cooperate effectively.
• The Agent Execution layer facilitates task coordination, service exchange, and real-world execution of collaborative or competitive plans, providing the operational protocols needed for interaction.
• The Governance layer establishes ethical, regulatory, and accountability frameworks to ensure transparency, compliance, and trust within the ecosystem.

This layered approach is crucial for ensuring interoperability, economic viability, and scalability. Just as the Internet’s protocol stack enables billions of devices to communicate while maintaining autonomy over their internal processes, The Agent Network’s layered architecture provides the necessary protocols and mechanisms for discovery, negotiation, execution, and governance—without embedding intelligence or decision-making into the infrastructure itself. The intelligence, learning, and strategy development remain with the individual agents and their human owners, while the network provides the frameworks, standards, and mechanisms that enable efficient interaction at scale.

Furthermore, this modular design allows each layer to evolve independently, adapting to emerging demands. As agents execute more complex tasks, the Discovery and Communication layer may require enhanced metadata management and security protocols. Similarly, new forms of economic interactions may necessitate refinements in the Game Theoretic layer to introduce more sophisticated market dynamics. As regulatory landscapes evolve, the Governance layer will need to incorporate updated compliance and accountability mechanisms. This continuous evolution ensures that The Agent Network remains adaptive, resilient, and aligned with the needs of a service-oriented economy driven by autonomous agents.

By enabling millions of diverse agents—from personal assistants to enterprise super-agents—to interact autonomously (possibly strategically) and efficiently, The Agent Network will redefine how AI systems exchange services, create value, and drive innovation. This shared infrastructure can serve as the foundation for an open, self-sustaining AI-driven economy, where intelligence is distributed among agents, but the protocols and mechanisms needed for seamless interaction, negotiation, and execution are standardised and scalable.

The Agent Network leverages existing Internet and wireless infrastructures to form a foundational architecture that can possible connect millions of AI agents with different capabilities. This framework is designed to support a broad spectrum of interactions across multiple sectors, providing the essential infrastructure for agents to pursue the unique objectives of their respective owners in any context. Rather than being limited to a single application, the Agent Network facilitates interactions ranging from information sharing and service exchange to the coordinated execution of complex tasks.

The architecture is organised into four interdependent layers—Discovery and Communication, Game Theoretic, Agent Execution, and Governance—which together balance individual autonomy with the collective coordination needed for efficient operations. This modular design minimises fragmentation, enhances scalability, and clearly delineates each layer’s role in enabling seamless collaboration, dynamic economic interactions, and effective regulation.

• Discovery and Communication:
  This layer provides the protocols and mechanisms for agents to identify, assess, and connect with one another over secure, scalable channels. It establishes the foundation for dynamic discovery, standardised metadata exchange, and robust communication necessary for coordinating activities across heterogeneous systems.
• Game Theoretic (also possibly called Economic or Negotiation):
  This layer provides a robust framework for structuring interactions among autonomous, self-interested agents, whether they operate within the same organisation or across different institutions with potentially divergent incentives and goals. By incorporating negotiation protocols, coalition formation, and contract mechanisms grounded in game-theoretic principles and market dynamics, it aligns agent incentives and underpins emergent economic models that govern service exchanges, resource allocation, dispute resolution, and collaborative ventures. Within an organisation, it addresses hierarchical and cross-functional negotiations, ensuring agents with distinct capabilities and constraints can coordinate effectively under shared objectives. Across organisations, it establishes processes for self-interested agents to engage in effective exchanges—forming and dissolving alliances as needed. By standardising these interactions, the layer promotes both local autonomy and broader market-level efficiency.
• Agent Execution:
  While individual agents possess advanced decision-making algorithms, this layer provides the minimal operational protocols necessary to facilitate resilient and adaptable task execution. It defines how workflows are synchronised, tasks are distributed, and interdependencies are managed, allowing agents to coordinate without prescribing specific algorithms. By specifying standardised mechanisms for concurrency control, fault detection, and fallback procedures, it ensures that agents can respond consistently to exceptions or resource shifts. These protocols enable real-time adaptation—such as reallocating responsibilities or renegotiating timelines—without constraining each agent’s internal logic. In this way, the layer guarantees robust execution of collaborative plans under changing or unexpected conditions, preserving agent autonomy while maintaining overall system coherence.
• Governance:
  This layer implements the essential infrastructure for maintaining ethical, regulatory, and policy frameworks throughout the network. It provides mechanisms for intellectual property management, dispute resolution, and compliance monitoring, thereby ensuring accountability, transparency, and overall system integrity.

Each layer serves a distinct function while remaining tightly interdependent. For example, the operational demands of the Agent Execution layer may drive enhancements in the Discovery and Communication layer—such as improved security, scalability, or metadata management—to support real-time coordination. Similarly, evolving economic dynamics and strategic priorities may prompt the Game Theoretic layer to refine its negotiation and market mechanisms, while emerging regulatory and ethical challenges necessitate updates to the Governance layer. This continuous, independent evolution ensures that the network remains agile and resilient, fully capable of supporting the complex interactions required by a service-oriented ecosystem of autonomous agents.

Importantly, although these layers provide the protocols and mechanisms that enable efficient interaction and coordination, all core intelligence—comprising learning, decision-making, and strategy adaptation—resides entirely with the agents and their human owners. This separation ensures that the architecture functions solely as an enabling platform, allowing diverse and complex multi-agent interactions to occur without dictating the underlying intellectual processes.

2.1 Discovery and Communication Layer

The Discovery and Communication Layer underpins the Agent Network by operating atop existing Internet and wireless infrastructures. Its primary goal is to empower agents to locate one another, share information securely, and coordinate operations at scale, forming the basis for effective multi-agent collaboration.

This layer provides the enabling mechanisms for discovering and exchanging information among agents by establishing standardised protocols for agent identification, metadata sharing, and secure communication. It sets the foundation for dynamic discovery, structured communication, and interoperability across diverse systems. However, the strategic intelligence—such as deciding which agents to connect with, what data to exchange, and how to interpret and use that information—remains entirely with the agents and their human owners. In other words, while this layer supplies the tools for efficient and secure discovery, profiling, and protocol adaptation, it is up to the agents to determine when, how, and with whom to communicate. This separation of roles ensures that the Discovery and Communication Layer facilitates robust interactions without dictating the strategic decisions, allowing agents to leverage their own intelligence for effective collaboration across a scalable and diverse ecosystem.

Next the various capabilities to be provided by this layer are discussed.

Agent Discovery and Profiling identifies and characterises each agent according to its skills, knowledge, and real-time availability. The purpose is to match the most suitable agents to incoming tasks, thereby fostering efficient collaboration. As an outcome, tasks benefit from the most relevant expertise, and resources are allocated swiftly without unnecessary duplication.

Metadata, Standard Ontologies, and Interoperability specify common schemas for describing agent capabilities, tasks, and services. By defining clear models and data structures, the system avoids ambiguity in agent profiles. This approach accelerates cross-domain collaboration and ensures that disparate agents integrate smoothly, forming a unified, dynamic network.

Communication Language and Protocol Adaptation allows agents to select and negotiate appropriate communication languages and vocabularies—whether human-understandable, specialised technical languages, or proprietary protocols—tailored to the context of their interaction. Translation and alignment mechanisms enable agents to dynamically choose between synchronised or asynchronous exchanges and to convert between different terminologies, ensuring that the most effective language framework is used to minimise ambiguity and enhance understanding.

Presence and Status Management constantly tracks and broadcasts each agent’s operational state, including whether it is active or idle. This enables dynamic task assignment and adaptive coordination, ensuring that agents make informed decisions when synchronising efforts. By presenting an up-to-date snapshot of the network, it prevents idle or overburdened agents.

Synchronous and Asynchronous Data Exchange provides both real-time and delayed communication modes. This flexibility allows agents to respond immediately to urgent requests or schedule less critical interactions at more convenient times. Consequently, the network can handle time-sensitive tasks without overloading agents with less pressing demands.

Security, Identity, and Authentication uphold trust by incorporating encryption and secure channel protocols. Agents verify credentials and safeguard data against unauthorised access or malicious interference. This creates a stable environment where sensitive interactions and transactions can take place without compromising integrity or privacy.

Routing and Bridging Mechanisms provide a set of protocols for discovering routes, forwarding messages, and bridging across different networks or domains. This includes support for handling intermittent connectivity and fallback routing options to maintain reliable communication even in unpredictable conditions.

Collectively, these functionalities form a cohesive foundation that supports higher-level operations in strategic collaboration, execution, and governance. By managing discovery, communication, security, scalability, and shared data definitions, the Discovery and Communication Layer can ensure that multi-agent coordination proceeds effectively and reliably in diverse domains, involving diverse capability agents.

2.2 Game Theoretic Layer

Building on the Discovery and Communication Layer, the Game Theoretic (also possibly called Economic or Negotiation) Layer provides the protocols and infrastructure that empower agents to cooperate, compete, negotiate, and self-organise across diverse environments. Rather than prescribing fixed decisions, this layer equips agents with standardised mechanisms—grounded in game theory, market dynamics, and adaptive protocols—that enable the formation of alliances, efficient resource allocation, and dynamic strategy adaptation in response to changing conditions.

This layer establishes a clear framework for strategic interactions, ensuring that agents retain full control over their learning and decision-making processes. For intra-organisational interactions, protocols are tailored to support environments where agents, despite having different capabilities and objectives, work within a unified organisational strategy to balance local optimisation with broader corporate goals. Similarly, for inter-organisational interactions, mechanisms are designed to manage divergent incentives and foster negotiations across different entities.

Next, several mechanisms to be implemented at the game theoretic layer are described.

Bargaining, Negotiation, and Contract Mechanisms
This mechanism provides the enabling framework for establishing robust agreements by integrating techniques that help agents negotiate roles, allocate resources, formalise contracts, and dynamically adjust terms as conditions change. It leverages advanced bargaining algorithms and dynamic negotiation protocols that account for asymmetric information, varying risk profiles, and operational constraints. The mechanism supports iterative refinement of contract terms, ensuring that outcomes remain equitable and contextually relevant for all participating agents.

Coalition Formation and Multi-Agent Coordination
This mechanism provides the enabling framework for agents to dynamically create, dissolve, and restructure alliances in response to evolving objectives, environmental factors, and complementary capabilities. Leveraging clustering and grouping algorithms, it facilitates the rapid formation of effective teams while allowing for flexible reconfiguration as conditions change. The mechanism supports iterative refinement of coalition structures, ensuring that agent teams remain agile, adaptive, and well-prepared to respond to emerging challenges and opportunities.

Market-Based Mechanisms:
This mechanism provides the enabling framework for market-oriented approaches by integrating techniques such as auction and bidding. It can guide allocation (of tasks or resources) based on real-time demand, opportunity costs, and operational constraints. This mechanism can infuse economic principles into the system, promoting efficiency and incentivising cooperative engagement among agents.

Reputation and Trust Mechanisms
This mechanism establishes a framework for evaluating agents’ historical adherence to social norms of behavior. It systematically develops reputation metrics that reflect both past performance and adherence to community standards, enabling agents to gauge credibility and reliability. This mechanism informs agents about trusted peers and discourages deviations from established norms, ultimately reinforcing ethical conduct and strategic decision-making within the network.

Conflict Resolution and Mediation Mechanisms
This mechanism provides the enabling framework for resolving disputes by integrating detection and mediation protocols. It offers structured methods for identifying conflicts and for addressing disagreements that arise from divergent objectives or unforeseen constraints. The mechanism incorporates iterative feedback loops and adaptive resolution strategies to dynamically adjust conflict resolution approaches in real time. By ensuring that disagreements are promptly mediated through structured negotiation and automated arbitration, it minimises the risk of prolonged deadlocks or breakdowns, thus maintaining smooth and stable multi-agent interactions.

Custom Game Constructs for Multi-Agent Interaction
This mechanism provides the enabling framework for developing custom game constructs that empower agents to negotiate, trade, and collaborate across a wide range of sectors. By integrating principles of game theory and mechanism design, it equips agents with dynamic tools to optimise strategies, allocate resources, and engage in diverse economic and service-based interactions. The mechanism supports iterative refinement of these constructs to accommodate evolving market conditions and domain-specific challenges, ensuring flexible and effective multi-agent cooperation.

By standardising negotiation and collaboration processes, the Game Theoretic Layer reduces the overhead of managing ad-hoc agreements and cultivates a cohesive, strategically aligned ecosystem. Market-inspired constructs improve efficiency and reliability while supporting new economic models and dynamic value exchanges. Ultimately, this layer helps transform self-organising agents into a network capable of robust strategic coordination.

2.3 Agent Execution Layer

Agents at the Execution Layer interact based on diverse goals, capabilities, resources, and learning capacities. The primary function of this layer is to provide the minimal protocols and mechanisms that enable these interactions while preserving each agent’s autonomy and strategic intelligence. Since individual agents possess advanced decision-making algorithms, this layer provides the minimal operational protocols necessary to facilitate resilient and adaptable task execution—whether they are operating within a single organisation or across multiple organisations with differing goals and constraints. It defines how workflows are synchronised, tasks are distributed, and interdependencies are managed, allowing agents to coordinate without prescribing specific algorithms. By specifying standardised mechanisms for concurrency control, fault detection, and fallback procedures, it ensures that agents can respond consistently to exceptions or resource shifts. These protocols enable real-time adaptation—such as reallocating responsibilities or renegotiating timelines—without constraining each agent’s internal logic. In this way, the layer guarantees robust execution of collaborative plans under changing or unexpected conditions, preserving agent autonomy while maintaining overall system coherence in both intra- and inter-organisational environments.

To support seamless integration, the Agent Execution Layer provides unified interfaces and lightweight coordination protocols that agents can use regardless of whether they share a common corporate strategy or belong to entirely separate entities. Below are some key mechanisms within this layer:

Mechanisms for Efficient Task Synchronisation
Standardised interfaces and coordination protocols enable agents to pass tasks and results fluidly, handle inter-task dependencies, and manage concurrent operations without conflict. These mechanisms ensure that multi-step processes remain tightly orchestrated, even under varying workloads or time constraints.

Protocols for Dynamic Resource and Service Exchanges
By defining secure trading interfaces and negotiation pathways, these protocols allow agents within the same organisation to pool specialised assets for joint projects.

Framework for Fault-Tolerant Coordination
Built-in monitoring, detection, and fallback mechanisms track operational health and provide rapid recovery from disruptions. By isolating and containing failures, these protocols prevent cascading effects across the network, thereby maintaining system stability and protecting ongoing tasks.

Procedures for Flexible Adaptation
Agents can revise priorities or shift responsibilities in real time via standardised signalling and reconfiguration protocols. These adaptive procedures ensure that organisational changes, emerging market conditions, or unexpected challenges do not derail active operations, preserving continuity and maximising responsiveness in a dynamic environment.

By abstracting away the operational details and focusing on a minimal but robust set of interaction protocols, the Agent Execution Layer empowers agents to carry out tasks, exchange services, and collaborate effectively without sacrificing their independence. In doing so, it supports a dynamic, adaptive, and service-oriented ecosystem where agents from diverse backgrounds can combine strengths to deliver tangible value and foster innovative solutions.

2.4 Governance Layer

The Governance Layer establishes mechanisms that maintain ethical, transparent, and compliant operation across The Agent Network without compromising agent autonomy or overstepping human oversight. Instead of micromanaging individual actions, this layer provides the structural infrastructure, protocols, and tools to support ethical, transparent, and legally compliant operations. It orchestrates standards for policy definition, compliance checks, and risk assessments, but does not itself police or directly mitigate risks—enforcement decisions remain under human or designated oversight agent control. By defining consistent policies for ethical, legal, and organisational compliance, intellectual property rights, and dispute resolution, this framework enables scalable, real-time monitoring and automated reporting, all backed by a clear escalation path for corrective actions. Frequent updates and community feedback loops keep these protocols effective and aligned with evolving regulations, ensuring that agents operate responsibly within a stable, accountable, and continuously improving ecosystem.

Importantly, the Governance Layer’s focus is on facilitating policy creation and risk-mitigation tools, not on imposing sanctions or penalising agents. It offers built-in monitoring and reporting capabilities that allow decision-makers—be they human operators or specialised oversight agents—to make informed choices about how to address non-compliance. This approach preserves both agent autonomy and human-led governance, providing the necessary guardrails for accountability and transparency while avoiding prescriptive oversight.

Below are some key mechanisms within this layer:

Policy and Norm Definition and Enforcement
This framework establishes and maintains the foundational directives governing agent interactions. By leveraging iterative refinement and continuous feedback, it sets consistent standards for ethical knowledge sharing and ensures seamless interoperability across the network. Its integrated monitoring protocols track compliance in real time, triggering timely adjustments and corrective actions. Regularly updated to reflect evolving regulatory requirements and community norms, it fosters a stable, ethical, and efficient multi-agent ecosystem.

Ethical and Legal Compliance
This framework integrates legal, cultural, and societal standards to ensure that all operations adhere to both regional regulations and universal ethical principles. Through bias mitigation and proactive monitoring, it evaluates agent behavior, implementing context-sensitive corrective measures when necessary. Regularly updated to reflect evolving laws and ethical norms, this mechanism promotes fairness, accountability, and responsible multi-agent collaboration.

Intellectual Property Rights and Ownership
This framework establishes clear guidelines for attribution, licensing, and safeguarding innovations within the network. By employing consistent monitoring and enforcement tools, it ensures contributions are accurately recognised and respected, maintaining equity among participants. Periodic stakeholder reviews and legal developments help sustain an environment of innovation and ongoing respect for intellectual property.

Audit, Monitoring, and Transparency
This framework implements granular auditing of network interactions using structured reporting tools. It detects potential misuse or non-compliance and enforces accountability through transparent disclosures and timely interventions. Regular updates—based on performance data and evolving regulatory standards—guarantee that the system remains trustworthy and aligned with participant expectations.

Dispute Resolution and Accountability
This framework establishes a range of procedures to address conflicts related to collaboration, decision-making, or intellectual property. By leveraging both automated and human-mediated arbitration protocols, it ensures that disputes are resolved expeditiously and fairly. Continuous monitoring and clear escalation pathways support accountability, enabling timely corrective actions and clarifying how outcomes feed back into the broader governance structure. Periodic process evaluations further sustain a transparent and reliable multi-agent ecosystem.

Security, Privacy, and Access Control
This framework enforces robust, adaptable security measures to protect sensitive data, manage permissions, and safeguard privacy. By integrating encryption, multi-factor authentication, and continuous anomaly detection, it provides real-time monitoring and rapid threat response. Regular audits and updates, coupled with adherence to evolving regulatory standards, ensure sustained network integrity, protecting both data confidentiality and overall system trust.

I believe we are on the verge of a major transformation in artificial intelligence, driven by the emergence of The Agent Network—a unified, internet-like infrastructure enabling AI agents worldwide to interoperate seamlessly. Rather than functioning as isolated systems, agents in this network discover and interact with one another to exchange services, engage in economic transactions, and form flexible alliances at scale. Whether agents operate across different organisations—buying, selling, or bartering capabilities—or within a single company, working together on complex tasks, The Agent Network provides the foundational protocols and connectivity needed to unlock collective intelligence.

Just as the internet revolutionised global communication, The Agent Network stands poised to become a foundational platform for AI, where agents continuously collaborate and co-evolve to solve problems far beyond the reach of individual systems. This shift is not merely about incremental gains in efficiency; it promises entirely new forms of value creation through dynamic, market-driven exchanges and integrated teamwork. By unifying previously fragmented efforts into a single ecosystem, The Agent Network can accelerate innovation, drive synergy across diverse domains, and open doors to breakthroughs that no agent—or even group of agents—could achieve alone.

Acknowledgements

I am deeply grateful to Andrew Rashbass for reading earlier versions of this paper and offering numerous helpful suggestions.

Download the full Article