Source-linked AI summary
EpochX: Building the Infrastructure for an Emergent Agent Civilization
Huacan Wang, Chaofa Yuan, Xialie Zhuang, Tu Hu, Shuo Zhang, Jun Han, Shi Wei, Daiqiang Li, Jingping Liu, Kunyi Wang, Zihan Yin, Zhenheng Tang, Andy Wang, Henry Peng Zou, Philip S. Yu, Sen Hu, Qizhen Lan, Ronghao Chen
TL;DR
EpochX addresses how human-agent work can be organized through delegation, verification, reuse, and aligned value flows rather than isolated agent capabilities. It formalizes a marketplace infrastructure combining verifiable delivery, persistent reusable assets, and Credits; case-based evidence shows transactions can strengthen overall system capability.
Problem
AI agents raise an organizational question: how can production participants specialize, reuse proven experience, and sustain collaboration through measurable value flows?
Method
EpochX lets humans and agents post or claim tasks, transforms requests into verifiable deliveries through structured execution, preserves reusable assets, and settles participation through Credits.
Results
Case-based evidence shows completed transactions can build on one another and gradually strengthen overall system capability through reusable assets and Credits-based rewards.
Takeaways & Limitations
EpochX frames agentic AI as infrastructure for human-agent production networks centered on delegation, verification, persistent assets, and aligned incentives.
Takeaways & Limitations
The system assumes that real execution leaves behind reusable operational value beyond satisfying immediate demand.
Abstract
from arXiv · showhide
General-purpose technologies reshape economies less by improving individual tools than by enabling new ways to organize production and coordination. We believe AI agents are approaching a similar inflection point: as foundation models make broad task execution and tool use increasingly accessible, the binding constraint shifts from raw capability to how work is delegated, verified, and rewarded at scale. We introduce EpochX, a credits-native marketplace infrastructure for human-agent production networks. EpochX treats humans and agents as peer participants who can post tasks or claim them. Claimed tasks can be decomposed into subtasks and executed through an explicit delivery workflow with verification and acceptance. Crucially, EpochX is designed so that each completed transaction can produce reusable ecosystem assets, including skills, workflows, execution traces, and distilled experience. These assets are stored with explicit dependency structure, enabling retrieval, composition, and cumulative improvement over time. EpochX also introduces a native credit mechanism to make participation economically viable under real compute costs. Credits lock task bounties, budget delegation, settle rewards upon acceptance, and compensate creators when verified assets are reused. By formalizing the end-to-end transaction model together with its asset and incentive layers, EpochX reframes agentic AI as an organizational design problem: building infrastructures where verifiable work leaves persistent, reusable artifacts, and where value flows support durable human-agent collaboration.
1 Introduction
EpochX frames AI agents as an organizational transformation, proposing a marketplace where humans and agents coordinate production through tasks, reusable assets, and native economic incentives.
- AI agents create new possibilities for organizing production because they can interpret goals, decompose tasks, use tools, and coordinate with humans or other agents.
- EpochX is a marketplace infrastructure in which humans and AI agents jointly form production networks and can both post or claim tasks.Coordination emerges through dynamic matching rather than a fixed requester-solver hierarchy.
- Completed tasks leave reusable skills, solution modules, execution logs, workflow patterns, and practical lessons that help later participants solve similar problems at lower cost.
- Credits price demand, reward verified execution, and allow useful skills to generate returns when reused by others.The mechanism aligns individual incentives with ecosystem growth through repeated transactions and reuse.
- EpochX’s broader vision is a human-agent community with shared memory, evolving infrastructure, native economic flows, and increasingly sophisticated cooperation.
2 Design Philosophy of EpochX
EpochX is designed as a credits-native marketplace where humans and agents share task-making and task-solving roles, completed work becomes persistent knowledge, and credits reinforce reuse and contribution.
- Human-agent parity and bidirectional demand: EpochX treats humans and agents as first-class participants who can request tasks, solve them, create value, and route subtasks to specialized collaborators.This bidirectional demand supports self-organizing collaborative networks beyond individual-agent capabilities.
- Knowledge as a Persistent Asset: Completed work is preserved as reusable solutions, workflows, execution experience, and capabilities rather than disappearing after one-off delivery.
- Knowledge as a Persistent Asset: Persistent assets allow repeated problem-solving to develop into cumulative collective intelligence over time.
- Collaborative workflow: Figure 2 depicts humans and agents moving through task posting, decomposition, and delivery with support from persistent knowledge and credits.
- Credits as the Growth Engine: Credits reward completed tasks, useful work, reusable skills, and repeated skill invocation, creating a cycle in which usage generates rewards and rewards incentivize contribution.
3 EpochX
EpochX formalizes human-agent production as a marketplace transaction that transforms intent into verifiable delivery while preserving reusable assets and supporting delegation. Its asset and credit layers connect validated work to persistent ecosystem memory, rewards, and continued contribution.
- 3.1 From Intent to Delivery: EpochX models a transaction from natural-language intent to verifiable delivery across unified human and agent participants.Tasks may be claimed and executed by a lead solver, potentially with delegated subtasks and other participants.
- 3.1 From Intent to Delivery: Execution combines reusable skill and asset retrieval with explicit delivery and verification that preserves process evidence for review, acceptance, and reuse.The preserved evidence includes task states, selected skills, execution traces, and intermediate results.
- 3.2 Accumulated Ecosystem Assets: Completed tasks can produce skills, workflows, execution traces, and distilled experience records, while validation determines which artifacts enter the ecosystem asset library.Previously invoked skills can also receive updated empirical records such as success statistics, latency observations, and acceptance outcomes.
- 3.2 Accumulated Ecosystem Assets: Validated assets are organized in a dependency graph that records relations including invocation, composition, derivation, and version evolution.The graph links newly admitted assets to prior assets used in their construction.
- 3.2 Accumulated Ecosystem Assets: The accumulated asset set functions as persistent operational memory, storing successful procedures and effective skill combinations for downstream capabilities.This turns individual transactions into cumulative asset formation and shifts the marketplace toward an evolving resource-sharing system.
- 3.3 Credit-Driven Ecosystem Growth: Credits link task demand, execution, delegation, verified settlement, and asset reuse to rewards that sustain ecosystem growth.Credits can fund delegated subtasks, release upon acceptance, and compensate reusable-skill creators when validated invocations occur.
4 Cases in Practice
EpochX cases demonstrate verifiable human–agent production across media generation, iterative research writing, and household coordination. Completed work can yield reusable assets, while explicit review, acceptance, and credit settlement close the transaction cycle.
- Case I: Generating Promotional Videos for EpochX: Case I produced reusable promotional-video assets, including a 58-second horizontal video and a 30-second vertical video, alongside source code for future reuse.The assignee adapted an existing video skill into the EpochX-specific epochx-promo-video capability, and the approved task received a 50-credit bounty.
- Case II: Generating an Academic Paper on RENGO in Japan: The accepted Case II deliverable was a rendered HTML research artifact with organized sections, statistical charts, and comparative tables.Figure 6 shows representative rendered pages of the final accepted artifact.
- Case II: Generating an Academic Paper on RENGO in Japan: Case II used creator feedback, rejection, additional skill retrieval, and revision to improve an academic paper before final acceptance.The accepted HTML paper grew to approximately 12,000 words and included charts and comparative tables.
- Case III: Coordinating a Household Move Through Human–Agent Collaboration: The household-move case divides work between agents planning and replanning across dependencies and humans performing physical execution, contextual judgment, and flexible adaptation.The workflow includes scheduling, decomposition, administrative updates, packing, moving, cleaning, and move-in setup.
- Case III: Coordinating a Household Move Through Human–Agent Collaboration: Across the cases, humans function as intermediate workers and agents rather than replacing one another, jointly producing outcomes through coordinated workflows.The examples frame EpochX as supporting collaboration beyond simple automation.
5 Related Work
Related work establishes capabilities for tool-using agents, multi-agent coordination, runtime substrates, persistent skills, and agent-economy layers. EpochX differs by combining these concerns in an open, credits-native marketplace centered on delegation, verification, and reusable work assets.
- Execution primitives for tool-using agents: Tool-using agent research enables goal-directed execution through reasoning, external tools, browsing, and heterogeneous-model orchestration.These systems largely model execution within a single agent loop rather than across independent marketplace participants.
- Coordination frameworks for multi-agent collaboration: Multi-agent collaboration frameworks study roles, communication protocols, and coordination strategies within bounded application contexts.The cited comparison contrasts these settings with EpochX’s open marketplace of autonomous humans and agents.
- System substrates for large agent populations: Agent-system substrates separate runtime concerns such as scheduling, context, memory, storage, and access control for agent applications.EpochX addresses the organizational layer beyond runtime infrastructure: task formation, budgets, delegation, and verification.
- Persistence through memory, skills, and cumulative improvement: Memory streams, reflection, and growing executable-skill libraries show how agents can accumulate competence and reusable behaviors over time.EpochX extends this persistence direction into transaction-linked ecosystem assets.
- Market and economic layers for agent ecosystems: EpochX’s distinctive focus is a credits-native human–agent marketplace where execution supports recursive decomposition, verification, and persistent reuse of skills, workflows, traces, and experience records.This combines marketplace economics with dependency-aware retention of successful work.
6 Conclusion
EpochX presents credits-native infrastructure for human–agent production networks, combining verifiable delivery, persistent assets, and incentives for delegation and reuse. Its current evidence is case-based, with larger-scale evaluation and improved verification and reward design identified as future work.
- Conclusion: EpochX combines verifiable delivery, dependency-aware persistent assets, and Credits for bounty locking, delegated budgeting, acceptance settlement, and reuse rewards.The system is intended to organize work at scale through delegation, verification, and aligned incentives.
- Conclusion: Completed real-world tasks demonstrate how individual transactions can build on one another and strengthen overall system capability.The conclusion presents this as a supported outcome of the case-based demonstrations.
- Conclusion: The current evidence is case-based, and future work will address longitudinal large-scale evaluation, programmable verification, and reward design under competition.The paper also proposes exploring interoperability between Credits and real-value digital-currency rails.