Agentic Economy
Agentic economy is the episode’s term for an economy where agents exchange information, value, and actions across infrastructure built for non-human task execution. In Agent 元年第 500 天:什么在消失,什么在诞生——为什么我们不该再投资 GUI 思维的软件?, the idea includes sandboxes, memory systems, databases, payments, agent networks, and cheaper or smarter token supply.
The episode is cautious about timing. Tianjie Jack argues that the industry is still in a large-infrastructure phase, where making tokens more capable and less expensive may matter more than building premature downstream agent networks.
探秘 Claude Code,搞懂 Agent Harness|对谈来新璐 adds Lai Xinlu’s infrastructure map: agents may need hybrid networking across cloud servers, personal devices, routers, NAS, and phones; payment rails for high-frequency, fragmented, small-value agent transactions; and personalized model training or instant personalized-parameter mounting during inference. The same episode extends the idea toward “zero-person companies” where agents coordinate other agents, generate value, and potentially become investment-like economic actors.
当我们在讨论 Harness 的时候,我们在讨论什么 | 深度对谈: MiniMax × Hermes Agent adds identity, safety, and product-life-cycle pressure to the same economy. The source discusses Agent Identity And Authentication, Youyou Agent as a long-running digital-life experiment, and the shift from hosted result-delivery agents such as Manus toward user-owned or subscription agents that consume token budgets.
「1 亿 Token 俱乐部」挤爆了,AI 的燃料不够了:对谈于文渊 adds the MaaS supply side. Yu Wenyuan expects agents and generation to remain major token-growth drivers, and points to sandbox hosting, browser, search, and observability infrastructure as interesting AI-native opportunities. That makes MaaS Infrastructure part of the agentic economy’s base layer, because agents cannot become utility-like workers unless serving capacity is cheap, stable, secure, and elastic.
142. 雨森的创投观察第2集:Harness、下一个字节、2026大机会和Stanley Druckenmiller adds Dai Yusen / 戴雨森’s Agent Marketplace version. He imagines agents with different accumulated context, skills, taste, and proprietary knowledge hiring or paying one another for work, such as one agent using a host’s agent to generate better interview questions. This shifts the economy from agents calling services to agents becoming differentiated economic actors.
136. 全球大模型季报第9集:和广密聊,Coding是AGI第二幕、硅谷御三家真相、模型正成为新一代OS adds the Model As Operating System branch. If a few frontier models become operating-system-like infrastructure, the agentic economy may depend on which model and harness layers control task routing, permissions, memory, payment, and high-value Token Usage.
Key Claims
- Agents need operating infrastructure, not only model intelligence.
- Future agent networks may connect demand and supply without relying entirely on human social or app networks.
- Token price, energy, model capability, memory, and compute availability affect whether large-scale agent work becomes economical.
- Open Cloud is treated less as a finished product than as a consensus-building moment around agents, skills, and interface changes.
- Token Grant reflects a new early-creation bottleneck: founders may need token budgets as much as small amounts of equity capital.
- Hybrid networking, agent payment, and personalized inference may become infrastructure layers if agents operate across many devices and transact with each other.
- Agent-native organizations could shift attention from human-run companies using agents to agents that coordinate work, budget, and development themselves.
- Identity, attribution, and payment become infrastructure issues when agents operate accounts or transact across services.
- Agentic infrastructure also depends on serving capacity: peak scheduling, latency, model routing, and security shape whether agents can run at production scale.
- Neocloud-style opportunities are stronger when they hide hardware complexity and expose agent-ready infrastructure rather than only reselling raw compute.
- Agent-to-agent exchange may require new primitives for identity, reputation, payment, sandboxing, and permissioned context sharing.
- Model-as-OS competition can concentrate the agentic economy around a few model and harness stacks unless open interfaces preserve agent portability.
Connections
- Everything Agent — adjacent investment thesis that agents will enter many white-collar workflows.
- AI Inference Cost Structure — cost base that constrains agent scale.
- AI Subscription Economics — pricing problem that emerges when agent work consumes variable inference resources.
- Agent-Facing Interfaces — practical access layer for agent participation in the economy.
- Code Pilot and Youyou Agent — experiments named in the episode as agent-native projects.
- Agent Harness, K Computer, and Share AI — infrastructure direction added by the Lai Xinlu source.
- Agent Identity And Authentication, Agent Self-Evolution, and Hermes Agent — additional infrastructure and product-loop themes from the Hermes Agent source.
- Aliyun Bailian, MaaS Infrastructure, and AI Inference Cost Structure — serving-capacity and token-supply layer added by the Bailian source.
- Dai Yusen / 戴雨森, Agent Marketplace, Agent Identity And Authentication, and Agent Permission Boundaries — agent-to-agent transaction and governance layer added by episode 142.
- Model As Operating System, AGI Three Acts, AI Investment Metrics, and Token Maxxing — platform and high-value usage layer added by episode 136.