Beijing Unveils Agent Policy, First to Write Harness Engineering, Token Economy and OPC into Official Documents

Key Highlights

Beijing released "Several Measures on Accelerating the Leading Development of Agents," ten articles in total, the first to write frontier concepts such as Harness Engineering, the token economy, and OPC (one-person company) into formal policy. The weight of this lies in the fact that macro policy language, usually abstract, has for the first time precisely absorbed the front-line terminology of the agent industry. Put simply, agents are no longer just a buzzword but have been incorporated into a city-level industrial blueprint. The document matters because policy vocabulary shapes where capital, talent, and procurement flow; by naming Harness Engineering and the token economy explicitly, Beijing is telling builders that the middle layer between model and application is now a recognized, fundable category rather than a hobbyist concern. By speaking the industry's own jargon in law, Beijing also lowers the translation cost between policymakers and builders, so future subsidies and pilot programs can target precise layers instead of blanketing the whole stack.

Capabilities and What Happened

The document's core direction is clear: shift from "billing by token consumption" to "billing by value." This means policy encourages the industry to step out of the old logic of selling compute and instead price around task outcomes and business results. The document explicitly encourages new models such as TaaS (Task-as-a-Service), AaaS (Agent-as-a-Service), and RaaS (Robot-as-a-Service), upgrading agents from a product form to a subscribable, measurable service form. At the same time, policy promotes embedding agents into terminals such as phones, glasses, and cars, letting capability leave the chat box and enter daily devices. The value-billing stance is significant: it pushes vendors to prove impact instead of volume, aligning the incentives of the whole supply chain toward delivered outcomes, which is the only pricing logic that survives contact with a CFO. If enforced through procurement, the value-billing principle nudges buyers to ask what a task is worth rather than what a token costs, which favors agents that actually close work over those that merely respond voluminously.

Technical Details

Writing Harness Engineering into policy deserves a close read: it refers to the middle engineering layer that "harnesses" large models—routing, memory, tool orchestration, safety guardrails—and it is precisely this layer that decides whether an agent can work stably on the job. The token economy is a redefinition of cost structure, emphasizing the cost-performance of unit intelligence rather than sheer volume. OPC (one-person company) points to a new organizational form: one person, assisted by multiple agents, can do what once took a team, and policy leaves institutional room for it. Recognizing Harness Engineering in law is a subtle but real shift: it treats the plumbing—the unglamorous orchestration and guardrails—as infrastructure worthy of support, which is where most of the hard, expensive engineering in production agents actually lives. Naming Harness Engineering also creates a category that startups can position inside, giving investors a clearer map of where to place bets between the model and the application rather than treating "agents" as one undifferentiated pile.

Comparison With Competitors

Compared with other localities' routine moves of compute subsidies or model filing, the special thing about Beijing's ten articles is "concepts first, framework building." It does not only talk about infrastructure but directly regulates the service model and pricing of agents, giving the industry an executable business language. For startups, it means the routes of agent servicization and on-demand billing have gained policy backing. Where rival policies often subsidize the input (GPU hours), Beijing is subsidizing the model (how agents are packaged and sold), which is a more mature stance because it nurtures markets rather than merely capacity, and markets are what turn a technology into an industry. Versus regions competing on cheap compute, Beijing is competing on market design, which is harder to copy quickly and tends to compound as more service providers cluster and standardize around the same rules.

Industry Impact and Use Cases

To put it bluntly, this document gives the agent industry both a "birth certificate" and a "roadmap." In the short term, models like TaaS, AaaS, and RaaS will cluster and pilot in Beijing; in the medium term, agents embedded in terminals will drive on-device compute and interaction innovation; in the long term, institutional designs like OPC may reshape micro-entrepreneurship and employment. For investors and entrepreneurs, this is a clear directional signal: the agent race is shifting from model parameters to service and engineering. The OPC clause in particular is a quiet revolution—if one person plus agents can incorporate and operate at team scale, the policy is implicitly endorsing a future where the smallest viable company is a human with a fleet of software coworkers. If OPC takes hold, the policy may quietly expand the supply of micro-firms, since one founder plus agents can now plausibly attempt what used to require a hired team, softening the hiring risk that stalls many early ventures.