01

From better models to systems that complete work

The measures identify online and continual learning, autonomous evolution, complex reasoning, planning, tool invocation, long context, memory and multi-agent coordination.

They also identify Harness Engineering around context, persistence, collaboration and scalability.

Current policyLast reviewed · 26 August 2026
02

Beijing wants agents inside existing industries

The policy directs agent technology toward industrial software, enterprise systems, scientific research, healthcare, manufacturing and other real workflows, including embedded FDE-style deployment.

Strategic directionLast reviewed · 26 August 2026
03

Token economics is becoming policy

The measures discuss inference efficiency and Token as a Service, Agent as a Service and Result as a Service. This is a business-model signal, not a promise of free resources.

Current policyLast reviewed · 26 August 2026
04

Agents and tiny AI-native companies are linked

The policy connects agent development with OPC communities, flexible compute, incubation, finance, IP support and simplified registration services, while legal structure and immigration remain separate.

Current policyLast reviewed · 26 August 2026