The Beijing Municipal Bureau of Economy and Information Technology, together with the Beijing Municipal Development and Reform Commission, jointly released the "Beijing Action Plan for Accelerating the Development of the Token Economy (2026-2028)" on September 18th. This marks the country's first special action plan for the token economy issued at the provincial level. The plan outlines ten key measures across six major areas: production-side supply, evaluation and distribution systems, consumption-side ecosystems, international cooperation, industry knowledge frameworks, and factor support.

This action plan introduces several unprecedented concepts for the first time. For instance, it establishes the first-ever grading and evaluation standards for token factories and creates a novel quality assessment system for tokens. Most notably, it is the first official government document to formally introduce the term "token engineer."

Four-Pronged Retraining Strategy for Token Engineers

The tenth measure of the plan explicitly mandates the implementation of a retraining initiative for token engineers. By integrating policy resources related to industry-education collaboration and super economic individual (OPC) industrial training programs, the plan supports token factories, digital industrial parks, and online training platforms. These platforms will conduct specialized training for software engineers in areas such as model optimization, inference acceleration, platform operations, and security compliance, with the goal of cultivating a robust talent pipeline for token engineering. These four training directions directly target the most urgently needed engineering roles within the token economy's industrial chain.

Prior to the release of Beijing's "Ten Measures for the Token Economy," the term "token engineer" had only surfaced in industry discussion papers. This policy marks the first time such a concept has been formally embedded in an official provincial-level government document. The relevance of this plan to programmers extends far beyond just the training measure. For example, the plan explicitly encourages software and information services companies to build token distribution platforms that provide developers with unified interfaces and metering and settlement services. This implies that software engineers will serve as the core workforce required to construct these distribution networks.

Additionally, the "Intelligent Agent Proficiency Enhancement Project" outlined in the plan calls for breakthroughs in critical software technologies, including memory management, security permission controls, and efficient inter-component communication. It also aims to cultivate a group of enterprises specializing in intelligent agent operating systems and new software tools. This signals that the core software ecosystem for intelligent agents will be heavily dependent on the technological advancements of software engineers. Currently, the Beijing No.1 Token Factory and the Jingsuan Token Factory have both been commissioned in the capital, with the first phase of the No.1 facility already achieving a daily production capacity exceeding 1.4 trillion tokens.

AI Coding Tools Shift from "Assistance" to "Core Driver," Forcing a Career Pivot for Programmers

From an industry perspective, token consumption in China has experienced exponential growth. As of March 2026, the average daily token call volume surpassed 140 trillion, a more than thousand-fold increase over two years. This massive scale of token production necessitates a large number of engineers capable of performing model adaptation and fine-tuning, optimizing inference efficiency, and ensuring platform stability. The day-to-day responsibilities of programmers have already undergone a structural shift.

In the workflows of many developers, AI coding tools have transitioned from being a mere "assistant" to the "main driver." Usage rates for Claude Code have skyrocketed from just 3% in early 2025 to 39% by mid-2026, representing a more than twelve-fold increase in just eighteen months. Meanwhile, a report released by the code quality platform Sonar during the same period provides corroborating evidence from the output side: 42% of all committed code globally is now generated by AI, a figure projected to rise to 65% by 2027.

Lyu Weifeng, Vice Chairman and Secretary-General of the China Software Industry Association, stated at a press conference on September 11th for the "Artificial Intelligence + Software" Special Action Implementation Plan that AI is driving a transformation in software production methods, consequently reshaping the talent structure and job requirements within the software industry. He noted that the weight of purely raw coding skills is diminishing, while competencies such as requirements comprehension, architecture design, system integration, and quality and security control are becoming increasingly important. Human-machine collaboration is emerging as a fundamental new skill, where accurately articulating development intent and evaluating and verifying generated outputs are becoming essential abilities for practitioners.

Data from the recruitment firm Career International further illustrates this shift. From the first half of 2024 to the first half of 2026, AI-related job postings grew by approximately 119%, increasing their share of all job listings from 0.72% to 2.01%—nearly tripling the proportion seen three years ago. Demand for AI agent development talent surged by 244% year-over-year, and the supply-demand ratio for AI engineers currently stands at a tight 2.62:1.