As the “AI + Software” policy gains momentum, how can AI agents be deployed at scale?

What scale of agent adoption does the policy target?

On 23 October, a Ministry of Industry and Information Technology meeting called for faster implementation of the “AI + Software” special action, support for agent software and the development of specialized industrial agent products and solutions, with applications explicitly directed at the digital and intelligent transformation of manufacturing.
The special action plan proposes that by 2028, intelligent programming tools and platforms will cover 20 thousand software enterprises above the designated size, a cumulative total of 100 intelligent technological-upgrade projects will be implemented at software enterprises, and 100 benchmark agent software applications will be developed in key industries; by 2030, key software will be fully upgraded with intelligent capabilities. These are policy targets covering both software production methods and customer-facing product applications.

On the supply side, the policy uses earmarked public funds to support key technology research and development, infrastructure construction and real-world applications, while computing-power vouchers reduce software companies’ computing costs. Eligible projects may also receive policy support such as recognition for first-version software products, easing the investment pressure of transformation.
On the demand side, the policy supports procurement of large-model and agent services, encourages software companies to pair up with users in key industries, and promotes connections with high-value applications in manufacturing, finance and other sectors. The simultaneous advance of funding support and customer demand underpins Guotai Haitong’s view that AI applications may move from “localized exploration” to “systematic deployment.”

Software companies have two paths to improvement: intelligent programming is evolving toward autonomous development at the project level and across the full workflow, potentially improving research, development and delivery efficiency; intelligent upgrades of existing software and new intelligent services open up room for product and service revenue.
Yang Lin of Guotai Haitong identifies three capabilities needed to benefit: deep involvement in customers’ core operations, the ability to turn industry experience into reliable agent capabilities, and strong delivery capabilities. Reliable execution of complex tasks, multi-agent collaboration and integration with existing systems are all technical directions emphasized in the plan. Software vendors able to bring these capabilities into real business operations have a better chance of achieving revenue growth and improved profitability.

HAND Enterprise Solutions already has cases of commercial deployment by customers. Its AIGC platform supports industry-specific models and enterprise AI applications, and the company develops business-scenario agents with large-model partners. Its logistics subsidiary Yipo has productized and released several agents, with commercial deployments at multiple customers; the connection lies in delivering products for specific business scenarios.
Nantian Information’s progress is mainly reflected in financial-sector products and model integration: AI models have been applied to its big-data platform and “Shensuanzi” risk-control system, while solutions including the ChatTask assistant platform, a financial product recommendation system, asset management tools and automated workflow assistants have integrated DeepSeek. The two companies’ disclosures respectively reflect customer commercialization and product integration stages; their deployment progress should be understood according to their respective disclosures.