The Zhitong Finance App learned that CICC released a research report saying that the AI office is an important layout point for agents to move towards general productivity, and is expected to become a potential market for large-scale model applications in the medium term. On the supply side of AI offices, improving model capabilities, improving training infrastructure, lowering costs, and market education accelerate the further penetration of office agents. Currently, domestic office agents use the C-side to gain customers. Profit is not a short-term goal. The bank believes that the B-side has a higher willingness and depth to pay, and may become a key commercial market for a long time.
CICC's main views are as follows:
Market space: demand for office agents is growing steeply, and long-term space may reach trillion US dollars
On the demand side, demand for intellectual work in Codex and Claude Code has been growing faster than development tasks since 2026, and products such as Microsoft 365 Copilot have also initially verified payment intentions; the bank believes that although office ARPU is lower than Coding, the user base of nearly 20 times is expected to support considerable market space. On the supply side, improving model capabilities, improving training infrastructure, lowering costs, and market education accelerate the further penetration of office agents.
Core elements: Office agents have different competitive priorities in five core dimensions
1) Model: In addition to the cutting edge of intelligence, more attention is paid to computer operation, multi-modality, and tool call capabilities, as well as cost economy and speed. 2) Harness & Context: The importance of the accumulation and management of context and workflow has become more prominent. 3) Customer portal: Relying on an existing customer portal is beneficial to reduce customer acquisition and deployment costs. 4) Product: Pay more attention to ease of use. Manufacturers often lower the threshold of use through UI interfaces and pre-built template skill libraries. 5) Ecology: Internal and external ecological collaboration helps expand the boundaries of agent capabilities.
Industrial pattern: Multiple types of manufacturers compete at an accelerated pace, long-term commercialization or mainly carried by the B-side
Model manufacturers have cutting-edge model capabilities and promote scenario generalization through third-party ecological cooperation; Internet vendors have full endowments in models, clouds, C-side traffic, and B-side customers, and the domestic “battle for entry” extends further from Chatbot to agents; traditional office software vendors control delivery entrances, and customer resource advantages help reduce customer acquisition and deployment costs. Currently, domestic office agents use the C-side to gain customers. Profit is not a short-term goal. The bank believes that the B-side has a higher willingness and depth to pay, and may become a key commercial market for a long time.
risk
Agent generalization falls short of expectations; inference costs fall short of expectations; security compliance risks