China Telecom's Office AI Agent Just Hit 1.5 Million Users. A Phone Company Is Quietly Winning Workplace AI.

On October 9, at the opening of PT Expo China 2026, China Telecom's executive director and chief operating officer Liu Guiqing announced a milestone that would have been unthinkable a year ago. TeleAgent, the company's office AI agent, has passed 1.5 million registered users, with more than 200,000 daily active users. The product was unveiled just months earlier at the World Artificial Intelligence Conference, and in mid-June it had only 27,000 registered users. That is roughly a 55-fold increase in under four months, a growth curve that would be remarkable for any enterprise software product, and especially striking for one that barely existed in the spring.

The announcement, distributed through Media OutReach Newswire by China Telecom Global, came with a thesis. "TeleAgent marks the shift of AI from understanding what is said to getting things done," Liu said, a framing that positions workplace AI as having crossed a threshold: from chatbots that answer questions to agents that complete tasks end to end. Western vendors spent 2026 doing the same thing, repositioning their work suites around autonomous agents. TeleAgent is the first large-scale signal that the same shift is happening inside China, and it is being built not by a software giant but by a state-owned telecommunications carrier.

The growth story, in numbers

Every figure here comes from China Telecom itself, so treat them the way you would any vendor milestone, but the components are worth unpacking. Registered users jumped from 27,000 in mid-June to 1.5 million by early October. More than 200,000 use it daily, roughly 13 percent of registered users, a respectable ratio for enterprise software.

In a survey of enterprise pilot and power users, 90 percent said the agent had become integral to their daily routines, with document drafting, industry research, and automated data analysis accounting for more than half of all usage. Seventy-nine percent reported significant productivity gains, 35 percent said the tool exceeded their expectations, and the Net Promoter Score came in at 62.9. Among the 74 percent of respondents who had previously used rival platforms, 89 percent rated TeleAgent on par with market leaders, while 16 percent rated it superior.

The most persuasive evidence is a pair of workflow examples. One consulting analyst compressed a recurring quarterly task, exporting 20-plus reports from five systems plus half a day of manual cleanup and charting, down to five minutes. Inside China Telecom itself, a regional sales and channel operations unit used the agent during a major commercial campaign to cut daily cross-system consolidation from hours to minutes, automating report distribution entirely. These are vendor-supplied case studies, but they describe exactly the kind of drudge work that decides whether enterprise AI actually gets adopted.

A 55-fold increase in under four months suggests the enterprise agent era is not waiting for perfect technology. It is arriving wherever the plumbing is ready.

What TeleAgent actually does

TeleAgent is a native AI application for the workplace built by China Telecom AI Technology Co., known as TeleAI, aimed at individual professionals, small and medium enterprises, and large state-owned conglomerates. The interaction model is deliberately low-friction: users drag and drop files into a workspace and let the agent handle execution across documents, spreadsheets, and briefings.

Individuals get a native desktop app for Windows, macOS, Kylin OS, and UnionTech OS with no server setup. The enterprise SaaS edition adds multi-device collaboration and scenario customization. And a private-deployment edition, released in mid-August, keeps models and data entirely within the customer's own environment, the sovereign-AI option that matters enormously in China and increasingly in regulated industries.

Under the hood, TeleAgent runs on China Telecom's own Xingchen multimodal foundation model, plugged into what the company calls a "five-in-one" intelligent cloud. An agent hub called "Xing Xiaochen" handles intent recognition, task planning, and tool orchestration, designed to be reused across the carrier's customer network. Partner integrations cover navigation (AutoNavi/AMap) and corporate travel (Umetrip), with 60 agent applications from 44 external partners in technical integration.

The economics are explicitly managed rather than left to chance. China Telecom says it uses an intelligent model router to direct requests to the most cost-effective foundation model, combined with context compression to reduce latency and inference costs, an approach it calls "token-based operations." In IDC's evaluation of Chinese enterprise agent products, TeleAgent scored 6.85 and ranked in the top three nationwide.

Security is the selling point, not the footnote

For an agent that touches files, systems, and memory across an enterprise, the security story is the product story. TeleAgent implements what China Telecom describes as a four-layer defense loop covering perception, reasoning, execution, and memory, with a lightweight security sandbox that strictly isolates file access. The product is certified by the China Academy of Information and Communications Technology (CAICT) and, the company says, cleared 99.7 percent of the 336 threat evaluation scenarios run by QiAnXin and other leading security vendors.

The more interesting moat may be the skills marketplace, which has grown from roughly 40,000 to more than 50,000 skills, 99 percent of them built directly by end users solving their own productivity problems. That is a genuinely unusual statistic. Most enterprise software companies build template libraries centrally and watch adoption stall. A marketplace built by users suggests TeleAgent has found the power users who turn a tool into a platform.

This also helps explain the survey finding that 16 percent of respondents who had used rival platforms rated TeleAgent superior. The comparison is self-selected and self-reported, so read it with caution. But the direction of travel is what matters. It appears to be winning not on raw model quality, where Western labs still hold cards, but on being tuned for Chinese-language workplace workflows, with persistent memory that adapts to user habits.

Why a phone company is building your office assistant

This is the part of the story that deserves a step back. China Telecom is a telecommunications carrier, not a software company, yet it is now one of the largest distributors of workplace AI agents in the world. The logic is infrastructural. Carriers already own the enterprise relationship, the billing channel, the network, and the data center. Adding an AI agent on top turns the network into the platform, and the private-deployment edition turns regulatory compliance into a feature.

The multinational angle is explicit: China Telecom plans to extend these capabilities through China Telecom Global as secure, localized digital transformation for multinationals. Whether enterprises outside China will run office workflows on an agent built by a Chinese state-owned carrier is an open question for procurement departments, not press releases.

For now, TeleAgent's milestone is best read as evidence for a thesis: the workplace AI race is no longer about who has the best model. It is about who owns the workflow, the deployment option, and the trust layer. A 55-fold increase in users in under four months, built by a telecom, running on a carrier's own model, with skills built by the users themselves, suggests the center of gravity is shifting from the labs to the infrastructure. The next phase of enterprise AI will be won by whoever makes the agent boring, local, and cheap to run. China Telecom appears to have read that memo before most of its competitors.