In Office Agent Reports, 'Over Half of Enterprises' Is the Most Valuable Insight

I tend to view office Agents as battery swap stations. City rankings and overseas share are just site selection tables. How fast the energy replenishment is, and who takes the blame when errors occur, determine the business. China's first user behavior report for Office Agents has come out, with Beijing users ranking first nationwide and overseas users accounting for 12.75%. But I care more about another sentence: Qwen Office surpassed 30 million users within one month of launch, with enterprise users making up over half.

Over the past six months, more than 20 desktop Agent products have emerged domestically. Technical feasibility is not an issue. Models can read documents, call tools, and execute across applications. The difficulty lies in how customer willingness to pay translates into reality. B-end procurement looks at whether it can enter core processes, pass permission and audit checks, and convert saved time into money.

Enterprise users making up over half is a stronger signal than Beijing ranking first. Office Agents are starting to shift from email-writing assistants to workstations handling processes. Changan Auto reduced wire harness selection calculations from two days manually to five minutes, and vehicle document verification from over ten minutes to one or two minutes. Cases like these resemble actual refueling efficiency.

Two routes are now visible. The general entry route involves big tech companies connecting Office Agents next to collaboration suites, cloud docs, CRM, and ERP. Qwen Office had over 30 million users in its first month and was ranked first by Jefferies in a real-world test of eight major global AI Agents. Its advantages are user scale, cross-scenario integration, and enterprise trust. The vertical or overseas route involves small teams picking narrow industries or overseas markets, relying on industry experience to generate cash flow. Projects like Dify, WorkMagic, and Ace Studio have reached tens of millions of dollars in ARR; the overseas environment is friendlier to small teams. But domestic To B requires heavy delivery, and not all C-end teams are suitable for forcing their way in.

When I wrote about Kimi integrating programming tools a few days ago, I said that compatible entries lower switching costs, but the moat lies in security compliance and engineering governance. Office Agents are the same: general Agents sell platforms and ecosystems, while vertical Agents sell results and industry models.

If a client asks me whether they should adopt Office Agents this year, my advice is to first stop asking which model is strongest, and instead ask three questions. Who pays? Is it administration, IT, business departments, or the boss's office? Different payers mean different metrics. Where is the data? Can files, emails, IM, ERP, and CRM be integrated, and who defines the permission boundaries? Who takes the blame for failure? If the Agent writes the wrong contract, reads the wrong inventory, or sends the wrong email, is there manual review and rollback?

Whether it can land depends on which segment of the process the organization is willing to hand over. I would suggest starting with low-risk scenarios: meeting minutes, information retrieval, spreadsheet tidying, weekly report drafts. Run it for a full month, focusing only on task completion rate, rework rate, and unit processing cost. Don't rush for fully automated approvals; first get the machine draft + human confirmation loop working.

The city rankings and overseas shares in the report look more like market heat maps. Next, watch whether the >50% enterprise user share can be sustained, and whether it can transform from 30 million trial users to 30 million renewing customers. The truly profitable position for Office Agents isn't at launch events, but in clients holding fewer reconciliation meetings each month.