On March 17, 2026, DingTalk launched Wukong as an enterprise AI-native work platform. Then-CEO Chen Hang’s line traveled widely: smash DingTalk, rebuild it with AI, refine Wukong. The substrate was hard too—approvals, calendar, messaging, docs CLI-ized into Agent-callable interfaces.

About four and a half months later, the standalone brand was gone. On August 3, Alibaba opened public beta for Qwen Office (QwenWork): QoderWork, Wukong, and MuleRun merged into one system and would no longer exist alone; the Wukong division upgraded into the Qwen Office division inside Alibaba Token Hub (ATH), led by Chen Yusen, who also owns DingTalk. 36Kr’s Zhidx timeline and the follow-up “Wukong vanished in four months” piece lock that window; Jiemian and peers note the org tempo—Qwen Office went public beta less than two months after Chen took DingTalk and Wukong.

This is not a rename story. Ask why a solo Agent brand could not survive half a year—and what actually fused into DingTalk / Qwen Office at product and org layers.

Mechanism 1: three capability chains; none alone “finishes the job” for enterprises

Publicly, the three products each owned a slice:

Ex-product	Focus	Missing slice
QoderWork	Desktop execution: local files, browser, apps	Enterprise identity, approval, org data
MuleRun	Agent runtime, skills, delivery / reusable flows; overseas / cloud orchestration skew	Desktop front door and DingTalk org field
Wukong	DingTalk collab, permissions, security; org field (~25 enterprise IM capabilities in public materials)	Full desktop execution and unified front-door mindshare

Racing made sense early: validate desktop, cloud orchestration, and org permissions separately. When the window tightened, the gap showed—enterprises do not buy “another chatty Agent name.” They buy one task that can click the PC, run the flow, and stay inside the right permission boundary. Any single product, even strong, can break on the next baton: desktop Agents miss approval; org Agents cannot touch local files; orchestration engines lack a single entry brand.

Qwen Office’s public pitch reconnects the three: describe a goal, Agent plans and calls tools in parallel, delivers editable Office files, web pages, or data reports; powered by flagship Qwen3.8 (including Qwen3.8-Max in public messaging), with desktop and web clients, DingTalk embeds to follow. Cold-start fuel follows the Token narrative: ~2,000 credits on signup, plus daily extras in week one of beta—not “teach the market who Wukong is,” but a Qwen-system usage door.

Coverage (including Caijing-style reporting cited across the wave) also flagged pre-merge friction: Wukong long half-finished, MuleRun skewed overseas/runtime, QoderWork stronger on reputation and scale—multiple brands raised comprehension cost and split R&D. Huxiu’s org read goes further: internal Q-Team / S-Team groups plus Wukong talent folded into one Qwen Office division—rebuild, not skin. Qwen is one of Alibaba’s highest-recognition AI consumer brands; hanging the office door on Qwen is cheaper than re-teaching “who is Wukong.”

So “Wukong disappeared” is more accurately: the brand died; the enterprise-AI org layer rose. Division upgrade, senior attendance in key meetings, and near-term willingness to de-emphasize MAU (Huxiu’s enterprise play: prioritize experience and scene delivery)—the group wants a unified productivity platform, not another launch-day brand.

Counter-intuitive: short brand life looks like failure. If the goal shifts from “ship an Agent” to “model + desktop + enterprise data + permissions + monetization in one product,” freezing a brand too early gets in the way. Four-plus months can validate org-field interfaces; it is not enough to win the entry war alone—especially once WorkBuddy took personal-side noise, Alibaba needed one enterprise-grade card, not three half-finished nameplates.

Mechanism 2: org into DingTalk / ATH; product into “one-stop deliverables”

At product layer, not a Wukong skin—a rebuilt front on three technical stacks, so desktop UI, SOP/orchestration, and org permissions close in one session. Public messaging claims a first: desktop Agent + cloud Agent + enterprise collaboration Agent in one product—subtasks can land on the right execution layer. DingTalk IM is initially wired; enterprise DBs and real workflows sit next in the story.

At org layer, the merge matters more than the name:

1. Single owner: Chen Yusen takes DingTalk and Wukong first, then all three Agents—fewer “three products, three KPIs.” Huxiu also flags why him: Agent problem-solving shown while leading MuleRun.
2. Division upgrade into ATH: Wukong → Qwen Office division beside Qwen consumer, Token Foundry, and MaaS—office sits inside the Token strategy, not as a DingTalk plugin. Separate owners for Token Foundry, consumer Qwen, and Qwen Office (with DingTalk) mark office as a first-class ATH line.
3. Scenes still lean on DingTalk: DingTalk remains cash flow and the gov/manufacturing SMB pool; Qwen Office cold-starts on DingTalk trust and permission models rather than educating a “new Agent company” from zero. Public comments that Wukong capabilities will keep serving many enterprises for a while mean: brand can retire; stock connections cannot be yanked overnight.

Entry-war numbers harden the “why merge.” Huxiu cites June 2026 desktop AI-office agent monthly visits: Tencent-group ~53.8%, ByteDance-group ~23.8%, Alibaba-group ~15.2% in third. With personal-side noise already lost, three half-finished brands only thin share further; one “Qwen Office” card packs DingTalk org assets with desktop/orchestration into a single purchase decision.

Commercial meaning shifts with it: standalone Wukong looked like project-style AI demos; post-merge, pricing and KPIs align more easily with ATH’s Token-processing narrative—seats, private cloud, and Token packs/credits on one enterprise contract. TMTPost’s entry-war analysis makes the same point: folding DingTalk into the Qwen system puts AI-hungry accounts inside a larger Token-burn perimeter. Wukong’s brand steps aside so org-permission assets can qualify for a unified entry.

Risk is in the same mechanism: three teams remaking one product is expensive; if customers only hear “Wukong is gone” without a migration path, half-finished-era trust burns again. Success is whether DingTalk tasks can be delivered in one Qwen Office run—desktop, cloud orchestration, and IM permissions in the same task—not whether the slogan is tidier.

Sources

• Wukong gone in four months: why giants care about AI Work (36Kr)
• Alibaba, ByteDance, Tencent push AI office (36Kr / Zhidx)
• Alibaba’s secret teams and Wukong reorg (Huxiu)