Qwen Work Hands-On: An Earnings PPT in 7 Minutes—Three Scenarios Through Alibaba's New Office Agent

On July 27, 2026, Alibaba officially launched its intelligent office product Qwen Work (entry: qwenwork.cn). It integrates three agents—QoderWork, Wukong, and Mulerun—with the goal of decomposing a natural-language request into multi-step tasks and directly delivering PPT, Word, Excel, web pages, reports, code, or multimedia files, while connecting to DingTalk, Feishu, local files, and the browser. The default base model is Qwen3.8-Max-preview, with million-token context, native multimodality, and code generation.

This article runs the full pipeline in three real scenarios: an earnings PPT, a podcast turned into a Feishu doc, and an e-commerce website. Every step's time cost, credit consumption, and pitfalls are recorded so you can reproduce them.

Before You Start

• Entry: desktop app or qwenwork.cn (web version opening later; currently requires desktop or DingTalk)
• Account: registration comes with 2,000 credits, valid for 90 days
• Pricing: Personal Standard ¥78/month (limited-time continuous monthly); Personal Premium ¥158/month; Business Standard ¥198/month/seat; Business Flagship requires contacting sales
• Expected consumption: the three tasks in this article used about 1,000 credits (roughly half a month of the Personal Standard allowance)

Scenario 1: A 12-Page Earnings PPT in 7 Minutes

Feed an earnings report to Qwen Work, have it interpret the data and insert visual charts into the PPT. Here we use Alphabet's 2026 Q2 earnings report.

Step 1: Upload the source file and give the instruction

After uploading the earnings PDF/document, enter something like:

Interpret this Alphabet 2026 Q2 earnings report; design and insert visual charts into the PPT covering revenue structure, Google Cloud, profitability, expenses, regions, balance sheet, cash flow, AI business, and potential risks.

Step 2: Watch the task decomposition

Qwen Work first does deep thinking and automatically splits the job into 6 subtasks:

1. Extract data
2. Write the PPT generation script
3. Run the script to generate the PPT
4. Validate the file
5. Check and fix
6. Final delivery

Step 3: Accept the output 7 minutes later

Measured time: about 7 minutes. The output is a 12-page PPT containing 11 charts (bar, pie, and line charts), all re-editable.

Data accuracy check: comparing the PPT's core figures against Alphabet's official earnings report item by item—

• Revenue up 24% YoY ✅
• Google Cloud revenue up 82% YoY ✅
• Net profit up 298% YoY ✅

💡 Tip: after delivery, manually spot-check 2–3 key figures against the original report. Qwen Work was stable in this case, but AI-generated content must always be verified by a human.

Scenario 2: A Podcast Transcript into a Feishu Doc in 2 Minutes

Testing cross-app execution: hand Qwen Work a podcast transcript, have it distill the content, write it into a Feishu doc, and send it.

Step 1: Authorize Feishu in the connectors

Open the connectors menu, attach the Feishu CLI, and follow the prompts to grant document-creation and messaging permissions.

Step 2: Feed the material and give the instruction

Upload the text transcript of Sam Altman's latest podcast and enter:

Extract the key points and standout quotes from this podcast, write them into a Feishu doc with structured output, and send it.

Step 3: Accept the delivery

Measured time: about 2 minutes. Qwen Work outputs a structured document containing a podcast overview, 10 core topics, quote highlights, and other takeaways, automatically creates the Feishu doc, and sends it to the specified location.

Known issue: garbled encoding in the Feishu doc

When we had Qwen Work write a WeChat-official-account article and convert it to a Feishu doc, the text turned into complete garble and the task failed to deliver. It invoked the Social-content skill; the article structure was acceptable, but the language leaned templated and still needed a human rewrite. The Feishu doc conversion step is currently unstable.

⚠️ Pitfall to avoid: for any Markdown / official-account-format → Feishu doc conversion path, verify with a small test first, or fall back to manual copy-paste. This is the most visible bug in the current version.

Scenario 3: E-commerce Website Generation (Usable, but Slow)

The last task tests web page generation.

The instruction

Build an e-commerce website for digital products.

Acceptance

Measured time: 1 hour 8 minutes. The final product is a site called "NOVA Digital" with 8 products and matching images, reasonably complete in visuals and interactions.

💡 Tip: web generation is the biggest time sink; run it on non-urgent tasks, or break it into steps—"first screen first → product list next"—to avoid tying up one task for too long.

Three-Scenario Summary and Credit Ledger

Total credits consumed: about 1,000 (= half a month of the Personal Standard allowance).

Who It's For

• Analysts / investment research / operations: earnings interpretation + PPT is the strongest scenario, with acceptable data accuracy
• Content creators: structured organization of podcasts, interviews, and meeting notes, fast and usable
• Enterprise teams: native DingTalk / Feishu integration is the differentiator, but mind the stability boundary of document conversion

FAQ

• Q: How does Qwen Work relate to Tongyi Qianwen? Qwen Work is an intelligent office product built on the Qwen model family, integrating the QoderWork, Wukong, and Mulerun agents, positioned as "task delivery" rather than "conversation."
• Q: Are the free credits enough? 2,000 credits ≈ 2 full rounds of testing at this article's scale. Enough for light trial; for heavy use, go straight to Personal Standard (¥78/month).
• Q: What about the Feishu doc garbling? A known issue in the current version. Copy-paste manually, or wait for an official fix to the encoding handling.

Try it at: qwenwork.cn