Executive
15 cases
Run video annotation overnight so it is done by morning
In a feature rundown from Volcano Engine's Wuhan event, GeekPark reports that one company scheduled its video annotation work for the night: ArkClaw opened the annotation platform on its own before dawn, operated it through the browser the way a person would, and the finished work was ready the next morning. The company is not named and no volumes or time savings are given.
Executive
Source: geekpark.net · Apr 2026
Test supply chain demand and supply analysis at a Thai chemicals group
SCG Chemicals (SCGC) told a customer panel at Qwen Conference Thailand 2026 that it tested Qwen/QwenWork on supply chain demand and supply analysis. Its digital and AI director said the tool handled the complex context well and cost noticeably less than some pricier models; this was a test, and no results figures were shared.
Executive
Source: techtalkthai.com · Oct 2026
Give every employee a morning and evening digest
Razorpay's engineering blog says more than 220 employees each run their own isolated Hermes agent. One example reads its owner's Slack, GitHub and Google Workspace, posts a 07:15 morning digest and a 19:00 end-of-day digest to a private Slack channel, and nudges a task tracker every two hours. Any new skill the agent writes for itself can be held for human approval.
Executive · Executive Assistant
Source: engineering.razorpay.com · Jul 2026
Guard focus time on the calendar and get a brief twice a day
An xAI employee describes a calendar bot that protects focus blocks, flags conflicts and school-pickup clashes, and drafts moves and invites only after they confirm. A second bot reads their inbox, Slack and meeting notes and sends an 8am and 5pm brief: what affects their team, open follow-ups, and what they need before upcoming meetings.
Executive Assistant · Executive
Source: x.ai · Sept 2026
Research a market and turn a doc into a slide deck
A GeekPark writer tested ArkClaw through Feishu. Asked to compare about nine OpenClaw hosting offerings and recommend vendors, it returned a fairly complete research brief within minutes. Using a connected third-party skill (Anygen), it also turned a Volcano Engine doc into an editable slide deck; the writer estimates under 10 minutes versus about three hours by hand, and had not yet finished testing a scheduled Feishu-doc workflow.
Executive · Marketing
Source: eu.36kr.com · Mar 2026
Let an agent take your calls and summarize them
Manus says each Cue agent gets its own phone number, email, wallet and computer. It can answer calls for the user and leave a summary in Cue, and it can pay for things within a budget the user sets.
Executive Assistant · Executive
Source: manus.im · Sept 2026
Split event venue research across a team of agents
In Manus's launch example, several Cue agents share one group chat. One researches venues for a New York product launch, another narrows them to a shortlist, a third drafts the presentation, and the user makes the final decision.
Executive · Executive Assistant · Marketing
Source: manus.im · Sept 2026
Give each team its own agent profile for ops work
A 17-person data startup (Artie) runs Hermes on one self-hosted machine with a separate profile per team. Its BizOps profile handles pipeline reviews, reconciliation and tracker updates, and flags exceptions for a human; staff mostly use it through Slack.
Sales · Finance · Executive
Source: x.com · Sept 2026
Research and rank sales leads into tickets
A three-person Polish software house had Hermes write briefs on 24 leads, each a ticket with a company profile, the likely problem, a first-contact suggestion and a hot/warm/cold rating. It also produced a 45-page competitor report and critiqued the team's first outreach email as too generic.
Sales · Marketing · Executive
Source: halfbitstudio.com · Jun 2026
Visualize survey spreadsheet data with one prompt
A Jiemian News reporter uploaded a questionnaire spreadsheet to QwenWork's investment-research expert kit and asked in one sentence for the data to be organized visually. Consistently formatted charts came back within seconds. The reporter says general chatbots needed several rounds of re-prompting to get similar output.
Marketing · HR · Executive
Source: m.21jingji.com · Aug 2026
Turn a quarterly earnings report into a visual brief
A Chinese tech outlet (ifanr) gave QwenWork Apple's quarterly results. It produced an HTML report that leads with the conclusion, charts revenue, profit, products and regions with a short note under each chart, and puts the raw data table at the end. The reviewer rated it strong and easy to check against the source.
Finance · Executive
Source: ifanr.com · Aug 2026
Produce a weekly business brief and inbox priorities
The same tester had Muse write a weekly business brief that needed little editing (9/10) and finish a competitor brief in the background after the app was closed (9/10). Inbox prioritization was weaker at 6/10: Muse wrongly reported a sent email as unsent and resurfaced already-fixed items.
Executive · Executive Assistant
Source: aiagentslibrary.com
Summarize monthly finances and flag odd expenses
Meta's small-business launch pitches Muse for reviewing a month's financial performance and flagging unusual expenses, analyzing sales and campaigns into a growth plan, and drafting replies to emails that need attention. Its one named early tester is an Iowa grocery owner who works about 65 hours a week and says Muse is taking over parts of the owner role.
Finance · Executive
Source: about.fb.com · Sept 2026
Draft a weekly business brief and post-meeting follow-ups
In the same test, a weekly business brief scored 8.6/10 (it came as plain text until a PDF was requested). The post-meeting follow-up email scored lowest, at 7/10: it invented no commitments but was not ready to send and needed two attempts.
Executive · Executive Assistant · Sales
Source: aiagentslibrary.com · Oct 2026
Prepare meeting briefs and triage the inbox
In a two-day, 10-task business test, a reviewer scored Dots 9/10 on meeting prep, but a promised visual deliverable was unfinished. Inbox triage scored 7.5/10: priorities were sensible, but it treated an already-resolved issue as still open.
Executive Assistant · Executive
Source: aiagentslibrary.com · Oct 2026
No case matches these filters yet.
How to read the evidence labels
- Vendor says
- The maker describes it on its own site or launch post. It shows what the product is meant to do, with no independent result.
- Tested by press
- A journalist or outlet ran it themselves and reported what happened, including what broke.
- User reports
- Someone using it at work posted about it. Any numbers are theirs, so we write "the poster says".
- Tested by Wize
- Our team ran it on test data and recorded the result.
Longer reads built from these cases
Our articles compare cases side by side: what worked, what broke, and where a person still decides.