Testing NetEase's Mail Agent AI: Invoices, Interview Conflicts & Email Categorization

The author explores NetEase Email Master's new Mail Agent , an AI-powered chat interface that reads emails and executes organization tasks. Three real-world tests are conducted: reimbursement invoices, interview notifications, and a month of work emails.

01 | Invoice Extraction with Unexpected Detail

Invoices arrive as attachments; manually copying numbers, dates, seller/buyer names, and totals is tedious. The author prompts:

我最近需要报销一些发票，请你把我邮箱里的附件发票提取出来，提取发票号码，开票日期，销售方，购买方抬头，价税合计，并注明信息来自邮件正文还是哪个附件

The agent returns an Excel file with columns for invoice number, date, seller name, and also includes seller tax ID, buyer tax ID, and ticket type — fields not explicitly requested. Results are split into three groups: 15 invoices with complete data, 5 items flagged for confirmation, and 1 non-invoice attachment (fuel card receipt) listed separately. The author notes that pending items still require original verification, but the structured list eliminates manual hunting.

02 | Interview Notifications: Conflict Detection & Reminders

Interview emails often scatter date, location, format, and changes across multiple messages. The author asks for a chronological table with company, position, format, location/link, material requirements, and reply status. The agent first clarifies the date range (recent vs. historical). The output includes two sheets:

Schedule : separates on-site and video interviews with date, time, location columns.

Notes & Reminders : flags two interviews both starting June 11 10:00 as a time conflict; lists dates with multiple interviews (two or three in one day); notes missing timezone (keeps original text); records "no reply seen in mailbox" and suggests verifying status on the recruitment platform.

The author highlights that seeing all interviews in one table reveals daily density invisible when reading emails individually.

03 | Work Email Categorization: Receipt Date vs. Event Date

The author requests a summary of work emails from September 24 to October 24, 2026, categorized by content with representative subjects, dates, and email links, plus a separate list of items needing reply, materials, or deadlines. The agent first states the search range and warns that receipt dates cluster on single days while event dates in email bodies span different months — a crucial distinction for historical review.

The result breaks into five top-level categories, each with sub-items and "view" links:

Foreign trade orders: inquiries, deposits, inspection, booking.

Supply chain procurement & incoming quality: delivery delays, inspection anomalies, contracts needing stamp & return.

Reimbursement & travel.

Recruitment & interviews: online tests, info sessions, technical interviews, reschedules.

Product R&D & project collaboration: weekly meetings, test reports, requirement changes.

The author finds this high-level map useful for quick triage before diving into specific threads.

Conclusion

The author recommends starting with a focused email set: specify exact fields for invoices, define date ranges and required details for notifications, and ask the agent to preserve sources and flag uncertainties. The chat-based interaction inside the email client feels natural and offloads repetitive extraction work, marking a practical step for AI-era email usage.