Qianwen AI Intern: 20 Days of Mastering Tech Editorial Tasks

On August 20 evening, Moonshot AI updated kimi-code 0.38.0 on GitHub without fanfare. But this quiet update was caught by Qianwen, an AI intern deployed by the editorial team via a scheduled monitoring task. Soon after, Qianwen pushed a 3100-word draft titled "Moonshot Bets Its $50 Billion Valuation on a Code Model" to the editor’s Feishu account.

The journey started on August 7, when Qianwen APP launched a "Work Assistant" for complex tasks. The editor set up a local "topic-finding" task for Qianwen to monitor nine major AI model companies’ official sites, code repositories, and media reports. When new updates emerged, Qianwen evaluated their worth and coverage depth.

Qianwen uses a scoring system: for the kimi-code update, it gave 3 points (first-tier company), 6 points (new model release), and 2 points (GitHub official confirmation)—totaling 11 points, exceeding the 9-point threshold to trigger drafting. It then gathered seven days of data, organized news points, and generated the draft quickly.

Tech editors often spend most time on invisible work: tracking scattered sources (official sites, GitHub, Hugging Face, API docs, media), verifying info, and judging story value. Qianwen took over this by running four crawlers hourly, with deduplication to avoid redundancy.

Training involved teaching Qianwen the full news pipeline: finding topics, evaluating value, reporting to editors, and drafting. The scoring rules included tiered company points, event type weights, and source credibility. Qianwen flagged gray areas (e.g., product partnerships) for editor input to refine rules.

Qianwen also helped with daily newsletters. The editor shared past issues for format learning, set two time windows (9:00-22:00 and 22:00-7:00), and added requirements: 2-3 backups per section, original links, and basic formatting. Qianwen delivered error-free, well-organized newsletters.

Sources

• Guancha.cn (Article ID: 1724039)