2026年10月底,蚂蚁集团旗下AI应用“灵光”上线2.0版本,转向Personal Agent方向 1。针对当前赛道缺乏统一用户接受度评估标准的现状,“灵光”提出以“托管量”为核心指标,强调用户将事务委托给AI的行为更能体现产品价值与用户信任 1。灵光产品经理寺二解释称,“托管量”代表用户敢于把多少件生活麻烦事托付给AI,是衡量信任度的基准 1。后台数据显示,只要用户在“帮你管”板块中托管了事务,其七日留存率就会显著高于App内的其他功能 1。
运营数据反映了这一策略的市场表现。2026年11月5日披露的数据显示,用户托管事务超过116万项,累计执行任务超360万次,整体用户规模突破2000万 1。不到一周后,累计托管事务量突破300万项,累计执行任务超过1100万次 1。
与此同时,Personal Agent赛道的其他玩家采用了不同的评估侧重。Instinct公司看重“第三周信用卡绑定率达40%”,以验证交易转化 1。Meta旗下的Muse关注“安全性与权限控制”,核心考量为低失误率和较少的监督负担 1。阿里千问和字节跳动等大厂模型团队普遍将“多步任务的工具调用成功率”视为重要指标 1。由前Teambition创始人齐俊元打造的Today AI,则看重“主动服务转化与上下文记忆深度” 1。行业仍在探索如何衡量信任及实现商业化闭环 1。
In late October 2026, the AI application "Lingguang," under Ant Group, launched its 2.0 version with a strategic pivot toward the Personal Agent sector 1. The release introduced "managed volume" as a core metric to evaluate user acceptance and product value, reflecting a shift from traditional engagement statistics to measuring how willing users are to delegate tasks to AI 1. This approach addresses a lack of unified standards in the competitive Personal Agent market during the second half of 2026 1.
Data disclosed on November 5, 2026, revealed that Lingguang had surpassed 1.16 million managed transactions and executed over 3.6 million tasks cumulatively, with a total user base exceeding 20 million 1. Within less than a week of this disclosure, the cumulative number of managed transactions broke through 3 million, while the total tasks executed rose above 11 million 1. According to internal backend data, users who engage with the "Help You Manage" feature demonstrate significantly higher seven-day retention rates compared to those using other app functions 1.
The emphasis on "managed volume" is intended to serve as a benchmark for trust, representing the extent to which users dare to entrust daily inconveniences to AI, as explained by Lingguang’s product manager, Si Er 1. This focus contrasts with metrics prioritized by other industry players; Instinct Corporation highlights a 40% credit card binding rate by the third week to verify transaction conversion 1, while Meta’s Muse concentrates on safety and permission control, aiming for low error rates and reduced supervision burdens 1. Meanwhile, major model teams like Alibaba’s Qwen and ByteDance generally regard the success rate of tool calls in multi-step tasks as crucial indicators 1, and Today AI, founded by former Teambition founder Qi Junyuan, values proactive service conversion and the depth of contextual memory 1.
评论
还没有评论,欢迎留下第一条。