Personal agents still lag: Tencent-media benchmarks Lingguang 2.0, Today AI, and Alice across app access, task execution, memory, proactivity, and interaction.

According to a TMTPost report published on November 4, 2026, its author spent five days testing Today AI and Ant Group's upgraded 2.0 version of Lingguang, two personal AI agent products available in China. The report notes that roughly two months after Meta's Muse became a hit overseas, products carrying the "personal agent" banner began appearing in the domestic market. Two representative ones are Today AI, led by Qi Junyuan, former head of the Doubao PC client, and Ant Group's Lingguang 2.0.

In positioning and product form, Today AI resembles a recreation of Muse. It builds a user profile by gathering large amounts of personal information, then sorts messy multi-task requests into itineraries, flags schedule conflicts between tasks, and identifies likely work needs and hobbies from local files, sharing related information at irregular intervals. The authors describe the result as feeling more like a living person. Lingguang, by contrast, is less like Muse and closer to a Chinese lifestyle butler with restrained data collection, which the report says caps its ceiling. Backed by the Ant ecosystem, Lingguang has access to Alipay APIs; its core capability is analyzing spending habits from billing records, echoing the Alipay AI assistant Abao, while some preset tasks guide users to build mini-apps themselves.

When it comes to actually doing things, Today AI connects scheduling and memory into a complete task network, so it can flag conflicts and make decisions inside more tasks by drawing on the user's profile, delivering more surprises. Lingguang behaves like a restrained butler: each assigned task is done well and it follows up daily, but it does not remember who the user is, so people must ask one more round and think one step further ahead, and the overall experience lacks coherence. The report adds that both products have impressed in testing, yet are blocked by other apps working in concert, confined within their own app walls.