AI Personal Assistants, AI Phones and Embodied Intelligence Have Yet to Find Real Demand

AI is an all-encompassing, epoch-making technology, but its concrete impact remains hard to predict. Professor Zeng Ming describes three stages: (1) infrastructure, where tokens become the unit of measure for on-demand AI services and 2026 is expected to mark the close of this phase; (2) an explosion of applications, mostly agent-based enhancements layered onto existing businesses; and (3) AI-native applications that redefine tasks, organizations and business models. The author argues that token costs no longer block high-value AI scenarios. In the past four years, only a few areas have truly exploded: AI Q&A from ChatGPT and Doubao, which is replacing search engines; AI programming and AI office tools such as Claude Code, which have achieved real commercialization by replacing labor; and AI video generation such as Seedance and Kling, which is popular mainly in China. AI Q&A has not yet monetized. ChatGPT has 1.1 billion weekly active users but only 50–60 million paying subscribers, a conversion rate of 5–6%. For direct-to-consumer paid services, conversion typically stays below 10%, which is why ChatGPT is exploring advertising. Meta's Muse AI personal assistant, along with Alibaba's Qianwen Go and Ant Group's Lingguang, reflects a pattern in which corporate narratives and ambitions outweigh real user demand. The author concludes that AI personal assistants, AI phones and embodied intelligence have not yet found genuine demand, and that companies are moving too early, driven by the urgency to find consumer-facing scenarios that justify huge capital expenditure.