AI Startup Directions Still Divergent: Getting the Tech Right May Still Mean Getting the Business Wrong

Just 100 days after founding his startup, former Alibaba Group vice president Ye Jun is already worrying about one question: will what he is doing be replaced by the next generation of models?

"Large models are advancing so fast now. If what I'm doing can easily be replaced by a new model, does this thing still have any meaning?" At the TGC Tech Growth Conference on the 23rd, Ye Jun spoke about his post-startup anxiety. His company, Shenkong Zhixu, is currently providing an agent platform for foreign trade and going global. As for judgments about the direction of technology, he admitted candidly that since AI emerged, reality has "slapped him in the face" many times.

This concern also surfaced in investors' discussions. Zhou Jie, head of General Atlantic's China region, said that for some applications that lean AI Native, the boundary between them and the capabilities of foundation models is not clear. As models continue to iterate, some functions are very likely to be covered by big companies.

Entrepreneurs need to judge whether what they have seized is a long-term existing demand, or a brief gap left during the course of model development.

Zhou Jie mentioned that in tracks where some large platforms have strong capabilities and ample resources, investors themselves also disagree internally. Some are optimistic about the market space for ToC, while others question how much advantage startups' products have compared with big companies' products. In her view, fields where big companies have obvious resources and users use them frequently are often "sensitive tracks."

Zhou Jie believes that the biggest point of disagreement in investing in AI today is that the curve of technological development and the curve of commercialization are not on the same timeline. In addition, commercialization faces costs, and whether the commercial value will be large enough once the technology emerges is also a question. She mentioned that some technologies ultimately land, but the commercial returns may not necessarily meet initial expectations, such as the SaaS industry in China's history.

The direction of technology is hard to see clearly from the start. Xu Dong, vice president of the Qwen large model business at Alibaba's ATH business group, shared another misjudgment about rhythm.

In early 2024, they were optimistic about the combination of hardware and multimodal interaction, believing that products such as phones, cars, and AI PCs might become important carriers. But over the past two years, progress in large model applications has been more concentrated in productivity fields such as office work and programming, while the commercialization of consumer electronics and multimodal interaction has lagged relatively behind.

If the boundaries of technology and platforms are still changing, where should entrepreneurs enter from in the next two to three years? In this discussion, industry figures are more optimistic about enterprise-oriented productivity scenarios.

Li Shuhao, founder and CEO of Taiton Technology, believes that the appeal of the enterprise market lies in the fact that customers already have budgets for continuous investment. He summarized the opportunities into three types of spending: marketing and sales, programming and product development, and management. If AI can help enterprises increase revenue or reduce costs in these areas, it will have a better chance of securing continuous payment. Targeting scenarios such as cross-border e-commerce and foreign trade, Taiton Technology released the marketing agent Navos Pro, hoping to drive AI into enterprises' business processes.

Zhao Liping, executive head of the global technology research team and executive managing director at CICC Research Department, believes that AI is still in an early stage of development, and it is difficult to see particularly clear technical barriers; data and industry Know-How may be more important. She suggested that enterprises do not need to wait until products are polished to perfection before testing the waters; they can accumulate industry experience through returned data in actual use and gradually build competitive advantages.

There has previously been industry discussion about when AI applications will reach the widely anticipated "GPT moment." Li Shuhao's judgment is that commercial applications are unlikely, like underlying general-purpose technologies, to see a unified inflection point at a certain point in time. Applications need to interact with consumers and producers over a long period, and are more about continuous accumulation in segmented industries. Biopharmaceutical R&D, marketing, e-commerce operations, and education each have different business logic and will also have different maturation times.