Author:Frank,PANews
As AI shifts from technical marvels to practical applications, the implementation of AI applications is accelerating rapidly to meet growing consumer demands. Meanwhile, with the continuous advancement in large model capabilities, AI seems to have entered an era where "everyone can create a product prototype."
During muShanghai AI Week, a roundtable titled "Innovation Practices and Path Exploration of the AI Consumer Ecosystem," hosted by PANews, focused on the real-world implementation paths for consumer-grade AI products. The guests included Feng Wen, Product Lead of the MiniMax Open Platform; Levy, CEO of FateTell; Anita, APAC Head of Sentient; and independent developer and electronic musician Gao Jiafeng, representing diverse fields such as open model platforms, cultural export applications, open-source AI ecosystems, and music creation practices.
In the view of the guests, the core issues of consumer AI have not become simpler with technological iterations. As model capabilities leap forward, the real barriers are shifting towards understanding scenarios, organizing data, user education, emotional value, and building open ecosystems.
AI Has Not Reduced the Difficulty of Entrepreneurship; The Real Barrier Remains the Application Scenario
A common contradiction in the AI industry is that models are becoming increasingly powerful, the threshold for entrepreneurship appears to be lower, but many products still struggle to find long-term viable scenarios. Applications that seem feasible today may quickly lose their relevance with the next model release.
According to Feng Wen, for consumer AI products, product ideas and scenario judgment are still more critical. As a large model and open platform provider, MiniMax places greater emphasis on underlying model capabilities, token-related product design, and the end-to-end experience for developers. However, from an entrepreneur's perspective, products should be designed based on the "intelligence level of the model six months from now."
His assessment is that, in the context of ongoing model scaling laws and continuous improvements in model capabilities, entrepreneurs should not be overly constrained by current model speed, cost, or capability limits. Instead, they should think more boldly about target users, specific scenarios, and the problems to be solved. Model providers will continue to offer cheaper, faster, and more cost-effective capabilities, while the application layer needs to more clearly answer "why this scenario."
Levy added another perspective on barriers from the application layer. He believes that while technology changes rapidly, the data and understanding associated with specific scenarios will not be quickly erased. Many previously thought that only fine-tuning a model could create a data moat. However, with the maturation of context engineering and prompt engineering, the data and structure accumulated in an application's context management can also influence model performance. This is particularly true for highly vertical, culturally specific, or personalized data that may not enter generic model weights, potentially serving as a differentiated foundation for consumer AI products against model iterations.
Anita offered a more cautious view on "AI lowering the entrepreneurship threshold." She noted that while AI has made it easier to generate demo prototypes, quickly build models, and launch initial products, the truly challenging aspects of entrepreneurship have not disappeared—in fact, they may be more prominent. These include how to acquire customers, build community stickiness, achieve commercial viability, and establish human connections beyond coding. She mentioned that concepts like "super individuals" and "one-person companies" are gaining attention, but those who succeed often possess more composite skills than just the ability to call large models.
From BaZi to Music: "Understanding Users Better" Becomes the Barrier for Consumer AI
As technical capabilities continuously advance, the value of consumer AI products must ultimately return to human needs.
The practice of FateTell offers a typical case. Levy introduced FateTell as an AI + Eastern metaphysics/BaZi application targeting overseas users, currently serving users in over 90 countries. The team initially avoided pure efficiency tools, instead focusing on spiritual consumption and emotional value.
In his view, understanding one's destiny, seeking explanations, and finding comfort are underlying psychological needs that cross cultures and endure over time. AI has historically struggled to build trust in this domain, but advancements in models like DeepSeek-R1 have objectively helped users and investors recognize the potential of "large models performing complex reasoning and explanations." The challenges FateTell faces are not limited to model capabilities but also include translating and interpreting Chinese cultural concepts like the Heavenly Stems and Earthly Branches, the I Ching, and BaZi for overseas users, and conveying their appeal through language, visuals, and interaction across different cultural backgrounds.
Gao Jiafeng, from a music creator's perspective, raised a similar issue: AI should not only deliver results but also preserve the process. He noted that tools like Suno make music generation very direct, but they also bypass the creative process, leading to a lack of user engagement and a sense of ownership. For musicians and ordinary users alike, creation is not solely about obtaining a "finished song"; the process itself is part of the experience.
He used football as an analogy: even if ordinary people can never surpass Messi or Ronaldo, they still play the sport out of passion. The same applies to music creation. Gao Jiafeng is developing the "Music AI GameBoy" (music AI game console), which attempts to use AI large or small models to drive musical code, combined with gamified interactions, allowing people with no musical background to participate in creation through play. For him, the real scenario is not "automatically generating a song" but returning the interactive process of music creation to the user.
After the Rise of Agents, User Education Logic is Changing
In consumer AI products, user education often determines whether a product can be genuinely used.
Feng Wen mentioned that among users of the MiniMax Open Platform, some have a development background but are still hindered by API documentation, parameters, error codes, and token usage. To address this, the platform provides model trial platforms, development guides, demo cases, and video tutorials to help developers transition from understanding to implementation more quickly.
With the development of Agents, the methods of user education are also changing. Previously, users needed to read documentation, understand interfaces, and troubleshoot errors. However, with performance upgrades in Agents, many users now rely on Agents to directly read documentation, search for solutions, select appropriate models, and automatically correct paths. Model providers need to optimize models, documentation, and platform experience, while communities, developers, and diverse product forms collectively lower the usage threshold.
For Sentient, the open ecosystem itself is part of user education and product implementation. Anita explained that Sentient focuses on the open-source AI ecosystem and related infrastructure, gathering developers through hackathons, grant programs, and other initiatives. She emphasized that a product must first identify its target users clearly: who they are, where they appear, and how trust is established through which channels. For developer tools, hackathons and ecosystem collaborations are effective entry points; for consumer products, KOLs, KOCs, and social media content are equally important.
In the context of rapidly declining AIGC costs, startup teams can produce trailers, visual materials, and promotional content at lower costs, enabling products to reach their first users more quickly. Gao Jiafeng also believes that product design should strive to be closer to users, allowing them to learn naturally through interaction and entertainment rather than relying on extensive manuals. This "learning by using" approach may be more suitable for consumer AI than traditional tutorials.
Hardware Enters the Real World, Personalization and Emotional Value Continue to Amplify
Looking ahead three to five years, the guests generally agree that the AI consumer market is still in its early penetration phase, but product forms will undergo significant changes.
Feng Wen predicts that in the next three to five years, smart hardware, robotics, and embodied intelligence will reach important inflection points. With improved model capabilities, AI will no longer exist solely in software interfaces but will also enter the real physical world, performing more interactions and tasks. Some products will be aimed at humans, providing efficiency gains or emotional value. Others may be directed at Agents, offering environments, tools, and infrastructure for AI to connect with the physical world. However, regardless of form, products should ultimately remain human-centric, allowing people to devote more time to human connections, family, the real world, and richer life experiences.
Levy believes that making predictions for three to five years in the AI industry is already very difficult; even three to five months are fraught with uncertainty. He observes that while cutting-edge users are deeply engaged with tools like Claude Code, most ordinary users are still in the early stages of AI adoption. In the coming years, AI will further address more fragmented and personalized needs. Compared to the relatively "one-size-fits-all" services of the mobile internet era, AI has the opportunity to provide more specific and segmented services for each individual. Simultaneously, unemployment anxiety and uncertainty driven by technological development may further amplify the demand for psychological companionship and other forms of spiritual consumption.
Anita summarizes this change as "technological democratization." She suggests that future distinctions between humanities, sciences, arts, and technology will be weakened. A small vendor might use AI to create advertisements and target information, thereby improving their business. The value of AI may not necessarily be making everyone a top programmer but helping people in various life scenarios access better tools. Meanwhile, fears of unemployment and loneliness will drive demand for emotional value, creating more opportunities for hardware, AI pets, companion devices, and multi-sensory interactive products.
Gao Jiafeng approaches this from the perspective of cultural transformation. He believes that future forms of content, such as music, film, and video, will be restructured. It is even uncertain whether a "song" will remain the smallest unit of music consumption. Current concepts like multitrack audio and audio tracks may be further deconstructed into more atomic units of creation. However, as forms dissolve, the emotional connections carried by IPs, brands, and specific individuals will become more important. What people seek is not always perfect works but imperfect, warm objects capable of establishing emotional relationships.
Although the guests did not provide a unified answer for consumer AI, discussions from different fields—model platforms, cultural applications, open-source ecosystems, and music creation—collectively point to the same trend: as model capabilities continue to advance, competition in consumer AI is no longer just about "who calls a stronger model," but about the ability to understand more specific users, real-world scenarios, and emotional needs.
The future AI consumer ecosystem may simultaneously include stronger open infrastructure, lower development barriers, more personalized services, hardware with stronger companionship value, and more new product forms centered around culture and the creative process. Models will continue to evolve, but what truly endures will be products that are needed by people, understood by them, and capable of establishing meaningful connections.
