汤道生×姚顺雨：腾讯AI下半场 拼的是“模型×产品”系统能力

This article summarizes a conversation between Tencent executive Tang Daosheng and Tencent's chief AI scientist Yao Shunyu at the 2026 Tencent Cloud AI Industry Application Conference. Key takeaways are as follows:

1. The "second half" of AI development has begun, shifting its focus from chasing larger parameter counts and leaderboard rankings to solving real-world problems and delivering actionable value. The core requirement is a balanced focus on foundational infrastructure, product value, and cutting-edge exploration, with competition now centered on the systematic capability to integrate models with products.

2. The first principle of product development remains unchanged in the AI era: it still centers on solving real user needs and creating user value. AI-era products must accommodate users' open natural language requests, leverage large model capabilities to translate these requests into concrete deliverable tasks. At the same time, AI is reshaping the entire product development workflow.

3. AI is still at an early development stage comparable to the early days of personal computers, far from reaching its final form. Its future development trajectory will diversify, with new application forms and product opportunities continuously emerging, leaving ample room for exploration.

This conversation shares the latest direction of AI industry development, offering valuable insights for brands to grasp AI-driven consumer trends and implement AI-powered product innovation:

1. The core logic of product development in the AI era still lies in anchoring to real user needs and delivering tangible value. Brands should not blindly chase technical concepts or large-parameter leaderboards when deploying AI; instead, they should identify user problems that need solving aligned with their own business scenarios.

2. Successful AI implementation requires the co-design of models and products. Brands can leverage large model foundations from mature platforms, embed AI into their own business workflows, and iterate models through feedback from user behavior data and scenario-specific needs, forming a positive cycle of implementation and optimization.

3. The overall AI track has already grown into a trillion-dollar market and remains in an early development stage. Its future development will diversify, and there are still massive AI innovation opportunities on the consumer end, so brands can seize this window for early layout. When building AI products, brands should also focus on cost-performance, prioritizing performance before optimizing costs, and balancing return on investment.

This conversation reveals the latest developments in the AI track, providing clear opportunity hints and risk references for sellers. Key takeaways are as follows:

1. The AI track is still in a period of rapid growth; the "second half" of AI has just begun, and the industry is far from reaching its endgame. New application forms and product opportunities will continue to emerge in the future. The overall track has already reached a trillion-dollar scale and is still expanding continuously, making it a high-potential growth market.

2. Sellers can access growth support through the Tencent ecosystem. Tencent has launched the second phase of its AI Co-Creation Camp, which is open to ISV and MSP partners to co-build industry solutions and implement benchmark cases together, providing ready-made cooperation channels and growth support for all types of sellers.

3. When deploying AI, sellers should avoid the pitfall of blindly chasing larger parameters and hyped concepts. The core is to focus on implementation to meet real user needs, while paying attention to the cost-performance of AI investment. Sellers should prioritize getting core tasks right the first time to reduce subsequent error correction costs, optimize investment and control operational risks on the premise that performance meets requirements.

The discussion on AI implementation in this conversation offers many valuable insights for factories advancing digital transformation and exploring AI empowerment:

1. The core of AI's "second half" is solving real-world scenario problems. When advancing digital and AI transformation, factories do not need to blindly pursue large-parameter, top-ranked models. Instead, they should anchor to real needs in production, product design, supply chain management and other links, and prioritize the model's ability to solve practical problems.

2. Factories can access implementation support through the Tencent AI ecosystem. Tencent has high-frequency scenario touchpoints covering both C-end and B-end markets, enabling it to embed large models into factories' actual business workflows. It also boasts mature AI infrastructure to ensure stable operation of AI systems, and has opened its AI Co-Creation Camp to support partners in exploring industry implementation solutions together.

3. The application of AI in industrial manufacturing is still at an early stage, and future AI development will diversify. When factories explore AI empowerment combined with their own production scenarios, they have the opportunity to build differentiated competitiveness. AI brings new business opportunities for factories to reduce costs, increase efficiency, and drive product innovation.

This conversation clarifies the development trend of the "second half" of the AI industry, sorts out core pain points currently facing the industry, and clarifies business directions for AI service providers. Key insights are as follows:

1. Industry development trends are clear: AI has shifted from the first half of competing on parameters and chasing leaderboard rankings to the second half of pursuing value in real scenarios. Core competitiveness now lies in the systematic capability of co-designing models and products. Currently, Agent and Coding Agent are the core development directions. The overall track has reached a trillion-dollar scale, is still growing rapidly, and offers ample market space.

2. The core pain points for current clients are two-fold: many clients blindly chase large model parameters and leaderboard rankings while ignoring the actual needs of their own business; second, most clients lack the ability to balance AI performance and costs, and have not established a mechanism for collaborative iteration of models and business.

3. Service providers can learn from Tencent's Co-Design model: help clients deeply integrate models with business products, build an evaluation system tailored to real scenarios, iterate and optimize models through business data feedback, and help clients build a reasonable cost-performance framework. Service providers can also join Tencent's AI Co-Creation Camp to jointly develop implementation solutions for segmented industries.

In this conversation, Tencent shared its experience and latest moves in building an AI platform, offering many valuable references for all types of AI platform operators:

1. The current core market demand for AI platforms is to deliver AI capabilities that can be implemented in real scenarios, helping enterprises embed AI into business workflows to create tangible value. Users no longer only pursue large parameters and leaderboard rankings; they value AI services that solve problems and offer reasonable cost-performance more.

2. The latest development path for platforms to reference: leverage your own multi-scenario advantages, adopt the co-design route of integrating models with products, embed large models into real business through high-frequency scenario touchpoints, build complete AI infrastructure to ensure the stable and reliable operation of large model applications, open up your ecosystem, and launch AI co-creation programs to attract partners to jointly build industry solutions and accumulate benchmark cases.

3. Development pitfalls to avoid: do not fall into homogeneous cutthroat competition of chasing parameters and leaderboard rankings; instead, pursue differentiated and diversified development. Recognize that AI is a long-term track, do not chase short-term explosive growth, make long-term layouts, iterate continuously based on your own scenario advantages, and seize the continuously emerging opportunities for new applications.

This conversation sorts out the latest developments in the "second half" of the AI industry and puts forward a new framework for industry development, offering high reference value for AI industry research. Key takeaways are as follows:

1. New industry trends are clear: AI development has officially entered the second half, with its focus shifting from pursuing parameter scale and leaderboard rankings to delivering value in real scenarios. Core industrial competitiveness has shifted from single model capability to the systematic capability of model-product integration. Currently, the overall AI track has grown to a trillion-dollar scale and remains in an early development stage comparable to the early days of PCs. It will diversify in the future, and Agent and Coding Agent are clearly defined as the current core development directions.

2. New problems facing the industry: the industry suffers from concept abuse and homogeneous competition. Many practitioners blindly chase parameters and leaderboards while ignoring the value of real scenarios. There are also misunderstandings about AI cost-performance: many only focus on costs at the model architecture level, while ignoring the impact of performance and systematic capability on cost-performance. Most enterprises lack an effective mechanism for collaborative iteration of models and products.

3. New research directions are proposed: a development framework that equally emphasizes three pillars — foundational infrastructure, product value, and cutting-edge exploration — along with the Co-Design model-product collaboration model, is introduced. It also splits cost-performance into two layers: performance priority followed by cost optimization. This framework clarifies the direction for industrial development and also provides new research topics for researchers.