2025 AIGC Overview: What Product Managers Should Focus On

As we look back from the point of 2025, the AIGC industry has undergone a necessary and brutal transformation. The first round of scenario exploration in domestic AIGC has essentially concluded, and the leading players have begun to take shape. The era where merely having an API interface could secure funding or attract users with a slogan like “disrupting search” is over.

For product managers, the core question is no longer “What can technology do?” but rather “Who can seize the power of scenario definition in a landscape of technological homogenization?” This article will analyze the survival rules for AIGC products in 2025 based on the report data.

I. Changes in the Model Layer: From “Reinventing the Wheel” to “Collaborative Innovation”

In the past two years, domestic large model vendors have focused on stories about “parameter scale” and “benchmark scores.” However, the market in 2025 offers a more pragmatic answer: Users care less about whether your model is self-developed or open-source; they care about usability, speed, and cost.

A significant phenomenon revealed in the report is the “DeepSeek Effect.” For instance, Tencent’s Yuanbao, as a flagship product, did not insist on a purely self-developed bottom-up approach but chose to integrate DeepSeek-R1 and optimized it at the system level. The result is clear: Yuanbao rapidly ranks among the top tier in inference depth and response time.

This provides an important insight for product managers: The technological gap is being bridged by open-source and collaboration, with the real barriers shifting to “system-level optimization” and “engineering capabilities.”

The current competitive focus has shifted to:

1. Cost-effectiveness battle: How to reduce token prices? (All major models are aggressively lowering prices, even moving towards free).
2. Edge breakout: 5B-level small models running locally on mobile and PC have become standard, not only for privacy but also to reduce cloud inference costs.
3. Deep inference: OpenAI’s o1 and DeepSeek-R1 have ushered in the “slow thinking” era, enabling AI to genuinely tackle complex logical problems (such as business analysis and medical diagnosis) rather than just casual conversation.

II. C-end Battlefield: Competition for Super Entry Points and the “Matthew Effect”

Report data shows that in March 2025, the total traffic of web-based AI products exceeded 600 million, but traffic is astonishingly concentrating towards the top players.

1. Winner Takes All: The AI Assistant Landscape is Set

In the crowded AI assistant (Chatbot) sector, Doubao, DeepSeek, Kimi, and Tencent Yuanbao occupy over 80%-90% of active users and new growth on both web and app platforms. This is a discouraging statistic for small and medium developers, indicating that the window for general-purpose AI assistants has closed. The leading products are doing two things:

• One-stop functionality: Evolving from simple chat to comprehensive tools for “searching, reading, writing, and drawing.” For example, Quark and ima.copilot are not just assistants but resemble an intelligent workspace.
• Comprehensive presence: Competing for “operating system-level” entry points. They are no longer satisfied with users merely opening an app; they aim to achieve “always-on” access through browser plugins, PC shortcuts, and mobile system integrations (like Huawei Xiaoyi and Lenovo Xiaotian).

2. The Emergence of the “Operating System”

Product managers should note the evolution of AI from “single-point efficiency tools” to “end-to-end super tools.” For instance, Doubao and Quark not only provide search but also integrate document editing, cloud storage, and video summarization. This strategy resembles the “super app” approach of the internet era—entering through high-frequency search/conversation scenarios and using long-tail features (documents, cloud storage, news) to keep users within their ecosystem.

3. Reconstruction of Search

AI search is the most penetrative sector, but product forms are diversifying:

• AI-enhanced search: Baidu, Quark. Adding AI to existing search boxes, steadily progressing.
• Native AI search: Mita AI, Perplexity. The report highlights Mita AI Search, which utilizes a “small model + large model” collaboration to achieve a “think before search” logic, excelling in ad-free and structured displays. This proves that even in a crowded search market, a refined vertical experience can still find space to thrive.

III. Next-Generation Interaction: AI Agents and “Intent Direct”

If chatbots are “you say, I listen,” then AI Agents are “you say, I do.” The report posits that AI Agents represent the ultimate ideal form of AIGC.

Currently, there are four main types of agents on the market:

1. ToB digital employees: Specific functions (like customer service, sales).
2. APP-form agents: Such as Zhipu’s AutoGLM and Alipay’s Zhixiaobao.
3. Terminal-level agents: System-level AI from mobile manufacturers.
4. Personalized avatars: Like products from Character.ai.

Product opportunity points: The report mentions AutoGLM and Zhixiaobao, showcasing the future interaction paradigm—Intent Direct. Previously, to book a ticket, one had to: open an app -> click on tickets -> enter dates -> filter -> pay. Now, it’s as simple as saying to your phone, “Help me book an early flight to Beijing tomorrow.”

This shift is revolutionary for product design. It implies that GUIs (Graphical User Interfaces) will gradually give way to LUIs (Language User Interfaces) or a deep integration of both. Future apps may not require complex menu trees but will need highly refined API interfaces for agent calls. “Tool invocation capability” will become the decisive factor for agent products.

IV. Content Sector: The Dichotomy of Professional and Entertainment

In the fields of AI-generated images and videos, product differentiation is evident.

1. Video Generation: From Showmanship to Practicality

Keling AI and Jidream AI (the team behind Jianying) have become the leading stars in domestic video generation. The report notes that these two products have surpassed expectations and lead domestically, primarily due to their product design lowering barriers.

Good technology is just the foundation; good products must solve the “controllability” issue. For example, providing camera control, frame alignment, and motion brushes allows professional creators to genuinely utilize the tool rather than just generating random videos. Additionally, they both have strong communities (inspiration libraries) that address the cold-start problem of users not knowing how to write prompts.

2. Entertainment Scenarios: Emotional Value as a Necessity

AI entertainment is not just about productivity.

• AI music: Products like “Give Me a Mic” and “Sing Duck” focus on social and subculture.
• AI companionship: Although it once faced growth fatigue, products like Xingye and Dream Island maintain specific user stickiness through role-playing and emotional connections.

However, the report also issues a stark warning: Shell-type AI photography and simple AI photo editing apps are seeing a significant decline in data. These functionalities are being rapidly integrated by established giants like Meitu and Jianying. For entrepreneurs, if your product is merely a feature, it is likely to be swallowed by the giants.

V. B-end and Development Tools: A Silent Explosion

Compared to the excitement in the C-end, B-end penetration is more pragmatic.

1. Victory of Vertical Models

Small and medium enterprises are no longer blindly trusting general large models but prefer “open-source models + industry fine-tuning.” In fields like law, healthcare, and finance, specialized models have shown impressive capabilities.

• Law: Contract review, case prediction.
• Healthcare: Medical record structuring, imaging assistance.
• Finance: Research report analysis, fraud detection.

2. Change in Development Paradigms: AI in Software Engineering

For technical product managers, it is crucial to pay attention to MCP (Model Context Protocol) and MaaS (Model as a Service) platforms.

• MCP: Aims to standardize AI tool interfaces, solving the connection issues between models, data sources, and tools. This is akin to the USB interface standard of the AI era.
• AI Coding: Products like Tongyi Lingma and Cursor are transforming “coding” into “logical orchestration.” In the future, the cost structure of application development will fundamentally decrease, shifting the core from “writing code” to “designing business logic.”

Conclusion: Product Managers’ Survival Guide for 2025

After reviewing the “2025 China AIGC Application Panorama Report,” we can clearly sense the shift in industry trends.

1. Abandon the “Large Model Myth”: Stop getting hung up on model parameters. Users do not care about the model used; they care about whether you can solve their problems. Choosing between DeepSeek and GPT-4o is merely a cost and effect trade-off, not a core product barrier.
2. Seek “Scenario Closure”: Pure chat or pure image generation are unlikely to survive independently. Products must delve into specific workflows. For example, not only generating a PPT outline but also directly exporting an editable PPT file, even assisting in layout and graphics.
3. Beware of “Intermediate State” Products: Either create a simple national-level entry point (the giants’ game) or become a deep vertical expert (opportunities for entrepreneurs). Products stuck in the middle as “shell tools” will face existential challenges in 2025.
4. Focus on “Edge” and “Agents”: This will be the next wave of traffic dividends. As mobile systems themselves become the largest AI Agents, how should your app coexist with them? This is a question every mobile internet product manager must consider now.

In 2025, AIGC is no longer a science fiction story but a utility like water and electricity. For product managers, this is both the best and worst of times. Fortunately, as the bubble bursts, we can finally focus on making great products.