Research on Application Scenarios of AI Agents in Media Content Production

Ren Haitao¹, Zhao Yanming²
(1. Baidu Era Network Technology (Beijing) Co., Ltd., Beijing 100193; 2. School of Language and Communication, Beijing Jiaotong University, Beijing 100044)

Abstract

[Purpose] With the breakthrough development of generative artificial intelligence technology, AI Agents are profoundly reconstructing the entire workflow paradigm of media content production, demonstrating revolutionary potential in reshaping content production logic and expanding creative boundaries. [Method] Based on the technological evolution of AI Agents and industry practices in media, this paper systematically explores their application scenarios and potential value across various stages of content production through typical case analysis and technical path deconstruction. [Results] The study reveals that AI Agents have evolved from rule-based systems (L1) to autonomous agents integrating large language models (L3-L5). Through four core capabilities—multimodal generation, intelligent review, dynamic distribution, and precision interaction—they are driving a transformation in media production from "human-led" to "human-machine collaboration." In the wave of intelligent transformation in the media industry, AI Agents are reshaping the entire content production process from "topic planning—editing and review—distribution and interaction." However, the application of AI Agent technology is accompanied by multiple challenges at ethical, technical, and ecological levels, including risks to content authenticity, pressure for talent transformation, and concerns over data monopoly. [Conclusion] In the future, mainstream media must strike a balance between technological application and value guidance to build a healthy and sustainable intelligent content ecosystem.

Keywords: AI Agent; Intelligent Media; Content Production; Multimodal Large Models; Generative Artificial Intelligence
CLC Number: G222
Document Code: A
Article ID: 1671-0134(2025)05-23-05
DOI: 10.19483/j.cnki.11-4653/n.2025.05.003
Citation Format: Ren Haitao, Zhao Yanming. Research on Application Scenarios of AI Agents in Media Content Production [J]. China Media Technology, 2025, 32(5): 23-27.

1. AI Agents and Their Value in Media Content Production

1.1 AI Agents and Their Development

The concept of "Agent" originated in philosophy, describing an entity possessing desires, beliefs, intentions, and the capacity to take action. In the field of artificial intelligence, this term has acquired new meaning: an intelligent "agent" characterized by autonomy, reactivity, and interactivity. The emergence of Large Language Models (LLMs) has brought new hope for the further development of AI Agents by providing a breakthrough technical solution at the foundational level. LLMs have introduced a new paradigm for deep learning, and their chain-of-thought capabilities and powerful natural language understanding promise to equip AI Agents with strong learning and transfer abilities, making it possible to create widely applicable and practical agents [1]. Research has classified the capabilities and effectiveness of AI Agents into the following levels [2]: L0—No AI: possesses only tool functions with perception + action capabilities; L1—Rule-based Symbolic Intelligence: uses rule-based AI; L2—Reasoning and Decision Intelligence: replaces rule-based AI with imitation learning (IL)/reinforcement learning (RL) and adds reasoning and decision-making functions; L3—Memory and Reflection Intelligence: replaces IL/RL-based AI with LLM-based AI and adds memory and reflection; L4—Autonomous Learning Intelligence: adds autonomous learning and generalization on top of L3; L5—Individual and Group Intelligence: adds personality (emotion + character) and collaborative behavior (multi-agent) on top of L4.

Currently, AI has evolved from no AI (L0) through rule-based AI (L1) and IL/RL-based AI (L2) to LLM-based AI (L3-5). L5 AI Agents signify the emergence of superintelligence, which will possess true digital personalities capable of performing tasks in human roles with superiority beyond 100% of adult human skills. From the most basic rule-driven systems (L1) to potential superintelligence (L5), each level's performance and functionality depends on different technical approaches, demonstrating AI's gradual development from simple task automation to complex, autonomous learning systems. The emergence of LLMs has significantly enhanced AI Agents' comprehension and generalization capabilities, enabling them to better handle multiple tasks and contextual information. This strengthens their natural language processing abilities, thereby providing more personalized and coherent interactive experiences. However, constrained by current limitations in computing power, models, data, and other factors, some AI Agent products still lack "memory capability," "reflection capability," and "planning capability," with product effects often relying on digitalization and previous automation methods.

Although many capabilities of AI Agents remain to be improved, their application has promoted a paradigm shift: migrating from "process-oriented architecture" to "goal-oriented architecture," and from "human-centered, AI-assisted" to "AI-centered, human-assisted." In this transformation, AI Agents demonstrate tremendous application potential and industrial value in multiple directions, including media content creation methods, editorial review, distribution forms, and interactive interfaces.

1.2 The Value of AI Agents in Media Content Production

The application value of AI Agents in media content production can be systematically categorized into four core dimensions. First, multimodal content generation capability: AI Agents can achieve automated content creation across text, images, audio, and video by integrating generative AI and multimodal large model technologies. In news production scenarios, they can automatically generate standardized news reports such as sports event coverage and financial briefs based on structured data, while also enabling script generation and intelligent video editing in film and television creation. Second, intelligent content governance mechanism: Based on deep learning multimodal semantic analysis technology, AI Agents can build an end-to-end content review system. Through the collaborative application of Natural Language Processing (NLP) and Computer Vision (CV), they can accurately identify non-compliant text, sensitive images, and inappropriate video content, while combining knowledge graph technology to achieve semantic-level compliance verification, significantly improving content safety governance effectiveness. Third, dynamic distribution network optimization: AI Agents can dynamically adjust content distribution strategies through real-time data analysis and algorithm optimization. Their core functions include automatically adapting resolution and encoding formats for different terminals, optimizing dissemination paths based on user reach effectiveness, and coordinating cross-platform communication matrices to maximize dissemination effectiveness. Fourth, precision user interaction model: Relying on reinforcement learning and transfer learning algorithms, AI Agents can build personalized recommendation systems based on user profiles. By analyzing user historical behavior data (including click heatmaps, dwell time, and social interactions), they establish multidimensional feature vector spaces to dynamically generate tailored content push solutions for individual users.

AI Agents have restructured the underlying logic of traditional media production and dissemination, not only significantly improving content production efficiency but also expanding the boundaries of creative expression through human-machine collaboration mechanisms.

2. Applications of AI Agents in Media Content Production

Leveraging their autonomous perception, reasoning, and execution capabilities, AI Agents are reshaping the entire content production process of "topic planning—editing and review—distribution and interaction" in the wave of intelligent transformation in the media industry.

2.1 AI Agent Applications in Topic Planning

As the starting point of content production, the "topic planning" stage determines topic direction, audience positioning, and dissemination strategy, with its efficiency and accuracy directly impacting media competitiveness. AI Agents have four core application scenarios in this stage. First, data-driven topic planning: AI Agents identify potential hot topics and user interest points by real-time crawling of multi-source data from social media, search engines, and public opinion platforms, combined with Natural Language Processing (NLP) technology. Compared with traditional manual screening, AI Agents can significantly improve topic discovery efficiency while covering long-tail demands. Second, user profiling and in-depth demand mining: Based on machine learning algorithms, AI Agents can build dynamic user profiles and analyze audience content consumption preferences, time-slot habits, and cross-platform behaviors. Third, hotspot prediction and risk assessment: AI Agents predict the evolution trends of social hotspot events through time-series analysis and causal reasoning models. Additionally, they can identify sensitive content and public opinion risks, providing compliance warnings for planning. Fourth, cross-platform content integration planning: Addressing the needs of media matrix operations, AI Agents can automatically analyze content styles and user differences across platforms (such as WeChat, Douyin, and Bilibili) to generate customized planning solutions.

Typical domestic media cases using AI Agents in the "topic planning" stage include the following. Case 1: Nanjing Media Group's "Xiao Jinling" AI Agent for Two Sessions reporting. "Xiao Jinling" is built on AI technology and the DeepSeek large language model, with deep analysis and precise matching as its core highlights. Relying on a vast database and authoritative information sources from the National Two Sessions, "Xiao Jinling" can not only provide audiences with the freshest and most concerned Two Sessions information daily but also conduct in-depth thinking and precise responses to various user questions such as "What new expectations can Nanjing residents have for the future?" and "What else do people want to say?" It can also integrate authoritative data and policy information from multiple fields including education, employment, and consumption, providing audiences with detailed data support while analyzing from multiple dimensions why these fields attract attention and their future development trends, enabling the public to gain more thorough and comprehensive understanding of livelihood issues involved in the Two Sessions, significantly enhancing the depth and professionalism of Two Sessions reporting [3]. Case 2: Chengdu Media Group's "Yuyan Zhixuan" platform. This platform integrates AI Agent technology to achieve full-chain automation of "data collection—hotspot prediction—solution generation." Taking "Fenxiang Finance," which focuses on Shaanxi urban content, as an example, it utilizes "Yuyan Zhixuan's" rapid short video generation and one-click distribution capabilities to immediately report on emergencies and current political news, condensing core content into 20-second videos that highlight an "instant information" approach, securing first-hand traffic from platforms and producing viral works with over 100,000 views almost daily. Efficient video production capabilities have enabled 80% of the public account's follower growth to come from its video account, achieving the goal of shaping news influence through video [4]. Case 3: Dazhong Daily's "Dazhong" new media platform. In August 2024, the "Dazhong" new media platform launched a new version 9.0 client, explicitly establishing "AI-powered platform" as one of its core strategic goals. Empowered by the Dazhong Party Media Large Model, the diversified functional advantages of AI Agents run through the entire content production chain, improving human-machine collaborative work efficiency and injecting inexhaustible momentum into content innovation. The platform has created multiple AI news columns such as "Morning AI Dazhong" and "Evening AI Dazhong," achieving "news following hot topics," and explores an "AI+ data journalism" model that allows AI Agents to work continuously and automatically integrate and generate livelihood information such as stock market trends, oil price fluctuations, and weather forecasts, which appear on screens after human-machine collaborative review [5].

2.2 AI Agent Applications in Editorial Review

The application of AI Agents in editorial review has evolved from single-function tools to full-process intelligent assistants, with value lying not only in efficiency improvement but also in building a new paradigm of "human-machine symbiosis" for content production. AI Agents have three core application scenarios in the "editing and review" stage. First, automated multimodal information integration: At the material acquisition stage, AI Agents enable intelligent processing of unstructured data. They can real-time crawl full-network text, audio-visual, and social media content, classify and store it after semantic understanding to provide a resource pool for secondary creation. Through real-time clipping and keyframe extraction, they condense hours of live video content into highlight clips, saving manual editing time, particularly suitable for time-sensitive emergency reporting. Second, from writing assistance to one-click generation: AI Agents have evolved from simple text generation tools to multimodal content production engines. They can automatically organize interview recordings into structured notes or even directly generate first drafts. Their video modules can also match media asset library materials based on text to generate short videos with subtitles and background music with one click. Such applications significantly lower content creation thresholds, allowing editorial staff to focus on creativity and in-depth reporting. Third, human-machine collaborative editing and review: The "de-hallucination" capability of AI Agents is particularly important in the editing stage. Through large model chain-of-thought reasoning and information source verification, they ensure content authenticity and logical coherence, automatically verifying historical facts and data sources to reduce manual verification burdens. Such technology automates mechanical editing and review work, freeing human resources for value judgment and creative review.

Current domestic media cases using AI Agents in the "editing and review" stage include the following. Case 1: Xinhua Zhiyun's "Xiaomai" AI Agent for full-process empowerment. As a vertical domain AI Agent specifically designed for media, "Xiaomai" deeply integrates editorial business logic. In the writing stage, it supports automatic organization of interview recordings and image search. In video production, it can complete live clipping, intelligent mixing, and text-to-video generation. Additionally, "Xiaomai" has strong adaptability and can seamlessly integrate with various systems and software. It can also adopt a collaborative model with customers to build customized AI Agents that meet business needs based on their existing business systems and operational processes [6]. Case 2: Cover News's "Xiaofeng Zhizuo" intelligent production chain. The "Xiaofeng Zhizuo" platform integrated in Cover News version 10.0 automatically generates news articles through algorithmic analysis of large amounts of data. Based on preset templates and rules, these articles can quickly cover various news events. For example, in sports reporting, the system can automatically collect match data and player information, combine it with competition background, and generate detailed match reports. Meanwhile, Cover News also uses AI technology for content review. By establishing deep learning models, the Cover Smart Media Review Cloud system can automatically identify and filter out sensitive, pornographic, violent, and vulgar content, greatly improving review efficiency, reducing labor costs, and ensuring platform content healthiness and compliance [7]. Case 3: Wenxin Yiyan and "Cihai" jointly launch "Ciwen" intelligent proofreading agent with precise knowledge traceability. "Ciwen" incorporates Agent RAG technology architecture based on Wenxin Yiyan's four capabilities of understanding, generation, logic, and memory, achieving four major technical breakthroughs: precise knowledge traceability capability, powerful multi-round reasoning capability, dynamic knowledge update mechanism, and intelligent interactive proofreading workflow, making the knowledge from "Great Cihai" learnable and flexibly usable. In practical application, the proofreading efficiency using "Ciwen" has increased by 300%, with error recognition accuracy reaching 80% [8].

2.3 AI Agent Applications in Distribution and Interaction

The application of AI Agents in distribution and interaction not only optimizes the efficiency of traditional information dissemination but also creates immersive interactive experiences that combine virtual and real elements, pushing media interaction from "one-way communication" to "two-way co-creation." AI Agents have three core application scenarios in the "distribution and interaction" stage. First, personalized content recommendation: AI Agents build dynamic user profiles by analyzing user behavior data (such as browsing history, click preferences, and dwell time) to achieve precise content matching. Personalized recommended content not only enhances user stickiness but also optimizes recommendation models through real-time feedback of interaction data, forming a closed loop of "user behavior—algorithm iteration—content adaptation." Second, intelligent customer service and proactive service: AI Agents have achieved a transformation from "passive response" to "proactive service" in media customer service scenarios. In breaking news reporting, AI Agents can automatically generate event timelines, FAQ databases, and rumor-refuting information, and push them to users in real-time through social media, significantly shortening response times. Additionally, multi-language-supported AI Agents facilitate cross-cultural interaction in international communication through automatic translation and localized expression. Third, virtual anchors and immersive experiences: Virtual digital humans, as the embodied carriers of AI Agents, are becoming new entry points for media interaction. Relying on real-time speech synthesis and expression-driven technology, they achieve human-like interaction. Through multimodal interaction (voice, gestures, expressions), they enhance immersion, shifting one-way communication to "participatory narrative."

Typical domestic media cases using AI Agents in the "distribution and interaction" stage include the following. Case 1: CCTV.com uses multiple interactive AI Agents to play with "AI Grand Ceremony." The comment section of the "2024 China·AI Grand Ceremony" live broadcast room specially launched an "AI Writing Assistant" to generate ceremony comments in different styles. For example, the "Panda Style" assistant would comment "The atmosphere of the AI Grand Ceremony is so warm, as comfortable as a bamboo forest." In addition, there are various styles available such as "Su Shi Style," "Zhenhuan Style," and "Zhihu Style." During the live broadcast, the CCTV host AI Agent "avatar" appeared in the CCTV.com APP live broadcast room, presenting an unprecedented novel experience through complete AI program announcements, thorough AI knowledge point explanations, and down-to-earth AI atmosphere comments [9]. Case 2: Wenxin Yiyan jointly launches "Yanbao" hot sports AI Agent with sports vertical platform "Zhibo8." During the Paris Olympics, Baidu Wenxin Yiyan leveraged Zhibo8's deep accumulation in sports content, enabling "Yanbao" to better understand sports and the Olympics, allowing users to freely chat about Olympic gossip and stories, make event predictions, and explain Olympic trivia. This is not only an innovation in traditional sports information dissemination methods but also an active exploration of AI technology in the sports field. Through "Yanbao," users can more conveniently obtain required sports information and enjoy a more personalized Olympic viewing experience [10]. Case 3: China Broadcasting Network builds an AI Agent matrix to optimize customer service and content recommendation experiences. China Broadcasting Mobile's customer service AI Agent learns from over 5,000 knowledge points of proprietary operational service data from the Broadcasting 5G network, with answer output rates reaching 50 Token/s, and has launched experience services for Broadcasting 5G users. Jiangsu Cable uses large models to optimize content recommendations, fully leveraging "AI+ converged media" production quality and efficiency, expanding content service boundaries, and creating more high-quality content with Jiangsu characteristics, style, and spirit to promote deep integration of audio-visual content and cultural IP. Guangdong Broadcasting Network upgrades interactive AI digital humans to achieve natural interaction for film and television recommendations and business processing. Fujian Broadcasting Network connects voice remote controls and digital human "Fujingjing" to DeepSeek, optimizing services such as AI TV watching and intelligent guidance to enhance user interaction experience [11].

3. Risks and Challenges of AI Agents in Media Content Production

Currently, AI Agents continue to expand their multi-dimensional application scenarios in media content production fields such as topic planning, editorial review, and distribution and interaction, with their application potential being continuously explored. AI Agents demonstrate significant advantages in improving content production efficiency, optimizing dissemination paths, and enhancing user interaction experiences, not only reconstructing traditional media production paradigms but also providing solid support for the intelligent transformation of the media industry. However, as application depth increases, related risks and challenges are becoming increasingly prominent.

First, content authenticity and ethical risk issues. The inherent "black box" and "hallucination" problems of AI large language models still exist, potentially causing harmful biases during specific goal achievement and increasing the frequency of operational anomalies. The hallucination defects of large models inevitably extend directly into AI Agents, causing agent hallucinations. Without strict review and verification, false information may be released, further disrupting the public opinion field. Second, technical dependency and talent transformation pressure. Excessive reliance on AI may lead to the degradation of traditional editorial capabilities, while journalists need to transform from "content creators" to "prompt engineers," mastering skills for collaborating with AI. This transformation places higher demands on practitioners' technical literacy, such as understanding AI Agent working principles and mastering prompt engineering techniques, and some organizations may face talent gap risks. Third, data security and monopoly concerns. AI Agents rely on massive amounts of data for training, and media institutions face challenges in user privacy protection and data security. Additionally, the monopoly of technology giants over the AI ecosystem may squeeze the development space of small and medium-sized media, exacerbating industry resource imbalance. Fourth, mainstream media needs to accelerate role transformation. In the wave of intelligence, facing the competition for content dominance from AI Agents, mainstream media is transforming from traditional content producers to "intelligent content ecosystem governors." Mainstream media needs to strengthen the establishment of credible content order by guiding AI values, thereby enhancing public opinion guidance capabilities and consolidating the main position of public opinion [12].

In summary, AI Agents have brought productivity innovation and communication paradigm upgrades to the media industry, constructing new momentum, models, and ecosystems for media industry development. However, they also accompany multiple challenges at ethical, technical, and ecological levels. In the future, mainstream media must balance technological application and value guidance, continuously optimize collaborative mechanisms, strengthen governance capabilities, and build a healthy and sustainable intelligent content ecosystem through policy regulation, industry collaboration, and talent cultivation.

References

[1] Liu Yao, Jiazi Guangnian. 2024 AI Agent Industry Report—The "App" of the Large Model Era, Exploring a New Generation of Human-Computer Interaction and Collaboration Paradigm [EB/OL]. (2024-05-05)[2024-09-13]. https://mp.weixin.qq.com/s/nHGv5C-xAvIFgOt_qWsL0g.
[2] Yu Huang. Levels of AI Agents: from Rules to Large Language Models [EB/OL]. arXiv: 2405.06643.
[3] "Xiao Jinling" AI Agent "Joins the Workforce," Raising a New Wave of AI [EB/OL]. (2024-03-03)[2025-01-23]. https://www.njdaily.cn/news/2025/0303/7437769654613464396.html.
[4] Quan Weifan. The New Future of "AI+ Media" [J]. Pioneer, 2024(04): 27.
[5] Liu Shuang, Song Yao. Media Large Platform Construction and Service Expansion Based on AI Agents—The First Anniversary of the Launch of "Dazhong" New Media Platform [J/OL]. (2025-02-16)[2025-02-24]. https://dz.dzng.com/general/0/NEWS2164490LFMRRVHWBNBWR.
[6] The Most Media-Savvy Large Model Editorial Agent—Xiaomai Is Here! [EB/OL]. (2024-11-21)[2025-01-12]. https://mp.weixin.qq.com/s/9iGcUMe6wfWZS5ZoeU9exA.
[7] Tang Jinlong. Practical Analysis of Artificial Intelligence in Media Content Production and Dissemination—Taking the Evolution of Cover News Client Version 10.0 as an Example [J]. Media Review, 2024(10): 19-21.
[8] Improving Knowledge Proofreading Efficiency by 300%! Wenxin Yiyan and Cihai Jointly Launch Intelligent System "Ciwen" [EB/OL]. (2024-12-30)[2025-01-13]. https://m.sohu.com/a/843380766_121798711/.
[9] Come to CCTV.com, Watch AI Grand Ceremony with AI Gameplay [EB/OL]. (2024-06-28)[2025-01-20]. https://baijiahao.baidu.com/s?id=1803083314122637205.
[10] Zhao Zizhong, Xu Qi. Large Models Deeply Empower Media Intelligent Creation and Integration—China Intelligent Media Innovation Development Report (2023-2024) [R/OL]. (2024-10-11)[2025-02-11]. https://www.fxbaogao.com/view?id=4545916.
[11] China Broadcasting Network Fully Accesses DeepSeek [N]. People's Post and Telecommunications News, 2025-02-21(5).
[12] Cai Jinjin. Mainstream Media AI Agent—The Anchor Point of the Next-Generation Intelligent All-Media Production and Dissemination System [J]. China Media Technology, 2024(11): 13-17.

Author Biographies: Ren Haitao (1984—), male, Master's degree, Head of Baidu AI Technology Ecology North China Region, research interests include intelligent communication, converged media technology, and artificial intelligence; Zhao Yanming (1985—), female, Ph.D., Associate Professor and Master's Supervisor at the School of Language and Communication, Beijing Jiaotong University, research interests include digital humanities and new media.

(Responsible Editor: Li Jing)