AI News AI资讯 8d ago Updated 7d ago 更新于 7天前 45

The website of the future may assemble itself for every visitor 未来的网站可能会为每位访客自动组装

Adobe introduces "agentic sites" that dynamically assemble personalized web pages in real-time based on user intent, moving beyond static audience segmentation. The system uses Large Language Models to retrieve and reorganize existing content grounded in a company's corpus, ensuring factual accuracy while maintaining low latency (1-2 seconds). Current inference costs are estimated at 1-2 cents per page, making the economic model viable for immediate experimentation in high-conversion sectors lik Adobe提出“Agentic Site”概念,利用LLM根据用户实时意图动态组装个性化网页,实现“一人一 Audience”的体验。 系统通过检索现有内容库作为事实依据,而非让模型凭空生成,确保内容准确性并控制延迟在1-2秒内。 单次页面生成的推理成本约为1-2美分,且随着技术发展成本有望进一步降低,经济可行性已初步显现。 网站未来需同时适配人类用户(视觉交互)与AI代理(结构化数据/工具调用),支持不同层级的任务委托。 尽管技术可行,但行业仍处于探索期,企业需在多种AI功能(聊天、生成式UI等)中确定优先落地场景。

65
Hot 热度
60
Quality 质量
65
Impact 影响力

Analysis 深度分析

TL;DR

  • Adobe introduces "agentic sites" that dynamically assemble personalized web pages in real-time based on user intent, moving beyond static audience segmentation.
  • The system uses Large Language Models to retrieve and reorganize existing content grounded in a company's corpus, ensuring factual accuracy while maintaining low latency (1-2 seconds).
  • Current inference costs are estimated at 1-2 cents per page, making the economic model viable for immediate experimentation in high-conversion sectors like e-commerce.
  • Future web architectures must accommodate both human visitors and autonomous AI agents, requiring flexible interfaces that support varying levels of user delegation and interaction.

Why It Matters

This development marks a paradigm shift from reactive personalization (recommendations) to proactive, generative web experiences, fundamentally changing how digital assets are constructed and consumed. For AI practitioners and enterprise leaders, it demonstrates that real-time, LLM-driven page assembly is technically feasible and economically sustainable today, not just a theoretical future state. It also highlights the urgent need to design web infrastructures that can serve both human users and machine agents simultaneously.

Technical Details

  • Agentic Site Architecture: The system interprets visitor signals (browsing behavior, search queries) to categorize intent (e.g., exploring, researching) and uses an LLM to compose a unique page layout and copy tailored to that specific intent.
  • Grounded Retrieval: Rather than generating content from scratch, the model retrieves relevant material from the organization's existing content corpus, ensuring brand consistency and factual reliability.
  • Performance Constraints: Adobe prioritizes speed alongside accuracy, targeting a page generation latency of no more than one to two seconds to ensure a seamless user experience.
  • Cost Efficiency: The current inference cost is approximately one to two cents per generated page, a metric expected to decrease further as model efficiency improves.

Industry Insight

  • Adopt Generative UI Strategies: Organizations should begin experimenting with dynamic page assembly for high-value conversion funnels, particularly where user intent varies significantly across segments.
  • Prepare for Agent-Centric Web Design: Web developers must consider how their sites will be accessed by AI agents, potentially implementing structured APIs (like WebMCP) alongside traditional visual interfaces to support both human and machine interactions.
  • Monitor Cost-Latency Trade-offs: As LLM inference costs drop, the barrier to entry for real-time personalization decreases, making it crucial for businesses to evaluate the ROI of agentic experiences against traditional static or semi-static personalization methods.

TL;DR

  • Adobe提出“Agentic Site”概念,利用LLM根据用户实时意图动态组装个性化网页,实现“一人一 Audience”的体验。
  • 系统通过检索现有内容库作为事实依据,而非让模型凭空生成,确保内容准确性并控制延迟在1-2秒内。
  • 单次页面生成的推理成本约为1-2美分,且随着技术发展成本有望进一步降低,经济可行性已初步显现。
  • 网站未来需同时适配人类用户(视觉交互)与AI代理(结构化数据/工具调用),支持不同层级的任务委托。
  • 尽管技术可行,但行业仍处于探索期,企业需在多种AI功能(聊天、生成式UI等)中确定优先落地场景。

为什么值得看

这篇文章揭示了Web体验从“静态页面+推荐算法”向“实时动态生成”范式转变的关键一步,为电商和内容平台提供了极具潜力的创新方向。对于AI工程师和产品经理而言,它展示了如何在保证低延迟和高准确性的前提下,将LLM应用于前端体验重构,并指出了人机协同与Agent交互并存的未来架构趋势。

技术解析

  • 核心架构:采用“意图识别+内容检索+动态组装”流程。系统首先分析用户的浏览行为和搜索查询,将其归类为特定意图(如探索、研究、购买准备),然后利用LLM从企业现有的内容库中检索相关素材,实时组合成个性化的网页布局和内容。
  • 性能与成本指标:Adobe强调实时性,要求页面生成延迟不超过1-2秒。在成本控制方面,当前单次页面生成的推理成本估计为1-2美分,且预计随模型优化将持续下降。
  • 内容 grounding 机制:为避免幻觉,系统不依赖LLM从头创作内容,而是以网站现有内容作为Grounding Corpus(基础语料),仅对内容的呈现方式、文案侧重和产品选择进行重组和适配。
  • 双模交互接口:未来网站需支持两种访问模式:一是面向人类的可视化界面,二是面向AI Agent的结构化接口(如WebMCP、A2A后端),允许Agent直接调用工具或获取结构化数据,无需经过传统视觉层。

行业启示

  • 个性化进入新阶段:传统的基于规则的推荐或受众细分已过时,基于LLM的实时页面组装能力将使“千人千面”升级为“一人一面”,极大提升转化率和用户体验,特别是在电商领域。
  • 基础设施需重构:网站架构需从单纯的CMS(内容管理系统)向支持Agent交互的混合架构演进。开发者需考虑如何暴露结构化API供AI代理使用,同时保持对人类友好的视觉界面。
  • 战略不确定性中的机会:虽然AI构建能力增强,但“构建什么”仍是难题。企业应从小规模实验开始,聚焦高转化场景(如零售),逐步探索Chat Interface、Generative UI与Agentic Site的组合策略,而非盲目全面铺开。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Product Launch 产品发布 Creative AI 创意AI