Parallel's Agrawal Builds Agent Search to Replace Ad-Supported Web

Parallel 创始人 Agrawal 打造 Agent 搜索以取代广告驱动的网络 前 Twitter 首席执行官 Parag Agrawal 认为，随着 AI Agent 取代人类浏览网页，依赖广告资助的网络模式将会崩溃，需要在 12 至 24 个月内建立一种基于博弈论的全新变现模型，为 Agent 的数据使用向内容发布商付费。 Parag Agrawal 认为，为人类点击量打造的传统搜索基础设施并不适用于新兴的 Agent 网络，AI Agent 未来的查询规模将达到人类的 1,000 倍。Parallel Web Systems 没有在前期构建庞大的静态索引，而是优先开发了一个搜索 Agent，通过实时抓取来优化架构，在降低 Token 消耗的同时将搜索延迟缩短至 200 毫秒。针对 Agent 绕过广告导致广告支撑网络模式瓦解的问题，Agrawal 提出了一个利用博弈论夏普利值（Shapley values）的框架，根据内容创作者的实际数据贡献直接进行收益分配。Parallel 致力于成为模型提供商和企业级 Agent 的核心搜索基础设施层，目前已正式集成至 Google Cloud Platform。 在训练和优化专为 AI Agent 设计的搜索引擎时，人类的点击流数据和浏览指标不仅不是优势，反而成了一种缺陷。 (0:00) 预计 AI Agent 查询和消费网络内容的规模将达到人类互联网用户的 1,000 倍，这需要一套全新的架构基础。 (0:00) Parallel Web Systems 通过推出一款可执行针对性、按需网络抓取的搜索 Agent 产品，避开了传统前期抓取和建立索引的高昂成本。 (5:00) Parallel 并非构建前沿基础模型，而是作为专门的搜索基础设施层运作，在第三方大语言模型（LLM）之上充当倍增器。 (10:00) Parallel 的 Turbo 产品优化了搜索计算，可将数万亿页面的网络数据压缩至约 1,000 个上下文 Token，将延迟从 3 秒缩短至 200 毫秒。 (10:00)

## Background 本期节目由红杉资本（Sequoia Capital）于 2026 年 8 月 25 日通过播客与网络平台发布，由合伙人 Sonya Huang 与 Andrew Reed 共同主持。节目围绕前 Twitter CEO、现 Parallel Web Systems 创始人兼 CEO Parag Agrawal 的全新创业展开，探讨他如何打破二十年传统搜索由人类点击主导的底层范式，专为未来查询量呈千倍爆发的 AI Agent 重构网络检索基础设施及基于沙普利值（Shapley Value）的内容经济收益分配体系。 ## TL;DR Parag Agrawal 深入剖析了传统以人类点击和 SEO 驱动的搜索引擎为何无法适配 AI 时代，并提出“人类点击数据本质上是 Bug”，Agent 需要的是高信噪比与精准提炼的事实 Token。Parallel 采取逆向策略，先推出深度研究型 Search Agent 替代传统人工外包调研并在真实场景中沉淀评估与索引，随后研发出 200 毫秒极速响应的 Turbo 检索层，成功打入 Google Cloud 企业级 Agent 生态。针对广告商业模式在 Agent 时代瓦解的困境，他提出利用博弈论中的沙普利值建立双向差分定价模型，以替代算力成本量化网页边际贡献，计划在 12 至 24 个月内向开放网络出版商分发真实商业收益。最终，他展望了智能体交互从被动即时“拉取（Pull）”走向全网动态事件主动“推送（Push）”的终极形态。 ## Atmosphere & style 整场对话充满高密度的技术洞见与深度的认知反思，既有顶级投资人与资深技术掌舵者之间的思维碰撞，也透露出褪去大厂光环后重回 Pre-PMF 阶段创始人的清醒与笃定。Parag 在剖析系统架构与博弈论数学模型时条理极其严密、逻辑层层推进，而在谈及公司早年命名为“Shapley Inc.”的趣事与对广告本质的学术热爱时，又展现出幽默而坦诚的工程师底色。对谈节奏紧凑流畅，从微观的 Token 压缩与毫秒级延迟，自然延展到宏观的互联网经济重塑，让听众清晰感受到一场面向“并行互联网”底层基建的范式革命正在发生。 ## Bulletin - `00:00` 前 Twitter CEO Parag Agrawal 创立 Parallel Web Systems，提出“人类点击数据是缺陷”的核心论点，致力于打破传统依靠人类点击流的搜索架构，转向基于大模型压缩与 Agent 任务反馈驱动的底层检索。 - `07:22` Parallel 采用先推长时延 Search Agent 替代高客单价人工外包、后建核心索引的逆向破局策略，并通过 Turbo 产品将 Agent 检索时延大幅压缩至 200 毫秒，且上下文 Token 消耗比传统检索方案减少 50% 以上。 - `24:58` Parallel 已与 Google Cloud（GCP）达成深度产品级集成，作为企业级 Agent API 的底层搜索与 Grounding 基础设施，与传统 Google Search 并列成为核心检索源。 - `34:45` 随着 Agent 消除人类浏览器注意力，传统广告变现模式面临瓦解；Parallel 引入博弈论中的沙普利值（Shapley Value）仿真估算模型进行双向差分定价，预测在未来 12 至 24 个月内将向开放 Web 出版商输送实质性经济回报。 ## Key takeaways - `00:00` 传统搜索引擎视人类点击为核心优化信号，但 Agent 执行复杂工作流需要的是任务有用性反馈，人类的浏览点击习惯对 Agent 系统而言本质上是噪声和 Bug。 - `00:46` 从 Twitter 这种依托数亿高频用户的成熟规模化平台转向早期创业，Parag 必须彻底卸载大厂 A/B 测试惯性，面向尚未成熟但查询频次将是人类千倍以上的全新 Agent 客户群体进行前瞻架构设计。 - `05:00` 大模型在专家级评估上的边际成本断崖式下跌，打破了 Google 和 Bing 凭借海量标注员和点击流垄断的壁垒，让初创团队重新具备挑战全网规模搜索的技术与经济可行性。 - `07:22` Parallel 采取逆向冷启动策略：先推出允许数分钟延迟的 Search Agent 替代金融、保险等领域的人工调研外包以自我造血，在按需动态抓取中积累评估数据并沉淀网页，渐进式建成分秒级底层索引。 - `10:00` Parallel 不做从零预训练前沿模型的 NeoLab，而是定位于基础模型的“力量倍增器”与互补品，聚焦于跨内存层级的毫秒级极端调度，从万亿网页中精准萃取出最高信噪比的约 1000 个 Token。 - `12:30` 传统大模型外挂 Google 会导致上下文充满冗余网页骨架与垃圾 Token；面向 Agent 的专用检索层能将 Agent 上下文 Token 消耗降低 50% 以上，直接突破大模型单次推理的内存与准确率瓶颈。 - `15:10` 人类的懒惰促成了以秒开排版和首屏摘要为主的 SEO 软文与二手聚合内容生态；而 Agent 能提供极高针对性的长 Query，让搜索系统可以直接绕过中介包装，直达 SEC 披露 PDF 等绝对权威源头。 - `23:19` 大模型预训练追求单位算力下的最大 Token 吞吐量，会主动跳过耗时漫长、依赖复杂 JavaScript 渲染的边缘动态页面，因此预训练爬虫语料根本无法顺带替代高时效性的实时 Web 索引。 - `27:38` Agent 引发的千倍搜索乘数源于两层跃迁：单次深度推理时模型内部数十至上千次子检索的裂变，以及完全脱离人类实时触发、不知疲倦监控开放世界的后台常驻 Agent。 - `34:45` 传统互联网依靠广告对高商业价值查询的超额利润来补贴全网免费信息，但当无名无姓的 Agent 取代人类眼球后，注意力变现闭环中断，导致站长既无法追踪转化又本能防御封禁 AI。 - `37:20` 头部大模型厂商与版权方签订的固定总额买断协议无法持续，随推理规模呈几何级激增，版权方单次分成被严重稀释；AI 时代亟需根据内容稀缺度与任务商业价值动态计费的双向差分定价机制。 - `42:30` 针对多源信息浓缩回答的归因难题，Parallel 借助博弈论沙普利值思想，通过测算“缺失某内容后需追加多少推理算力拉平质量”来锚定内容边际贡献，并训练轻量模型克服全排列计算的算力瓶颈。 - `50:00` 从财报电话会到技术 API 文档，数字世界的受众正在发生倒置——内容创作者的第一目标读者正在从人类转向 Agent，未来网页必须同时兼顾人类视觉体验与机器解析结构。 - `52:32` Web 与 Agent 的交互终局将从“即时被动拉取（Pull）”演进为“全网变化主动推送（Push）”，闲置算力将被常态化分配在监控全网动态上，一旦出现特定事件信号即自动唤醒下游 Agent 介入决策。 ## Detailed walkthrough ### `00:00` 人类点击是缺陷：为 Agent 重构搜索 在访谈开场，Parallel 创始人兼 CEO Parag Agrawal 开宗明义地提出了一个打破过去二十年搜索行业常识的激进观点：**人类点击数据本质上是一个“Bug”（缺陷）**。传统搜索引擎将人类在蓝链上的点击反馈奉为圭臬，但对于正在借助搜索来完成复杂工作流程的 Agent（智能体）而言，系统真正需要的是来自 Agent 自身的反馈，而不是人类的浏览与点击习惯。 现代大模型在信息压缩与表征方面展现出惊人能力。大模型前沿研究中的诸多突破，完全可以直接迁移并赋能到搜索引擎的网页索引（Indexing）与排序（Ranking）机制中。Agent 的出现带来了双重转变：它不仅是一项重塑系统架构的底层技术，更代表着一种前所未有的全新“客户形态”。正因如此，构建全网规模搜索引擎这一曾经被视作不可逾越的天堑，如今变成了一个高度可行的工程与商业命题。 ```mermaid flowchart TD accTitle: Traditional Search vs Agentic Search accDescr: 传统人类搜索依靠点击数据与人类标注，而 Agent 搜索依靠大模型压缩与 Agent 反馈驱动。 subgraph Traditional["传统人类搜索范式"] A["数百亿网页抓取"] --> B["预先全量倒排索引"] B --> C["人类用户查询输入"] C --> D["点击流数据 / 人类标注反馈"] D --> B end subgraph Agentic["Parallel Agentic 搜索范式"] E["大模型信息压缩"] --> F["渐进式与动态索引"] F --> G["Agent 智能体查询 (千倍规模)"] G --> H["Agent 任务反馈与真实评估"] H --> F end ``` --- ### `00:46` 从 Twitter CEO 到二次创业：遗忘规模化，面向“尚未到来的客户” 红杉资本（Sequoia Capital）合伙人 Sonya Huang 与 Andrew Reed 欢迎 Parag 登上播客。作为前 Twitter CEO，Parag 在将公司出售给 Elon Musk 之后选择重回创始人轨道，创立了 Parallel Web Systems。Sonya 提问：从掌管拥有数亿用户的成熟巨头，到重新从零做一家初创公司，Parag 需要在认知上进行怎样的“知识卸载”（unlearning）？ Parag 坦言，这两种工作环境的底层逻辑完全相反： - **在 Twitter 担任高管时**：公司处于极度成熟的“后产品市场匹配”（Post-PMF）阶段，规模极其庞大。系统的反馈回路来自数亿名每天深度使用产品超过 30 分钟的高活跃用户，优化依赖于海量真实流量的 A/B 测试与细微迭代。 - **在 Parallel 创业初期**：公司处于“前产品市场匹配”（Pre-PMF）状态。团队的核心前提是一个前瞻性假设——在未来几年内，互联网上将会诞生一种完全不同的新型客户（即 AI Agent），并且这群客户使用互联网的频率和广度将比人类历史上高出上千倍（1000x）。 在目标客户群体尚未完全成型时，团队必须抛弃依赖海量即时反馈的“大厂惯性”，学会在缺少现成大盘数据的前提下，每周自我迭代并验证底层架构。 --- ### `03:07` 拆解搜索本质：十亿级对十亿级的匹配挑战 播客新主持 Andrew Reed 抛出核心问题：硅谷大部分人对搜索的理解仅停留在“抓取、索引、排序并给出答案”的表层认知。从第一性原理来看，无论服务对象是人类还是 Agent，网络搜索的本质到底是什么？ Parag 将网络搜索定义为一个**“十亿级对十亿级（billion to billion）的极致匹配问题”**： 1. **网络抓取（Crawling）**：追踪并访问全网每一个可公开触达的 URL，读取其内容。 2. **构建索引（Indexing）**：将网页信息进行结构化和内存化组织。索引本质上是一种**延迟优化策略（latency optimization）**——通过预处理和就地常驻内存，确保查询到来时无需实时耗时加载远程网页。 3. **多阶段检索与排序（Retrieval and Ranking）**：当用户或 Agent 提出查询时，系统必须迅速解析意图，从数百亿、数千亿个网页的浩瀚空间中层层筛选收敛，最终精准缩减到 5 个、10 个乃至最核心的 1 个绝佳结果。 --- ### `05:00` 破局双寡头：打破 Google 与 Bing 的基建壁垒 面对 Google 和 Bing 统治了二十多年的全网规模爬取与索引壁垒，Andrew 质疑：这样重资产、高成本的基础设施，初创企业凭什么敢于入场竞争？ Parag 回顾三年前创办 Parallel 时的核心思考。过去挑战者无法匹敌巨头，核心瓶颈不在于写不出代码，而在于**数据飞轮缺失**： - **人工评估（Human Ratings）成本极高**：判断“结果 A 是否比结果 B 更优”需要雇佣海量外包标注员； - **点击数据（Click Data）被巨头垄断**：没有搜索大盘流量，就拿不到人类点击信号来训练排序模型。 但生成式模型彻底改变了博弈规则： - **评估成本断崖式下跌**：基于前沿大模型构建的“专家级评估”，能以极低廉的成本大规模合成高质量训练信号。 - **Agent 本身就是客户**：由于系统的最终使用者是 Agent，系统需要的是 Agent 在执行复杂任务中反馈的“有用性”，而不是普通人浮躁的点击流。既然“人类点击数据是 Bug”，初创公司便无需再追逐巨头把持的人类点击日志。 - **渐进式建库（Incremental Indexing）**：初创公司绝不能在第一天未服务任何客户前，就贸然把全部资金砸入全网静态爬取与基建中。关键在于设计一条让索引伴随客户业务价值自然生长的演进路径。 --- ### `07:22` 逆向破局策略：先做 Search Agent，后做搜索引擎 针对“客户第一天就要求全网覆盖”的悖论，Parag 透露了 Parallel 极具逆向思维的冷启动战术：**团队在上线搜索引擎产品之前，率先推出了一款 Search Agent（搜索智能体）产品。** Parag 指出，索引归根到底是对“延迟”的妥协。如果放弃传统搜索即时响应（百毫秒级）的限制，直接与替代方案中的“人类外包团队”竞争，战局就会完全改观： - 当用户要求执行“深度研究”（Deep Research）时，往往具备数分钟甚至 10 分钟以上的耐心。 - 在拥有充足时间的前提下，只要拥有合理的网页优先级调度图谱，Agent 可以在**接收到查询之后实时按需爬网**，从而绕开静态预建全网索引的资本门槛。 - 与其在第一天硬碰 Google，不如先在“高质量人工外包信息整理”这一生态位展开竞争——因为人类对着传统搜索框手动挖掘整理数据的效率，远比早期搜索智能体更容易被击败。 ```mermaid flowchart LR accTitle: Cold-start Incremental Indexing Strategy accDescr: Parallel 先推出 Search Agent 解决长耗时深度研究任务，在实时抓取中积累评估数据并沉淀网页，渐进式建成大规模索引。 Query["客户深度查询请求<br/>(允许1~10分钟延迟)"] --> Agent["Parallel Search Agent<br/>(按需动态网页抓取)"] Agent --> Replace["替代传统外包人工<br/>(保险核保/数据补全/金融分析)"] Replace --> Value["为客户交付精准结果"] Agent -. 沉淀评估反馈与优质网页.-> Index["渐进式沉淀规模化索引<br/>(为秒级底层搜索引擎铺路)"] ``` 在产品初期，各大企业迅速将 Parallel 的 Search Agent 嵌入到了过去重度依赖人工的复杂业务流中： - **保险核保与理赔（Insurance Underwriting & Claims）**：自动化调取并核实分散在公开网络中的企业资质与风险记录。 - **销售数据富化（Sales Data Enrichment）**：针对潜客背景、公司动态进行跨站自动化调研。 - **金融建模数据抓取（Financial Modeling）**：以往对冲基金或金融机构需要将数据需求连夜分发给离岸人工团队进行搜集，次日再输入模型；使用 Parallel 后，整套数据收集流程由 Agent 自主在全网挖掘完成。 通过切入这些重度依赖网络数据的人工外包环节，Parallel 不仅实现了早期的商业化自我造血，更在真实严苛的生产环境中源源不断地积累了高质量的模型评估（Evals）数据，为其最终演进为底座级搜索引擎奠定了基石。 ### `10:00` 不是 NeoLab，而是大模型的“力量倍增器” 在外界看来，Parallel Web Systems 汇聚了顶尖的系统工程师与大量 AI 研究员，这让主持人在初次接触时感到疑惑：这究竟是一家传统的 Web 系统架构公司，还是一家当前定义下的“NeoLab”（新型前沿 AI 实验室）？面对这个标签，Parag Agrawal 明确表示 Parallel 并不将自己归类为 NeoLab。在他看来，NeoLab 的核心产出物是一个个基础大模型，而 Parallel 的产出物则是“模型的互补品（complement）与乘数（multiplier）”。他们致力于让基于模型构建的 Agent 获得超能力；当其他实验室推出更强大的底层模型时，NeoLab 之间是在攀登更高的技术壁垒相互竞争，而对 Parallel 而言，每一次底层模型的突破，反而是为其技术解锁了三到四个全新的落地场景。 Parallel 并非不涉足前沿研究，但其研发重心绝非从零预训练超大规模模型，而是专注于极致的压缩、排序与系统级调度。Parag 形象地指出，现代搜索本质上是一个在万亿网页海啸中捞针的超大规模匹配难题——每一个 Agent 查询的潜台词其实都是：“请从全网一万亿个网页中，精准找出最对的那 1000 个 Token”。过去，系统在读取时段（Read time）可以奢侈地消耗整整 3 秒计算资源来筛选；而随着团队刚刚发布的 Turbo 产品，这一全流程时延被硬生生压缩到了令人惊叹的 200 毫秒。研发人员必须在极度苛刻的时延与算力预算下，重新调整预训练模型架构、精巧安排不同层级的内存体系，并训练极精简的排序模型来完成这千枚 Token 的提取。 - **定位差异**：NeoLab 追求以模型本身为产品终端，而 Parallel 追求与任何先进模型正交互补，做模型之上的效能放大器。 - **核心搜索命题**：每一轮 Agent 检索的实质，是从全网万亿级网页中提纯出最契合上下文的 1000 个 Token。 - **性能飞跃（Turbo 产品）**：从原先消耗 3 秒读取时效，直接缩短至 200 毫秒级别，对跨内存层级（Memory hierarchy）的高速访问与精简排序模型提出了极高要求。 --- ### `12:30` 为什么“Deep Research 套壳 Google”注定不够用？ 当前业界常见的“深度研究（Deep Research）”范式，本质上是让大模型在一个 Agent 循环中进行推理推导，遇到信息缺口时便调用一次 Google 或其代理接口。主持人提出疑问：这种方案究竟有何致命缺陷，为什么专门面向 Agent 的搜索是一项彻头彻尾的“全新能力”？Parag 坦言，调用 Google 确实能解决一部分问题，但对于严肃的 Agent 工作流而言远远不够。如果改用 Parallel 的 Agent 搜索基础设施，Agent 在运行中所消耗的 Token 数量往往能直接减少一半以上，同时在推理准确度与响应速度上实现双重提升。 在信息获取与推理中，一切终究会归结为质量、成本与延迟的三维权衡。大模型技术的演进一方面在挤压出更高的智能上限，另一方面在通过蒸馏将成本压低至十分之一。而搜索作为 Agent 的认知外脑同样遵循这一规律。当 Agent 受限于上下文窗口长度或内存开销时，Parallel 能够在搜索结果中提供极高信噪比的信息片段，剔除冗余干扰。“每一次你能在搜索结果中向模型输出全部有效信号并大幅降低噪声，你就赋予了模型做更多事情的能力。”这背后的真正赌注，是跨越基础模型层、Agent 调度层与底层搜索层，实现全链路的“智能算力动态分配”。 ```mermaid flowchart TD accTitle: Deep Research 与 Parallel 检索流对比 accDescr: 对比传统大模型套壳 Google 与使用 Parallel 专用检索层在 Token 消耗与信噪比上的巨大差异 subgraph Legacy["传统模式: 套壳 Google 搜索"] A["Agent 推理循环"] --> B["调用传统搜索引擎 API"] B --> C["抓取混乱网页与冗余文本"] C --> D["塞入海量噪声 Token<br/>(高延迟 / 易爆上下文 / 高幻觉)"] D --> A end subgraph ParallelFlow["Parallel 模式: 专为 Agent 优化的检索层"] E["Agent 高信息量 Query"] --> F["Parallel 智能检索系统<br/>(200ms 极速响应)"] F --> G["精准提炼权威 Token 片段<br/>(剔除网页噪音与广告干扰)"] G --> H["注入精简纯净上下文<br/>(Token 消耗减半 / 准确率提升)"] H --> E end ``` - **Token 减半效应**：Parallel 搜索结果剥离了无关网页骨架与噪声，Agent 所需处理的 Context Token 数量可降低 50% 以上。 - **上下文扩容复用**：节省下的上下文预算让有限内存的模型能够突破单次推演瓶颈，以更低成本应对复杂度呈倍数增长的问题。 - **全栈算力分配**：不再将检索视为黑盒，而是将计算资源合理分配在模型推理、Agent 决策和搜索匹配三大环节之间。 --- ### `15:10` 人类的“懒惰”催生了网页垃圾，而 Agent 终结了 SEO 套路 人机交互界面的本质差异决定了搜索底层架构的重构。Parag 调侃道，人类并不是脑容量不足以容纳长句，而是“本质上太懒了”。用户在向 Google 输入关键词时，恨不得只敲两个打错的词根，然后全靠下拉框自动补全来逃避输入；而大模型驱动的 Agent 完全相反，它们几乎不会拼错单词，能够清晰、详尽地给出极具针对性的长 Query，搜索系统再也不必费尽心机去“猜”用户到底想要什么。 谈及充斥在各大搜索引擎前列的 SEO 软文、返利联盟广告（Affiliate advertising）与 AI 生成垃圾（Slop），主持人笑称自己虽身经百战却依然屡屡中招。Parag 深入拆解了这类内容之所以合法存在的人性逻辑：以查询一家上市公司的最新财报营收为例，最权威的数据必然深埋在 SEC（美国证券交易委员会）官方披露的数百页 PDF 中（例如第 73 页）。如果 Google 盲目追求绝对权威，将这份 SEC 原始 PDF 排在首位，普通用户点击后往往要忍受长达 3.5 秒的漫长加载，接着还必须痛苦地翻找定位。人类的懒惰促成了一个庞大的中介内容生态——有人专门把权威数据扒下来，做成秒开的网页，把核心数字放在首屏（Above the fold）。这种内容哪怕被戏称为“人类垃圾内容”，却精准满足了懒惰人类的即时需求，从而在传统 SEO 机制中大获全胜。 然而，Agent 的行为逻辑打破了这一切。Agent 不需要靠好看的前端排版与首屏摘要来救赎，更不必在笨拙的浏览器里胡乱点击摸索。Parallel 所做的事情，是直接定位并截取全网最权威源头（如官方备案原件）的关键文本段落，绕过所有中间商与 SEO 干扰层，精准投喂进 Agent 的 Context Window 中。 > “在面对 Agent 时，我们再也不必被逼着接受那种妥协。过去之所以必须妥协，是因为在标准的浏览器和网络协议限制下，我们根本没法稳定、精准地在整个互联网上将用户直接导向某一段高亮的极小文本段落；但面对 Agent，我们能把这一小段核心信息直接带入它的上下文，剩下的推理由它自己搞定。” - **查询界面的根本转变**：人类依赖残缺关键词与容错推测；Agent 提供零拼写错误、高度具体且富含语境的长提示。 - **SEO 垃圾页面的经济学成因**：由于人类拒绝翻阅 300 页厚重的财报 PDF，催生了将深层数据抓取并置于首屏的高速二级网站。 - **权威直达机制**：Parallel 规避了“用户友好度 vs. 权威度”的历史妥协，直接跨过中间包装页面，将源头事实摘录注入大模型。 --- ### `19:00` 解析一次 Agent 检索调用的底层逻辑 当开发者向 Parallel 的 Agent 搜索 API 发送一次检索请求时，系统并不会像传统引擎那样机械地进行关键词匹配。首先，系统会调度专有模型对该查询进行深度的意图解析与上下文丰富（Query Enrichment），精准推断出 Agent 真实的下游推理意图，从而决定请求在后续复杂体系中的路由分发方式。而在底层的存储与检索层面，Parallel 并没有依赖单一扁平的巨型索引库，而是维护了一组针对互联网不同领域、不同数据子集专门构建的高度垂直的差异化索引矩阵。系统依托内存分层架构展开极速多路召回，为后续在毫秒级时延内精挑细选出最关键的信息 Token 奠定底层支撑。 ### `20:00` 面向 AI 的多层检索架构与极致信号提取 Parag Agrawal 详细拆解了 Parallel 针对 AI Agent 重构的搜索技术栈。在这个架构中，第一层系统会针对不同的底层子系统分别定制和重写查询（Query Rewrite）。这些底层系统分工明确：一部分是海量的主索引（Big Index），一部分是追踪即时动态的实时索引（Fresh Index），还有一部分是知识图谱或结构化索引。当一个 Agent 发起请求时，系统首先需要决定将查询分发给哪些子系统，并针对每个索引的特性进行查询改写。 ```mermaid flowchart TD accTitle: Parallel 多层搜索架构示意图 accDescr: 展示从查询分发、多级检索引擎到多轮重排，最终提炼出1000个高信号Token的完整数据流。 subgraph Dispatch ["查询理解与分发层"] Q["Agent 原始查询"] --> QR["查询重写与路由"] end subgraph Subsystems ["底层专用索引集群"] QR --> BI["海量主索引 (Big Index)"] QR --> FI["实时索引 (Fresh Index)"] QR --> KG["结构化图谱 (Knowledge Graph)"] end subgraph RankingPipeline ["多级过滤与精排流水线"] BI & FI & KG --> R1["粗排召回<br/>(数百亿网页 ➔ 数万篇候选)"] R1 --> R2["细粒度段落提取与深度重排<br/>(逐级调用更大模型与丰富特征)"] R2 --> Final["高信号压缩与输出<br/>(提取最核心的 ~1000 Tokens)"] end ``` 接下来的流水线则是一个极高强度的信息压缩过程：从数百亿乃至上千亿的网页和文档中，系统首先通过检索层筛选出数万篇候选内容；随后在更精细的排序层中，由逐级变大、架构各异的模型介入，结合更丰富的特征，从这几万篇文档中定位并抠出具体的段落与片段。整套庞大计算流水线的最终目标极其明确——为调用端的大模型带回信号密度最高、最纯净的约 1000 个 Token。Parag 解释道，Parallel 提供的不同版本搜索 API，其本质区别就是在满足不同延迟（Latency）与成本（Cost）约束的前提下，在流水线的各个节点动态分配不同规模的算力。 - **抽象概念与落地细节**：Parag 直言“在抽象层面，这套架构听起来很简单”，但真正的技术壁垒在于每个专用模型背后的训练数据如何收集与清洗、模型如何针对推理进行极致调优，以及索引本身如何利用现代硬件的存储层级（Memory Hierarchy）来实现成本、质量与毫秒级延迟之间的最佳平衡。 --- ### `21:40` 十亿对十亿的“永恒匹配难题”与质量北极星 当被问及 Parallel 在搜索质量上的“北极星指标”——是否类似于传统 Google 时代“用户是否在前三条结果中找到答案”的标准时，Parag 表达了更深刻的算法哲学。他认为，“十亿级需求与十亿级信息的匹配问题是一个永恒问题（a billion to billion matching problem is a forever problem）”。在这个领域，永远不存在所谓“大功告成、无需再优化”的终点，真正的权衡仅在于边际优化的收益何时会低于付出的算力成本（not worth the squeeze）。 在 Parag 构想的超前 AGI 世界观中，这个终点甚至可能永远不会到来。因为随着底层基础模型的智能水平不断提高、推理成本每隔几个月就大幅下降，未来的核心竞争力将彻底聚焦于三点：拥有廉价且高智能的模型、拥有最优质的信息输入，以及在这两者之间完成近乎完美的供需匹配。“无论是企业内部私有数据还是广袤的开放网络，只要你能把需求与信息连接起来，并做出独一无二的深度洞察，哪怕只比别人好上 20%，就能形成压倒性的竞争壁垒。既然如此，为什么不全力以赴去推高上限呢？” --- ### `23:19` 预训练爬虫与实时索引的本质分歧：为什么大模型厂商无法顺带吃掉搜索 红杉资本的主持人提出了一个行业内广泛存在的论调：网络搜索对于模型公司而言是如此底层且关键的能力，加之预训练阶段本就需要耗费巨资抓取全网数据作为事实来源（Source of Truth），模型厂商理应顺理成章地将搜索基础设施内化。Parag 对此给出了坚决的反驳，指出这种观点完全脱离了实际工程现实：“从现实中的工程数据来看，根本不支持这种假设。用于预训练的爬虫数据，对于构建一个高时效性的实时网页索引几乎毫无用处。” Parag 深入对比了两者的核心差异：Parallel 的定位是成为基础模型的“互补品”（Complement）。大模型在预训练时追求的是吞吐量，关注的是单位算力下能获取的最大 Token 数量，因此极度缺乏耐心去等待那些加载极其缓慢、依赖复杂 JavaScript 渲染的边缘网页，因为这类网页的算力产出比可能要低上一到两个数量级。相反，Parallel 必须是一个“完美主义者与穷尽者”（Completionist）——越是预训练模型嫌麻烦跳过、由于未曾收录而在知识库中留下盲区的动态网页，Parallel 越需要花费高昂代价去精确渲染与索引。如果一个信息早就被预训练大模型熟记于心，实时搜索再去检索它就失去了最大的价值。 --- ### `24:58` 拥抱巨头：携手 Google Cloud 落地企业级 Agent 搜索与 Grounding Parag 指出，在企业采购大语言模型用于实际业务的场景中，十有八九都需要赋予模型访问外部网络的能力，并配备专门针对 Agent 优化的搜索基础设施。因此，搜索已成为大模型推理（LLM Inference）最自然的延伸与强邻近业务。既然模型公司需要最顶级的搜索能力，他们就面临着经典的“自研还是采购合作（Build vs. Buy）”的选择。 在此背景下，Parag 透露了一个重磅合作进展：Parallel 已与 Google Cloud（GCP）达成深度产品级整合，正式成为其企业级 Agent API 的搜索与事实锚定（Grounding）服务商。这意味着在 Google Cloud 上构建 Agent、聊天机器人或调用 Gemini 及其他开源模型的企业客户，在连接网络搜索时拥有了两个核心选项：Google Search 或是 Parallel Search。 - **与“搜索之王”的戏剧性共存**：作为由前 Twitter CEO 创立、挑战传统搜索范式的初创公司，Parallel 的产品不仅直接打入了“搜索之王”Google 的云计算生态，两家公司的技术、训练、产品和商业化团队还进行了深度调优，确保 Gemini 模型搭配 Parallel 检索时能够达到顶尖效果。 - **关于厂商属性的讨论**：面对主持人“拿到搜索之王背书意义非凡”的惊叹，Parag 幽默且克制地表示，当今的 Google 既是原生模型实验室又是超大规模云厂商（Hyperscaler），很难单纯定义这次合作代表了模型厂商的意图还是云巨头的生态策略，但无论如何，这标志着新型 AI 搜索合作范式的开启。 --- ### `27:38` 从人机交互到自主监控：搜索量暴增千倍的乘数效应 讨论随后转向了核心命题——所谓“Agent 带来的搜索需求是人类千倍以上（1000x more）”究竟是如何发生的？究竟是单次研究的深度加深，还是触发搜索的主体本身发生了质变？Parag 剖析了两个层面的乘数效应（Multipliers）： 第一层乘数来自单次人机交互内部的深度展开。即便不启动耗时漫长的“深度研究”（Deep Research）模式，一个响应时间仅有数秒的常规搜索 Agent，为了回答一个看似简单的问题，在后台也会并行或串行发起 5 到 20 次检索；如果是在复杂推理任务中，单次 Prompt 触发的检索次数会直接跃升至数百乃至数千次。人类一次提问，在模型层面就被放大了一个数量级以上的搜索请求。 第二层也是量级更庞大的乘数，则来自于彻底脱离人类实时触发的“后台常驻 Agent”（Background Monitoring Agents）。Parag 分享了一个真实的企业级客户案例： 某金融信贷机构拥有一个包含 10,000 家小微企业的贷款资产组合。在以往，公司需要依靠繁重的人工流程，每月逐个排查这些企业的网络动态以评估信用风险的起伏。而在引入自主 Agent 之后，开发者只需编写一套工作流，这群无休止监控真实世界与开放网络的 Agent 就会在每个周期自动爬取、交叉比对海量动态信息。在这样的场景下，人类单次启动或常规维护，对应的是几十万甚至上百万次的底层网页检索调用。Agent 正在从根本上重塑整个数字世界的信息吞吐规模。 ### `30:00` 后台智能体驱动的搜索爆炸与 Turbo 的诞生 后台自主运行的智能体正在彻底重塑信息获取的频率与搜索规模。以往由人工每月梳理一次的被投企业全景看板，如今完全可以交给智能体每周自动执行：自动发起海量搜索，汇总生成投资组合仪表盘，并提炼出一整套待办行动项。在日常办公中，会议准备也是极具代表性的场景。红杉资本内部拥有自己的智能体工具（红杉合伙人们开玩笑提到“以前这活都归我们优秀的年轻分析师 James Flynn”），而 Parag Agrawal 则深度依赖 Notion 的智能体。他通过一次性编写 Prompt 构建定制智能体，将 Parallel 的内部数据与 Parallel Web Systems 的外部搜索 API 打通，每次开会前该智能体就会自动发起数十甚至上百次网页搜索以生成背景调研文档。随着每建立一个新用途的智能体，搜索量就会成倍级联扩散。 这种后台智能体不知疲倦运转的上限，目前仅受限于“产出价值与算力支出”的权衡。Parag 坦言，现阶段大家在智能体上的算力花费或许还不够理性，但只要其带来的增量价值足够明确，行业就会持续部署更多的后台运行程序，搜索调用量也将随之飙升。针对这种爆发式的查询需求，Parallel 在创立最初几年制定了清晰的工程演进路线：在“质量（Quality）、成本（Cost）、延迟（Latency）”三个核心维度中，他们选择前两年完全忽略延迟，集中火力攻坚质量与成本。因为对成熟工程系统进行优化以及将大模型蒸馏为小模型属于“已知的工程技术路径”，而模型检索效果本身则是更艰难的未知探索。当 Parallel 在各个价格区间都做到了最高质量的搜索效果后，团队迅速转入延迟攻坚，最终推出了名为“Turbo”的产品，成为市场上响应极快、质量极高的智能体网络搜索方案。 - **工作流裂变**：原本一个月一次的手动分析被压缩为每周自动例行执行，每次运行自动发起海量检索并输出决策仪表盘。 - **会议准备倍增效应**：单一会议的前期调研即可触发上百次后台静默搜索，随着自定义智能体数量增多，查询规模呈指数级递增。 - **三维工程策略**：创业早期战略性放弃延迟指标，优先解决检索质量与推理成本，依靠模型蒸馏成熟后再针对延迟突破，打造出极低延迟的 Turbo 检索系统。 --- ### `32:50` 极早期的普及阶段：人与智能体的查询比例 尽管智能体引发的查询乘数效应惊人，但 Parag 强调整个行业仍处于智能体普及的极早期阶段。“只要走出科技圈的信息茧房，大众甚至连很多基础名词都没听说过。”回顾 ChatGPT 出现之前，普通人每天在 Google 上的搜索量通常不过 20 到 30 次；而如今像 Parag 这样深度使用各类后台智能体的人，其全天调用的搜索总量已然达到过去的 100 倍至 1000 倍。若将公司内部没有明确指派给具体个人的后台批处理任务全部计入，调用规模突破千倍更是轻而易举。但这类用户目前依然是极少数的先锋群体。 从宏观网络流量来看，分水岭已经初见端倪。正如 Cloudflare 近期的网络监测数据显示，互联网上 AI 产生的流量在网页读取量（Page Reads）维度已经与人类流量旗鼓相当。尽管网页读取数据中包含了大量的网络爬虫程序和机器人，与最终用户的显式搜索尚不能完全画等号，但智能体全面主导互联网流量只是时间问题。这种访问主体的剧变，正在动摇整个现代互联网赖以生存的根基。 ```mermaid flowchart TD accTitle: AI 流量增长与搜索需求倍增模型 accDescr: 展示人类传统搜索模式与后台智能体高频并发检索之间的调用量级对比 subgraph Human["传统人类模式（基准量：20-30 次/天）"] H1["人工意图驱动"] --> H2["手动单次检索"] H2 --> H3["浏览器展示点击"] end subgraph Agent["智能体倍增模式（100x - 1000x 扩张）"] A1["后台任务触发"] --> A2["多智能体协作分解"] A2 --> A3["高频并发 API 搜索 (Turbo)"] A3 --> A4["信息聚合与自主决策"] end Human -.->|"范式转移"| Agent ``` --- ### `34:45` 广告驱动的互联网经济体系正在瓦解 作为曾经深度参与广告系统搭建的前 Twitter CEO，Parag Agrawal 对网络广告有着截然不同的学术与商业理解。他坦承：“我知道大多数人都讨厌广告，但我从理智上非常热爱广告，因为正是广告让如此多不可思议的内容和技术免费向全人类开放。”如果没有广告的强力支撑，Google 搜索不可能免费，Twitter 也不可能免费，开放互联网上海量的优质信息更不可能零门槛获取。广告本质上是一种极度高效的“差别定价机制（Differential Pricing）”：平台在绝大多数无商业转化意图的长尾搜索查询上都在赔钱，但凭借少数高商业价值查询所带来的暴利，维系了高额利润率，进而无形中补贴了全网的免费信息流转。 然而，整个广告经济的底层基石建立在“人类稀缺的注意力转化为商业结果”这一假设之上。当人类不再亲自浏览网页，取而代之的是冰冷无情的智能体时，这套逻辑立刻陷入瘫痪： 1. **商业对齐断裂**：网站站长原本依靠 1000 次人类访问转化 20 个月度付费订阅会员。但面对无名无姓的智能体爬取，站长完全无法建立转化统计模型，不知道它们最终是否会带来购买，只看到内容在被持续调用。 2. **两难的防御阻断**：站长一方面花费重金做 SEO 吸引人类访客，另一方面却本能地选择封禁智能体访问，唯恐自己的商业化内容被“白嫖”。这造成了极大的割裂——因为智能体背后代表的依然是有需求的人类，但旧时代的商业闭环却无法对智能体的这次访问完成变现。 --- ### `37:20` 头部买断协议的困局与 Parallel 的差异化定价解法 在 Parallel 介入之前，高价值内容版权方在 AI 时代几乎只有华山一条路：挤入互联网内容金字塔的最顶端，与头部前沿大模型实验室（AI Labs）签订包含训练授权、法律兜底以及部分推理访问的固定金额买断协议（Fixed-fee Contracts）。这种模式不仅将绝大多数非头部的内容创作者完全拒之门外，即使对于顶尖媒体机构而言，它也是一种无法持续的病态商业结构。 Parag 指出，随着大模型推理调用规模呈现几何级数爆发（今年增长 7 倍，明年再增长 7 倍，累计激增 50 倍），内容版权方手里的固定授权费却分文不涨。任何一家签了两年买断协议的出版商，都心知肚明在合约到期续约时，自己的价值分成必然会被大模型公司大幅稀释甚至大幅砍价。这种静态的价格契约完全违背了 AI 推理规模扩张的客观规律。 Parallel 的解题思路直接脱胎于 Parag 过往的广告变现与数据交易实战（包括他当年主导向 OpenAI 出售 Twitter 数据资产的经验）。一个可持续的全新经济协议必须具备以下核心机制： - **双向差分定价（Two-dimensional Differential Pricing）**：既要根据内容本身的稀缺品质给出差异化高价，更要根据智能体调用该内容所执行任务的“商业价值高低”实行动态计费——价值极高的商业推理调用必须支付极高的数据费用。 - **无感与规模化（Scalability）**：彻底摆脱耗时费力的一对一线下商务谈判，建立一套自动接入、细粒度核算并规模化分发收益的底层清结算基础设施。 ### `40:00` 大模型厂商的付费动机与内容归因难题 大模型厂商目前为何愿意向内容版权方付费？Parag Agrawal 坦言可以去直接问那些模型公司，但根据他的观察，背后的驱动力主要由三部分构成：其一是在推理时获取最新鲜的数据（Fresh Data），以便在 ChatGPT 或 Claude 等面向终端用户的产品中实时呈现；其二是作为基础模型训练所需的预训练语料；其三则是为了给之前已经完成的抓取与模型训练提供法律责任保护（Liability Protection）。厂商给出的报价往往是这三者以某种比例打包的组合，但外界很难确切得知他们对每一项的具体估值。 此时主持人提出一个关键矛盾：在一个极其复杂的搜索场景下，哪怕是一个看似简单的用户提问，Agent 背后也可能执行多达 10 次子检索，最终浓缩提炼为一段仅有一段文字的回答。在这一长串链条中，究竟该如何量化每个信息源的增量贡献？ Parag 笑着表示：“在 Parallel，我们最喜欢做的事就是构建模型。”但他强调，在动手建模之前，必须先从理论和认知上厘清什么是“激励对齐”（Incentive Alignment）。如果纯粹从思想实验出发去测量增量价值：假设把某一个内容源从语料库中强行剥离，再运行 Agent 去评估输出质量下降了多少。如果为了弥补缺失该内容造成的质量损失，我们投入更多算力或调用更好的模型，刚好能把质量拉平——比如这多耗费了 1 美分的算力成本——那么从最终用户价值的角度来看，两者的质量体验是一致的。这意味着，该内容源所蕴含的价值在直觉上就约等于 1 美分。 ```mermaid flowchart TD accTitle: 基于替代算力成本的内容价值评估逻辑 accDescr: 展示当缺少某个内容源时，通过追加算力弥补输出质量，以此反推该内容价值的归因机制。 subgraph Evaluation ["归因仿真评估"] A["从语料库移除特定内容源"] --> B["运行 Agent 生成回答并评估质量降幅"] B --> C["投入增量算力 / 调用高阶模型弥补质量"] end subgraph Valuation ["边际价值锚定"] C --> D["测算弥补质量所需算力成本（如 1 美分）"] D --> E["确立该内容源的等价边际贡献值"] end ``` --- ### `42:30` 博弈论切入：沙普利值（Shapley Value）与估算模型 为了将这种“增量价值”的直觉严密公式化，Parallel 采用的核心数学框架正是博弈论中的“沙普利值”（Shapley Values）。主持人惊呼这简直是“耳畔天籁”，因为这触及了深层的博弈论与机器学习特征归因（如 SHAP）。沙普利值专门解决一个理论难题：假设多人合作创造了“1+1+1>3”的超额价值，如何将这个更大的蛋糕公平分割给各方，才能让所有人都有持续合作的激励。 然而，沙普利值在现实中最致命的痛点是计算复杂度极高：它需要模拟所有可能的合作子集排列组合。在创业公司分股权时这显然不现实；但在机器学习中，人们可以通过剔除特征来评估重要性。对于 Web 搜索，Parallel 同样可以进行大规模离线仿真——模拟剔除特定 URL、域名或特定内容集合后 Agent 的表现差异，建立全面的评测（Evals）数据，再训练机器学习模型去逼近真实沙普利值。 Parag 指出，如果对每一次 Agent 查询都做完备的理论沙普利值全排列计算，算力花销可能要好几美元，远超过付给出版商的 1 美元甚至 Agent 本身的运行成本。因此必须依靠精准的估算模型。只要信息对称，各方就会形成合作的稳定机制。这如同早期互联网广告的次价拍卖（Second-price Auction）——博弈论设计让广告主敢于披露真实出价，随着市场建立反馈闭环并演进为基于 ROI 的自动竞价（Autobid），系统最终达成了极高的效率。沙普利理论驱动的机制同样旨在让内容所有者与 AI 内容检索者之间形成可持续的自洽生态。 - **沙普利值的核心意义**：在多方协同贡献总价值超越个体总和时，以公理化方式分配利益，保证所有参与者的激励相容。 - **算力现实妥协**：全量排列组合计算成本过高，Parallel 的解法是利用仿真数据训练轻量预测模型，对价值边际进行高精度估算。 - **与广告拍卖的相似性**：就像搜索广告从手工出价走向以 ROI 为核心的自动竞价，内容结算体系也需要通过机制设计引导真诚合作。 --- ### `45:00` 宏观经济账本：12 至 24 个月内向出版商输送真金白银 向传统内容版权方讲解博弈论和沙普利数学固然需要花点时间，但只要抽离出底层逻辑，出版商就能迅速理解： 1. **差异化溢价**：内容越独特、越具有不可替代性，在模型归因中分到的报酬就越高； 2. **场景价值传导**：一名从事高薪金融工作的银行家检索使用数据，与一位退休老人阅读消遣内容，由于前者属于高商业价值工作流，其背后产生的信息调用付费会高出许多。 从宏观经济体量来看，这套逻辑完全站得住脚。未来知识工作中消耗在 LLM 推理上的资金将极其庞大；只要将其中的 2% 到 10% 重新分配给 Web 上的数据供给端，其绝对金额就已经远超除 Facebook、LinkedIn 等围墙花园（Walled Gardens）之外的所有开放 Web 现有商业模式总和。 面对 Stack Overflow 流量断崖式下跌、传统人类互联网内容生产激励崩塌的现状，Parag 展现出了明确的乐观预期：伴随 Web 上 Agent 的调用量呈现年均数量级（Order of Magnitude）的爆发式增长，根据他的推算，**在未来 12 到 24 个月内**，这套基于沙普利机制的估算模型就将能够向广泛的开放 Web 内容所有者输送极具实质意义的真金白银。 --- ### `47:30` 从“Shapley Inc.”到“Parallel”：双重受众与并行互联网 在谈到公司名称的由来时，Parag 透露了一个鲜为人知的内幕：Parallel 的原始注册名称其实就叫“Shapley Inc.”。主持人和投资人对此哈哈大笑。Parag 回忆，创立之初他整个人完全沉迷在这个博弈论问题空间里，即使知道“Shapley Inc.”作为一家面向 B2B 的商业产品来说是个很糟糕的名字，他还是硬着头皮先注册了。当时 Shapley.ai 还能买到，但 Shapley.com 停放着一个卖鱼骨的域名页面，显然无法作为长期品牌。在那段时期，Parag 出席各种行业活动的名牌上写的都是“NewCo”或“Stealth Company”（隐形公司），绝口不提 Shapley。 团队花了六到八个月才最终敲定了“Parallel”这个名字，主要源于两个顿悟： - 一是工程技术层面，当时系统内部正在将大量串行流程全面改为“并行化”（In Parallel）处理； - 二是他们具象化地意识到，一个**专为 AI 打造的并行互联网（Parallel Web）**正在浮出水面。 未来的内容发布者在写下网页时，脑海中将不得不面对截然不同的两批受众（Dual Audiences）：一批是人类读者，需要考虑排版、视觉和情感体验；另一批则是 AI Agent，需要考虑机器如何解析结构、提取语义。主持人也深有共鸣地补充道，类似企业财报电话会纪要（Earnings Transcripts）这类高价值信息，现实中通过 Agent 过滤消费的人数已经远远甩开了实际去听音频或硬啃逐字稿的人类。网络上许多最具商业价值的切片，实际上已经率先翻转为“Agent 优先”（Agent-First）的生态。 ### `50:00` 为 Agent 写作：从财报发布到 API 文档的受众倒置 Parag Agrawal 指出，虽然当前世界还没有完全完成转变，但信息传播的核心逻辑已经开始颠覆。他打趣道，如果自己今天是一家上市公司的 CEO，在做财报电话会议（Earnings Call）时，最关心的绝对不仅是电话线另一头听讲的人类分析师，而是必须确保自己说出的每一句话，都能被自动化 Agent 准确转录并正确理解。因为真实世界中做出买卖决策和信息分发的，正迅速变成各种 AI 智能体。 这种转变在 Parallel 自身的日常运营中更是早已成为现实。Parag 提到，Parallel 在编写和发布 API 文档及技术说明时，脑子里设定的目标读者压根不是“在网页上到处乱点、手忙脚乱查找文档的人类开发者”，而是开发者用来构建 AI 解决方案的 Agent 本身。这些 Agent 在读 Parallel 的文档、调他们的接口并集成 SDK。因此，Parallel 测试和评估自身文档好坏的方式，就是直接拿 Agent 来跑，看 Agent 能否无缝理解并执行。 - **受众倒置（Audience Inversion）**：从上市公司财报到底层技术文档，第一消费群体正从人类视听者转向自动化大模型与 Agent。 - **文档自动化测试**：以 Agent 为“第一公民”进行文档评估，确保机器能够零歧义理解接口规范与代码范例。 --- ### `51:15` 智能体演进三阶段：从单点工具到长程多 Agent 协同 在被问及 Parallel 目前所处的阶段以及未来世界的全貌时，Parag 梳理了智能体演进的“三层阶梯”。 第一层（Level 1）是当前大部分企业和开发者所处的现状：人们构建出较为简单的单体 Agent，这些 Agent 只是将网页检索当作一项外挂工具（Tool Use）。这是因为早期的基础模型在训练时就被教会了模仿人类使用各种现有软件和搜索框，因此最自然的路径就是给模型配上熟悉的浏览器或搜索 API，让它们去完成即时任务。 第二层（Level 2）则展现出了更强烈的自主性和复杂性，Parallel 的一部分先锋客户已经迈入了这个阶段。在这里，系统不再是单一的 Prompt 串联，而是构建了精密的多智能体编排系统（Multi-Agent Systems）。不同子 Agent 之间可以相互唤醒、分工合作、分配不同的上下文与决策权。最典型的例子是当下的 Coding Agent 框架，而在前沿科学探索中（例如 AI 科学家系统），这种模式更是展现出惊人的威力——通过持续投入巨量算力与海量数据，多 Agent 可以在极长时间跨度内协同攻关极其复杂的科学难题。 ```mermaid flowchart TD accTitle: 三层 Agent 演进范式 accDescr: 展示从工具调用到多Agent协同，再到Web端主动推送的三阶段跃迁 subgraph Level1["Level 1：单体工具调用 (现状主流)"] L1["单体 Agent"] -->|"即时拉取 (Pull) / Web Search Tool"| W1["静态网页与搜索索引"] end subgraph Level2["Level 2：长程多 Agent 系统 (正在发生)"] L2Main["主控 Agent (如 AI 科学家)"] --> L2Sub1["子 Agent A (检索/推演)"] L2Main --> L2Sub2["子 Agent B (验证/代码)"] L2Sub1 <-->|"相互唤醒与协同编排"| L2Sub2 end subgraph Level3["Level 3：从 Pull 到 Push 的事件驱动 (未来愿景)"] WebChanges["全网持续变化<br/>(内容更新/卫星影像/舆情)"] -->|"持续监控与算力分配"| ParallelPlatform["Parallel Web 基础设施"] ParallelPlatform -->|"事件触发：'Call Me' Push"| L3Agent["目标 Agent / 人类工作流"] end Level1 --> Level2 Level2 --> Level3 ``` --- ### `52:32` 从“Pull”到“Push”：全网变化驱动的事件型 Agent 经济 Parag 提出，网络与 Agent 交互的第三层（Level 3），也是最具颠覆性的终局状态：整个 Web 体验将彻底从“拉取（Pull）”变为“推送（Push）”。 在第一和第二阶段中，交互模式本质上都是即时被动的：Agent 调工具或唤醒子 Agent，发出一个请求说“立刻去帮我查这个”。然而一两年后，真正的范式转移将是“注册意图，等待触发”。用户或企业会告诉 Parallel：“如果网络上发生了某种特定事件，请立刻呼叫我（Call me），这样我的 Agent 或团队就能立刻介入执行任务。” Parag 经常跟团队分享一个关于未来算力利用的思考框架：假设未来 Agent 无处不在，且能力极强，那么只要某件事在当下有价值且仍有空闲 GPU 算力，系统就会毫不犹豫地在今天把它做完，绝不会拖到明天。那么到了明天，Agent 究竟该做什么工作？答案是：**响应世界的增量变化**。 明天的算力将被用来处理全新进入系统的输入流——可能是另一个 Agent 刚刚产出的计算成果，可能是卫星图像里捕获的物理世界变动，可能是突发的客户评论，或是人类突发奇想的一个全新洞见。而其中最庞大、最动态的数据源，正是“整个网络上每时每刻发生的所有变化”。 在 Parag 的宏大愿景中，Parallel 不再只是一个被动应答单次查询的搜索引擎，而是代表所有客户持续将庞大算力常态化分配在整个 Web 变化上的中枢神经。开发者只需向 Parallel 定义长期有效的行动条件：“一旦全网信息中出现了支持某种行动的信号，叫醒我，我来跑我的 Agent。”这彻底终结了单点时刻的被动检索，开创了全网常驻监听与算力自主分配的新图景。对谈在 Sonya Huang 和 Andrew Reed 对 Parag 的赞赏与感谢中落下帷幕。 - **工作流驱动力翻转**：从“Agent 主动发起查询”转向“全网动态事件流主动唤醒 Agent 处理任务”。 - **计算分配常态化**：并行分配算力全天候监控全球 Web 内容演变，把每一次网页信息的跃变，转化为下游自动化商业决策的触发信号。 ## Notable quotes > `00:00` 人类点击数据本质上是一个“Bug”（缺陷）。 — Parag Agrawal > `03:07` 十亿级对十亿级（billion to billion）的极致匹配问题 — Parag Agrawal > `10:00` 请从全网一万亿个网页中，精准找出最对的那 1000 个 Token — Parag Agrawal > `12:30` 每一次你能在搜索结果中向模型输出全部有效信号并大幅降低噪声，你就赋予了模型做更多事情的能力。 — Parag Agrawal > `15:10` 在面对 Agent 时，我们再也不必被逼着接受那种妥协。过去之所以必须妥协，是因为在标准的浏览器和网络协议限制下，我们根本没法稳定、精准地在整个互联网上将用户直接导向某一段高亮的极小文本段落；但面对 Agent，我们能把这一小段核心信息直接带入它的上下文，剩下的推理由它自己搞定。 — Parag Agrawal > `21:40` 十亿级需求与十亿级信息的匹配问题是一个永恒问题（a billion to billion matching problem is a forever problem） — Parag Agrawal > `21:40` 无论是企业内部私有数据还是广袤的开放网络，只要你能把需求与信息连接起来，并做出独一无二的深度洞察，哪怕只比别人好上 20%，就能形成压倒性的竞争壁垒。既然如此，为什么不全力以赴去推高上限呢？ — Parag Agrawal > `23:19` 从现实中的工程数据来看，根本不支持这种假设。用于预训练的爬虫数据，对于构建一个高时效性的实时网页索引几乎毫无用处。 — Parag Agrawal > `34:45` 我知道大多数人都讨厌广告，但我从理智上非常热爱广告，因为正是广告让如此多不可思议的内容和技术免费向全人类开放。 — Parag Agrawal > `40:00` 在 Parallel，我们最喜欢做的事就是构建模型。 — Parag Agrawal > `52:32` 如果网络上发生了某种特定事件，请立刻呼叫我（Call me），这样我的 Agent 或团队就能立刻介入执行任务。 — Parag Agrawal > `52:32` 一旦全网信息中出现了支持某种行动的信号，叫醒我，我来跑我的 Agent。 — Parag Agrawal ## Mentioned Parag Agrawal, Sonya Huang, Andrew Reed, James Flynn, Elon Musk, Parallel Web Systems, Twitter, 红杉资本 (Sequoia Capital), Google, Bing, Google Cloud, Gemini, OpenAI, ChatGPT, Claude, Notion, Cloudflare, Stack Overflow, SEC (美国证券交易委员会), Turbo, Search Agent, Shapley Inc. ## Video description Parag Agrawal is making a bet that goes against two decades of web search: agents will query the web a thousand times more than humans ever have, and the infrastructure built around human clicks is wrong for them. The former Twitter CEO, now founder and CEO of Parallel Web Systems, explains why Parallel treats human click data as a bug and trains on agent feedback instead. He unpacks the counterintuitive choice to ship a search agent before a search engine, building an index incrementally, and how the new Turbo product cut agentic search to 200 milliseconds. But the problem Parag keeps returning to is economic: the ad-supported internet collapses when agents show up instead of people. His fix draws on Shapley values to pay content owners for the value their pages provide agents, with real dollars reaching publishers, he predicts, within 12 to 24 months. Hosted by Sonya Huang and Andrew Reed, Sequoia Capital [Source](https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI4289055692.mp3)

[0:00] Parag Agrawal: 人类点击数据本质上是一个“Bug”（缺陷）。 [3:07] Parag Agrawal: 十亿级对十亿级（billion to billion）的极致匹配问题 [10:00] Parag Agrawal: 请从全网一万亿个网页中，精准找出最对的那 1000 个 Token [12:30] Parag Agrawal: 每一次你能在搜索结果中向模型输出全部有效信号并大幅降低噪声，你就赋予了模型做更多事情的能力。 [15:10] Parag Agrawal: 在面对 Agent 时，我们再也不必被逼着接受那种妥协。过去之所以必须妥协，是因为在标准的浏览器和网络协议限制下，我们根本没法稳定、精准地在整个互联网上将用户直接导向某一段高亮的极小文本段落；但面对 Agent，我们能把这一小段核心信息直接带入它的上下文，剩下的推理由它自己搞定。 [21:40] Parag Agrawal: 十亿级需求与十亿级信息的匹配问题是一个永恒问题（a billion to billion matching problem is a forever problem） [21:40] Parag Agrawal: 无论是企业内部私有数据还是广袤的开放网络，只要你能把需求与信息连接起来，并做出独一无二的深度洞察，哪怕只比别人好上 20%，就能形成压倒性的竞争壁垒。既然如此，为什么不全力以赴去推高上限呢？ [23:19] Parag Agrawal: 从现实中的工程数据来看，根本不支持这种假设。用于预训练的爬虫数据，对于构建一个高时效性的实时网页索引几乎毫无用处。 [34:45] Parag Agrawal: 我知道大多数人都讨厌广告，但我从理智上非常热爱广告，因为正是广告让如此多不可思议的内容和技术免费向全人类开放。 [40:00] Parag Agrawal: 在 Parallel，我们最喜欢做的事就是构建模型。 [52:32] Parag Agrawal: 如果网络上发生了某种特定事件，请立刻呼叫我（Call me），这样我的 Agent 或团队就能立刻介入执行任务。 [52:32] Parag Agrawal: 一旦全网信息中出现了支持某种行动的信号，叫醒我，我来跑我的 Agent。

[10:00] Parag Agrawal: Every query is essentially: give me 1,000 tokens from a trillion web pages on the web and make sure they're the right thousand tokens... Until last week we would give ourselves 3 seconds to throw a bunch of compute at read time to do it. This week... we shipped a product which now does it in 200 milliseconds. [25:00] Parag Agrawal: For them, it's like 'I'll take X trillion tokens'... For us, we are completionist. [35:00] Parag Agrawal: I know most people hate ads. I used to do ads and build systems for ads. I love ads... ads is the reason that so much amazing content and technology is available for free to all of us.

[0:00] our view at parallel. is that human click data He's a bug. an agent doing work with search should rely on agent feedback not human feedback. We believe that These models are really good at compressing information and we can benefit from a lot of the research that have gone into building models. and apply it to search indexing and ranking. And so you can now make many, many arguments, and that's the arguments we made back then, that actually now it's way more tractable as a problem because of the existence of agents, not just as in technology, but as a distinct customer. *music* Parag, thank you so much for joining us today. We're delighted to have you on the show. For those who don't know, Parag of Twitter CEO fame was CEO of Twitter before selling it to Elon and is now back on the founder arc. You founded a company called Parallel Web Systems, which is scaling up. agentic search for the agentic web. We're very excited to have you here today to talk about the future of search and the future of the internet. So let's get started. What's Parallel? - Thank you, Sonia, for having me. Thanks, Andrew, for joining us. I'm really excited about this conversation. At Parallel, we're building a bunch of technology in order to allow agents to search and use the web. So just like humans forever have figured out how to use browsers and search engines and clicking and browsing around the web to access information, agents need to do the same things. We started parallel with the bet that agents would do it a thousand X more than humans ever have. and as a result we need to reinvent the technology that can power search for agents. but also the business models that go alongside it. And that's what we're doing at Barul. Okay, I want to go back into what's what makes human and agent search so different. But before we get there, you've told us that, you know, you're, you're unlearning a lot of the lessons that you learned from running Twitter as you build build parallel. Why is that? Listen, when I was at Twitter in leadership roles. Twitter was a post product market fit, extraordinarily scaled business. where your feedback loops were from the hundreds of millions of customers using the product for 30 plus minutes every day. In that world you operate differently than a pre-product market fit company based on the premise that In a few years... a new customer is going to show up on the internet and let's go build technology for the not yet here. Customer. that we are all learning every day and every week. Fantastic. Okay, let's talk about Parallel. I have a question, actually. Can I jump in? No, you cannot, Andrew. I'm Andrew. I didn't get an introduction, but I'm also happy to be an inaugural guest on the Training Data podcast. Andrew is making his podcast debut on Training Data. We are honored to have you finally here. Thank you. Um... I feel like all of us in Silicon Valley have a very surface level idea of search. Like what is search? Like the job to be done is getting the answer makes sense. But we all know there's crawling and there's an index and there's ranking. But maybe let's take a big step back and explain. What is the problem of web search, be it for... a human or an agent, and then we can dive into the differences. The problem of web search, and we all know it and experience it, when we want to find something and we do not know where it is on the web, we go to a search engine like Google. and the search engine then hopefully surfaces the answer to us in the most convenient of locations. That's the base problem. Now, in order to do this, what is the search engine doing? The search engine is going and crawling the web, which is finding every URL that's out there, trying to read it. trying to organize that information in what might be called an index so that by having all of this stuff in memory in one location, you don't have to, once the query comes in, you don't have to spend time loading up pages because you already have them. You've already done pre-processing to organize them. And then when the query does come in, You quickly understand the query, you find the most relevant results, and then there are many many, because you're essentially taking hundreds of billions of pages, and narrowing it down to what? 5 or 10 or ideally 1 in terms of what someone is looking for right now. So you go through these multiple stages of retrieval and ranking [5:00] in order to surface the most relevant. result. So that's the broad problem. One way to think about it is it's a billion to billion matching problem, right? So you have hundreds of billions of pages, Thank you. and hundreds of billions of queries over time. And you need to figure out how to match make across these two. That sounds like an enormously expensive infrastructure challenge. And I think for the longest time, basically only Google And Bing! had you know done full web-scale crawls and indices. Why did you think about, you know, you as a young company could go off and tackle that problem? It seems like a problem of the giants. So, it is expensive in the long term. So what's actually interesting is that when I was starting the company three years ago, You could... Imagine a world where One, some of the reasons it was difficult for others to compete, which was not having access to great feedback in terms of is this a better search result than this? To collect that feedback at scale, there was a problem around human ratings. There is a problem around click data. You need to access those. Now, it is, of course, expensive to crawl the entire web and index it. But as you think about what agents and the large models that we now have access to have enabled is... the ratings data, can be now created by experts way more cheaply. Our view at Parallel is that human click data He's a bug. an agent doing work with search should rely on agent feedback, not human feedback. We believe that these models are really good at compressing information and we can benefit from a lot of the research that have gone into building models. and apply it to search indexing and ranking. And so you can now make many, many... arguments and that's the arguments we made back then that actually now it's way more tractable as a problem because of the existence of agents not just as in technology but as a distinct customer. And then we figured out a way to go about building this business, which did not require us to go spend all of the money on infrastructure upfront before we could service a customer. Right? So if you can paint a path of incrementally being able to build an increasingly larger and more sophisticated index. Over time, as you solve problems for more and more customers, That's... The insight that actually convinced me that this was a viable problem. for after take on. And how do you make that happen? Because I imagine this is one of those things where... customers want full coverage day one so how do you go about making that happen when when we first launched the product right we launched in We did not launch our search product first. We launched a search agent product first. our search agent could go essentially crawl the web. after a query arrived. So if you are doing deep research You have patience to the extent of a minute and we've had products which sometimes take 10 minutes of research That's a lot of time to be able to crawl a lot of pages if only you have enough of a map to know how to prioritize crawling, right? So you can make up for shortcomings, like index is oftentimes, you can think of it as a, latency optimization. So if you give up on that dimension, if you're competing with humans, that's why our search agents were competing with the alternative being outsourcing to humans to curate amazing data. Right? So we said It seems like humans sitting on search engines are far way easier to compete with. than a search engine on day zero. So by building a product that was a search agent to do real work on top of web data, we were able to incrementally go build our index. Well, there's some examples of what people built with your search agents. In the very early days, there were search agents being built with us for doing some kinds of insurance underwriting workflows and claims processing workflows. people in sales who are doing all kinds of sales data enrichment. people in finance who would previously and historically go to build a model, collect data by sending it overnight to a set of humans who would curate data that would feed into a modeling exercise they would do, would start using us to do that instead. And so we were going initially after replacing where there was outsourced. human work on top of web data. in order essentially to collect Evals. [10:00] to run agents, to figure out what search for agents. should look like in the first place with empirical use cases. instead of theoretical evils. I see. You're turning off the... crawl for inference time compute effectively. Yeah. You guys... were started before the term like NeoLab came into existence and you have you know, web systems product, but you also have like a bunch of AI people. From the outside, before we met for the first time, like it wasn't obvious to me how much of a web systems company or an AI company that you've built. The more time we spend together, the more clear it's like, by some definitions, this is a neolab. Do you want to talk about where... the research side of what you're doing comes in, not just the infrastructure and system side of what you're doing. so we don't categorize ourselves as a neoliberal. Well, of course you would not. We... No, because I don't think... Our output is immoral. Like I somehow... maybe my definition is broken. My definition of a new lab is an output as a model. I think our output is a complement to a model. What we build is something that multiplies on top of a model in order to... give either you can call it the model gets better or the agent built with the model gets better and has superpowers right so we've always want to be in a place where whenever someone ships a better model a neolab for example if somebody else ships a better model They have a higher hill to climb. For us, somebody ships a better model, they have now unlocked four more use cases where we can be valuable. Now... whether we need to do work that might be framed as research. that So in that sense we do have to do research but we are also not pre-training extraordinary large models. In fact our job is to compress things down into tiny ranking models, right? Like I think if you go back to my framing around like this 100 billion, 200 billion matching problem, every query is essentially Give me... 1000 tokens from a trillion web pages on the web and make sure they're the right thousand tokens. Like that's the query that we are getting effectively from to our search engine. And so now what do you have to do? And until last week we would give ourselves 3 seconds to throw a bunch of compute at read time to do it. this week, last week, we shipped a product, which now does it in 200 milliseconds, right? So you have now that much time to go figure out how to pick those, right? So you will do, We're trying to organize information in memory across the memory hierarchy in a way that we can access it fast. We're trying to figure out how to train what model Uh... to select the best 1000 tokens. And so there's a large amount of research that goes into figuring out how to take a pre-trained model of some kind, adapt its architecture for this new problem, and allocate effectively compute. in a way that produces the best output. with a limited compute/latency budget. So a lot of these use cases seem like the deep research kind of shaped use case. And when I think deep research, at least in the earlier incarnations, it was... Effectively like an agentic loop with the model reasoning and then basically just calling a search engine. calling Google or some proxy to Google. - Yeah. - Right? Why is that insufficient in your eyes versus what I'm hearing from you is agentic search is kind of a net new capability. So... Calling? Google. for every query in a deep research thing. It gets you somewhere. Right? Okay. If you use... parallel search. you will for the most part use under half the tokens in your agent. it will become more accurate and to be faster into it. And so you can think of it as... But every time, if you use only half the tokens, if your model is context limited or memory limited, you can now do More problems. You can do the same problems cheaper. or for after. So everything, to me, when you have a infinite appetite for information and relevant information for all kinds of work. It's at its core an optimization problem around quality, cost and latency. Every model advancement is about how do you squeeze out more intelligence and then how do you distill it down? into keep most of it at a tenth of the cost, right? And that applies to search as well. And every time you can produce... all of the signal with less noise in your search results to give to a model. [15:00] you now give the model the ability to do more. And so a lot of the bet here is intelligent compute allocation. across the model layer and the agent layer down to the search layer The other interesting thing to observe is the interface actually changes. Oh. when you think about serving agents versus humans. Humans rely on keyword search. With agents we've had to innovate quite a lot on the What does an agent tell our search engine? The more we know, the better we can do. Interesting. So humans can only hold a few words in memory, basically. versus an agent. No, I think we can. We're just lazy. yeah right like we can we can have a conversation we can hold a lot of words but when you start typing We We want to tell Google like two incomplete words with a typo in there. and hope for the best or rely on some form of a drop-down autocomplete thing to avoid typing those three words in the first place. So we're fundamentally lazy. Turns out like models less so. Few typos. better specified queries, perhaps longer queries Less for the search engine to guess what the agent might want. So you get to solve a different class of problems. One other thing about humans is I feel like we know how to decipher... the pre-AI slop that populates many of the top-ranking Google results sites you know, for best product for X, Y, and Z, and you get all the affiliate advertising things. Um, And yet, obviously, they rank highly. I'm very susceptible to it. Are you? Yeah, I'm like, I feel like I have this incredible, I can see it coming a mile away. How do agents deal with things like that, right? Where these are, you know, well-trafficked pages, they have seemingly good answers, but you just know they're not. real. Perhaps with parallel agents, you have to deal a little bit less with that. Let me tell you why those... pages. exist in the first place. Let's work through a simple example. If you ask for... of public companies most late most recent financials. like just the headline revenue number. We can sit here and know that there exists an authoritative filing with the SEC. which has that number. perhaps on page 73 of a PDF. That is the authoritative number. Now imagine Google decided like I care about authority. And whenever you ask that query, That's the first result you see. Right? And then you click that, this PDF takes what? Like 3.5 seconds to load instead of the second that another page does. You're already frustrated. And then you see a... or in speech. That's the alternative. But here you have to now grab your way to page 77 to find your answer. We're lazy. we're not going to do the work. Right? So now there exists an entire class of content on the web, which is like, okay, this information is needed by a lot of people. it is worth putting it on a page that loads fast, where this information is above the fold. All of you can go there. It's, 99.99% right. So you're not like so skeptical when you go there that this is 100% wrong. And it's added real value in the process because it identified out of the 300 page page earnings report to empty bits of information that should be above the fold. And so you can call it pre-AI human slop, or you can call it catering to a lazy human and being successful at SEO. The good news is with agents, We're not making the agent click around and fumble around and crap a PDF, right? We're taking ı excerpt from the most authoritative place on the web. and trying to bring it to the agent's context window. And so we aren't forced into this We are trade-off. And this trade-off existed with humans in the first place because like with standard browsers and protocols and everything, we just didn't figure out how to have... us point exactly, consistently across everywhere on the web to the exact right, highlighted, tiny, paragraph, right? With agents we get to bring that to the agent's context. and then let it figure out what's next. Can you walk us through what actually happens when one of your developers sends a query off to your agentic search API? It's in some sense pretty standard. So we've done some models to figure out what this query is and enrich it to figure out how it will flow into the rest of our system. we have a bunch of indexes which organize different subsets of the inf of the web [20:00] in different ways. And so the first layer will essentially craft queries for each of these different systems. Each. One of these systems perhaps is our sort of big index. One of these is perhaps our fresh index. One of these is perhaps like you can some people will describe it like a knowledge graph. Some people will describe it like a structured index. There are a bunch of these, right? So you're now deciding which ones this query needs to go to. you're figuring out what is the query rewrite for each of these. Then each of these has a big retrieval layer and then a ranking layer, and then more ranking layers. So you're going to try to boil down tens, hundreds of billions of URLs or documents. down into thousands, tens of thousands. down into specific excerpts and paragraphs in those tens of thousands with more and more bigger models running at each stage with different architectures pulling more features. to ultimately get down to here are the 1000 tokens I want to bring back to this AI. which has the highest signal. And now if you look at our various versions of our search API, They just throw... for different latency and cost constraints. different amounts of compute. at various points in this journey. to hit those limits. So in the abstract, it's Simple. Right? What's interesting is what each of these models that I described, how you curate and collect the training data for them. Right? How you build those models and optimize them. The index itself, how you use the memory hierarchy to store it to be able to hit certain cost quality latency thresholds. What's the North Star from a quality perspective? Like, you know, on Google, there's that, you know, did you get this result you wanted in the first three answers or something? What is the equivalent North Star for you? So I think of, I don't know if I'm right on this, but like my... take is that a billion to billion matching problem is a forever problem. Thank you. And... So the real question is at what point incremental optimization isn't worth the squeeze. so I don't think there is a thing as like okay We're done. on improving this thing. The question is at some point, it's going to get harder and harder to improve this thing. And it just won't be Worth it. But I'm hoping that we don't get there actually because like if you think of what we're doing with AI. we will have more intelligence that gets cheaper. every few months. and as a result of it, it will come down to having great models which are cheap. having great information. Thank you. matchmaking across a need and all of the information available to you, whether it's your own or on the web, and doing something unique and differentiated with it to produce more knowledge. And anytime you can do something 20% better. than somebody else. that might give you an edge. So why wouldn't you? So, in the super AGI-built worldview, it feels like If you can push on quality across web search and in the model layer. Thank you. Why wouldn't you? I've heard a point of view that this is so fundamental to the model companies that they're just going to own it. And in part, because as they're, collecting data for pre-training, that's a very, very expensive infrastructure exercise. That kind of is your source of truth for the crawl. What do you think of that? I don't see... Empiric data on the ground to support that view. To build a fresh web index. I don't think that crawl is particularly... useful. and let me frame y. So if you think of what we are building. We are building a complement to models. we like to crawl things that people don't like to crawl for pre-training Because like if the model already was trained on it, Like it's not useful. model companies for the training. aren't patient enough to go in a completionist way, try to wait for really slow random JavaScript to load. because... the number of tokens you get per amount of compute you throw at it Thank you. is like one order of magnitude. So two order of magnitude, too bad. And so, like, is it worth the extra... effort to get div tokens, for us it is because we are completionist. Right? For them, it's like I'll take X trillion tokens. So that's one. Now I do think this is a Core. [25:00] part of every agent. My world view is that if you are buying LLMs. for doing work. for 9 out of 10 use cases you will want them to have access. to the web and create search infra. optimizing for agents. So it is a real adjacency. for all kinds of LLM. inference. And that supports your view that model companies could, should have the best in class infra for it. So now, they can build it or they can buy it and that's the conversation and we will see who builds and who buys and who partners and how things evolve. Are you partnering with any of the model companies that you can share? We want to, we will. Mm-hmm. I can't share anything on that. We did announce, and I don't know your definitions of model companies, we announced today actually, that we are working with Google Cloud. to be a search and grounding provider for their enterprise agent APIs. So if you think of grounding Gemini models or other models available on GCP, when you build agents on GCP or chat apps on GCP or do any other inference with LLMs on GCP, when you want to attach web search to it, uh Your options... R. Google search or parallel search and Parallel search is product integrated. The integration is optimized, we've spent time with technical teams and training teams and product teams and commercial teams to make sure that when people use Gemini models with parallel. like they get exceptional and great results. So yes, there are these partnerships now. emerging, I bet that there will be several of these. they will all look somewhat unique. That's a big deal from the search king. Congratulations. Google's the original NeoLab. I don't know how to frame Google as a model lab versus a hyperscaler. And I don't know what precise lessons to learn from this one, whether it applies to other labs or to... Hyperscalers And so we will see. The way that I use voice agents and I need to get a reservation for dinner at night, what restaurants should I go to? And then that request gets fulfilled. by the agent. When we talk to a lot of the parallel customers, there's this you know, background agent, whether the monitor product, these agents that sort of are always watching the world or watching the Web. And when something happens, then they go off and take actions and do something with it. It might be worth, because if we think about what does 1000x more mean, there's like the depth of research and then there's just like the... What is actually initiating the tasks? Is it a human initiating the search or is it an agent itself? You want to talk about that dynamic a little bit? Yeah. So, There's a bunch of dimensions here. Let's go back to for a moment on search agents, right? If you run a typical search agent and I even without doing deep research, It'll do somewhere between 5 to 20 searches, even if it answers within a few seconds. Thank you. because why not? So already if you transition from using a ChatGPT, Medium, or I don't know all the categories, but like somewhere not on the high tier, like Instant Low Medium. Every time you write a prompt to it in ChatGPT, it will do five to ten searches. As you dial it up, it will do hundreds and thousands of searches. So one interesting thing to observe is like a human action to a multiplier on number of searches that happened, right? So just by using an AI app. you're kind of multiplying your way to perhaps one order of magnitude more searches. Now, A lot of our initial takes on the product and the market were to go after bigger multipliers than even that. So we were much more interested when you said I have a Thank you. portfolio of 10,000 small businesses where I have given out credit to. for all of them, every month. I have this human process that runs. to feel out how my risk is going up or down. Now, can we... and it relies on a bunch of web data. We're trying to use agents for doing this. So here a developer is effectively the multiplier there is hundreds of thousands or a million in terms of the number of web searches that happen because a human goes [30:00] programs that instead of now doing this process every month, We can do it every week. So, doing a lot of searches, when this agent runs every week Du. create a dashboard on the portfolio and a collection of action items that somebody needs to look at. Right? Now you can go to another example which is even more interesting. Do you use something to do meeting prep documents for you all that's an agent there's a sequoia one there's a sequoia agent yeah yeah i was i was gonna say james flynn but he's one of our great young guys So I use... Notions agent and you can build custom agents which look into all the internal data that we have a parallel plus all of the web data using Parallel's APIs. to Create meeting prep docs. One time I went and created one prompt to build this custom agent. Now it does. tens and hundreds of web searches for every meeting I have. Every time I build a new agent for a new use case, that keeps multiplying. So I think the path to these background agents doing more and more and more work. all the time for us, it's only going to be bounded by value versus spend. You know, like I don't think it's rational right now. Like I'm probably spending more on it than I should be. but it's not too much, so I don't care yet. So there'll be some rationalization in all of these agents. But I think we will deploy background agents to the extent that there is incremental value in doing that compute. And the same thing applies to searches. All of these background agents would do a bunch of searches. And so our first set of products were really obsessively focused on these. in part because we were building, growing the index and we decided that our company was based on three dimensions. quality cost latency and for the first couple of years we said Let's focus. Let's ignore latency and let's just nail the other two. Because optimizing systems, distilling to smaller models, is much more known art? than unknown research. So once we achieved the best quality search and search agent products. at every price point. we started working on latency. And that's what we shipped with a product we call Turbo now. It is the... fastest highest quality agentic web search. on the market by lot. Do you think there are more agentic queries than human queries on the web now? I don't think yet. Just given some of the multipliers you mentioned and given the background agents, it seems... I don't have to remind you, but we are early. We are very early in agent adoption. You go step outside of our bubble People haven't heard the word 'fable'. So we are very, very early. I think there are now... People like me who are probably operating at the 1000x. I don't know. What do you think? How many Google searches a day did you do three years ago before Charged GPT. 20, 30? Yeah, I would have guessed something like 20, 30. I think today, if you just look across at all of my agents, like... I bet they're doing thousand X more than that. Maybe. 100 to 1000. if you count some of the things that happen at my company which isn't like assigned to a human it might easily be more than a thousand X but I think we are the outliers rather than the So I think we're very, very early on this journey. I do think I recently saw, I think it was Cloudflare that said that in their monitoring of web traffic, the AI traffic is about the same as human traffic. in terms of page reads, which is slightly different from searches. because it includes perhaps all crawlers that are out there. and a bunch of other stuff. But I think it's... It's going to happen. So maybe this is a good segue to talk about a topic that I know you are passionate about, which is... the economics of the internet as we know them. Some of the fundamental assumptions there, human eyeballs, scarcity of attention, seems to be falling right in front of us right now. Are the economics of the internet broken now and what's going to happen? Yeah, no, I think this was perhaps... part of the thing that totally convince me that this needed to be done by me and build parallel. I know most people hate ads. I used to [35:00] do ads and build systems for ads. I love ads. Oh wow. I love shopping. Okay. So you get good ads. I don't know if I love ads but I intellectually love ads because ads is the reason that so much amazing content and technology. is available for free. to all of us. Google search wouldn't be free without ads, Twitter wouldn't be free without ads. And these are truly useful pieces of technology. a lot of content on the web out there. that we can access for free. Wouldn't be free. if not for ads. And so ads is a very efficient monetization scheme. It's a very efficient monetization scheme. because it is exceptional at differential pricing. Right? So, Most queries. Google loses money on. Some of them make it up and it's an extraordinary business with extraordinary margins. Same with Twitter. Most users... Exactly. You're welcome. I'm subsidizing the free information you guys are getting. You are subsidizing all of us. Thank you. But I think ads is extraordinarily efficient at differential pricing and monetization on the web, which is why it has been a dominant business model. Now, the core assumptions, as you noted, going into it around... limited human attention to translate. into outcomes. If humans don't show up and their agents show up on the web, Like what does this? mean? How does the business work? and yeah I think this was the If we don't figure out a new business model, we are seeing it already right, like people are gonna say, okay I don't want my content to be accessed by an agent. because I have a business model for humans. So I actually want to go... pay someone to SEO, optimize myself so more humans show up. But then if their agent shows up, I'm going to cut it off. Which seems... confusing and disconordinate. Like, okay, ultimately this agent is acting on behalf of a human. But we haven't found business model alignment in a way that... You can't monetize that visit. You can't monetize that visit. So let's say you... And in the business, subscriptions. So, you get a thousand humans and you convert 20 of them into monthly subscriptions. Right? You don't know how that loop works yet with agents. You can't distinguish, you don't have statistics, you don't know if these will lead to subscriptions, if they'll just keep stealing your content. as nameless agents. And so there are real challenges around the old business models breaking. and us not figuring out real scalable new business models. So if you own high quality content and you currently... Put it on the web. Your one available business model pre-parallel was two. be in the head and be able to transact with the lab on some fixed fee contract. That's literally which includes some amount of training and liability and then the inference time axis. That was your one option. That option is not available to most content on the web. It's a very head phenomena and even for the head It is a broken... business model because... when AIs or inference grows, let's say 7x this year. and another 7x the next year. on this 50x. their deal size is not going 50x. None of them, after signing a two-year deal, believes that their share isn't going to decline materially at renewal. And so... These are like fixed price constructs in the world of AI infants. and so which doesn't drive for sustainability. for all of these businesses. Now, Our solution is trying to learn all the lessons from my work at ADS. on having transacted, like I sold Twitter data to OpenAI. having transacted on that side. to figure out what actually might work and be incentivized. So what might work? Efficient differential pricing. paying for differentially a lot for extraordinarily high value content. paying a lot for high value work accessing the same content? So it's differential in both dimensions, quality and value of work being done with it. and... a way of doing this scalably and not just with bespoke deals. So, those are the properties needed for any reasonable solution. And the biggest property of It all is [40:00] incentive alignment, right? Like at what point do people want to collaborate into this enterprise. Why do the model companies need to pay anything at all? You should ask them. my understanding. is... one you want, fresh data during inference time to be able to display it in products like ChatGPT or Cloud. Do you want training data? And 3. You want some liability protection. for training that you already did. *sniff* And so the payments are some combination. of these three. Thanks. And I can't. be sure of like how they value each of these three. You said even a very simple query will go do 10 searches? And so how do you do attribution between all the different sources that, you know, boil down to one paragraph response? Yeah, at parallel we like to build models. I heard so no I think it's a it goes back to my point around incentive alignment so what are you let's before we try to build a model let's try to figure out like if you were going to try to do this intellectually theoretically how would you go do it So you have to ask the question of, okay, how much incremental value did somebody's content add. right? So you can You can run all of these simulation exercises. One piece of content out. of the corpus. And then you say, let's run the agent. Let's see if the quality of our outputs declined. How much? To Claw back that quality, perhaps if I threw a little bit more compute or a better model in some way Could I claw back that quality? Oh I could, it cost me a cent. Huh. I could get The quality I lost by not having this source My alternative was to throw more compute. a cent worth of compute. to get that same quality. It might be a different result, but as far as the end customer value is concerned, it's the same quality. And so you're like, okay, this source is worth close to a cent, feels like, right? That's intuitive. Now, the formalization of this kind of an intuition... is the core framework we use, it's called Shapley Values. yes let's go what's a Shapley value it's music's my ears it's a game theoretic you're you're a bunch of game theory yeah yes Shapley value is this very theoretical mathematical concept. It is used in sharp values and feature importances for those who are ML folks here. Let's say the three of us collaborate on something, and the whole is bigger than the sum of parts. in that moment. The theoretical question is, okay, how do I divide up this sort of bigger pie that we created by collaboration? So all three of us have incentive to collaborate, right? And Shapley values is a mathematical way of effectively answering this question. Now, that sounds amazing, right? Like you could, if you're co-founders, you could figure out how to divide equity. Unfortunately, it's not that useful. It's not that useful because in order to compute Shapley value, you need to simulate all words where some subset of the two of us collaborated, but the third one doesn't. And play out those realities to then impute back to today in terms of how we should divide the pie, which in practice you can't do. most places. In ML models, when you do feature importances, you kind of can. You can hold a feature back, Run your model and see how well it did. In web search, we can run simulations of if we did not have access to this URL or this domain or this collection of them, How would the agent perform? We can, if you're good at evals, if you're good at assessing quality, you can build that data by running various scenarios. Collect a bunch of this data. and then you can trained models to... Your favorite thing. Favorite thing. So, the challenge with Shapley values is computing Shapley values in our context. Thank you. is way more expensive than the amount of dollars... we spend on the agent, forget the amount of dollars we want to pay a publisher. Right? So to compute a content owner, to compute that a content owner gets a dollar, if I decide to do the full Shapley value computation, that might take several dollars. So it doesn't make any sense. except you can really well estimate them if you build the right kind of data and models around it. but we have confidence that our estimations are good and that This is sound theoretically that If there was perfect information symmetry, people would want to collaborate. So the same way in the old ads days, people [45:00] Second price auctions and believed, for better or worse, that people would reveal their true bits. in an ad auction and end up paying less than that. uh Once feedback loops get established in a market, like today everything is like an autobit, right? Like most people are measuring ROI when they're doing advertising and running on autobid instead of making up bids. I think, a solid foundation based on incentive alignment that Chaplin Mad drives. will ultimately maximize participation. of content owners. into this. as well as content seekers via AI. in an optimized system. And we have some positive evidence to support it. Like we've had some interesting partnerships that we've been able to do and announce. And yeah, like imagine... sitting with content owners and explaining Shapley math. Takes a moment. But ultimately once Once you pull out the properties, right, that you participate in the value, if you have unique differentiated data. you get paid more. if a banker in an expensive job reads your data versus my retired dad reads your data. . the banker ends up paying more for that REIT. because it's a part of high value work. Thank you. and The macro math also seems to work. Like if you're going to spend a lot of money on inference on LLMs for knowledge work, If we allocated a like what 2-10% of it. to data on the web. That's way bigger... than all, Web data business models. today. outside of like Waldgarden's like. Facebook and LinkedIn. The macro math supports it. It is a scalable approach. And as... agents on the web grow an order of magnitude year on year It's like by my calculations, like we're... 12-24 months. from This Math. being able to give meaningful dollars for a very wide range of content owners on the web. That's very exciting, especially at a time when, you know, you see the traffic data, the stack overflow, plummeting, you see a lot of the human internet as we know it. going away because of incentives, that's very exciting to see how you're thinking about incentive alignments for people to keep publishing. Yeah, it's why we've tried the company. Why is this company called Parallel? the company's original land? Was Shapley Inc. Really? It was. Do you know this? I knew that, yeah. When I incorporated, I'm telling you, I was obsessing about everything to do with the problem space. So while knowing that the first set of Also one. Shaplink is a terrible name for a B2B product. It was not going to be the long-term name. So we... But I incorporated it as Shapley Inc. for lack of a better word. And Shapley.ai happened to be available. Shabli.com? is a parked pond domain. So, It was not going to be a name. I went around not talking about my company at Shapley. and my badges at all events used to call it NUCO. or stealth company? And it took us almost six, eight months to figure out what the real name of the company will be. We end up at parallel in part because it, Click. One, at the time we are doing a lot more in parallel. And two, we started visualizing this sort of a parallel web built for AIs and how its properties are different. And this metaphor that when you publish, you're now thinking of, as we all are now, of two audiences. Okay, I'm gonna create a page. I know humans will read it What should it look like to them? And then how should I make sure that agents can read it to So it feels like you're dual publishing to two audiences. And so we had this like this parallel web for agents will emerge. And so that's why we Started liking battle. Love it. I was just thinking about this recently with the... you know, things like earning transcripts, right? Like I feel Very confident that more people are consuming earnings transcripts through agents than are actually certainly listening to the audio and honestly probably reading the transcript itself So there are these very high value portions of the web that have flipped already to being agent first and obviously the communication [50:00] has not yet flipped. Um, But I think if I were a public company CEO, today and I was doing an earnings report I'll make very clear that the thing that I'm saying will be transcribed and interpreted correctly by the agents, not just by people listening. Theoretically, yeah. Same for us when we're publishing docs for our APIs. Like, it's like our customers are... building AI solutions. They are using AIs to do it. It's there. Agents... Reading our docs. End quote. in RFD case. It's not humans fumbling around docs pages for the most part. So... In fact, for us, the primary audience is a agent and that's how we test our docs. A parallel web for agents. That's very cool. Maybe close us out. Tell us, just give us a snapshot of where Parallel is today and... If everything that you hope... To build comes true. What does parallel look like? What does the world look like? What's your role in it? I think of... three levels in this journey. Level 1 is people are building simple agents that use the web more as like a tool, like a web search tool, because it's familiar. Because the first set of agents we built, to your point, was just like model and give it the same tools, because these models have been trained. to use all the tools humans have been used to and let them do work. By and large, if you think of most work, it's there. today of our influence it's there today. We have now started seeing some subset of customers who are in this world where They're building more sophisticated, multi-agent systems which use sub-agents and have agents wake each other up or orchestrate in interesting ways, right? People are seeing that as sub-agents, most familiar one is sub-agents with encoding agent harnesses. But we see a lot more of that. Like if you're building an AI scientist, some of those systems are very interesting, very sophisticated, very long running. And like, just... throw large amounts of compute and data at a really hard problem. And I think the third layer for the web specifically is... The web goes from pull to push. So today by and large across the first two modalities I described either an agent isn't calling a tool or a subagent. But it is telling it a request saying, go find this for me. Right now. And I think where we will... end up. in a couple of years. is a variety of use cases will be The web. Parallel. Call me. If this happens, so that My agent can do some work or a human can do some work. So... I use this line with my team all the time, which is like, imagine like agents are everywhere and they can do a lot of things. Right? Thank you. Is there something you can do today? and it's worth doing. And we still have GPUs available. We'll just go do it. We won't be like, oh, let's just do this tomorrow for the sake of it. If it can be done today, do it. So what will we do tomorrow? We will do it tomorrow. work. In response, to something that circles through either another agent's work. or something changes in the world, as visible in satellite imagery, or some... customer commentary that happens, some agent finishing some compute, a human having a new insight to trigger work. But there are going to be a few feeds like this, which will drive new... agentic work tomorrow. And one of those fields, is going to be everything that changed on the web. Which... is really exciting for me because then you are framing not a point-in-time need. but a long term Here is what is actionable for me If something like this happens as evidenced by all of the information on the web, That is changing all the time. Call me. and then I'll run my agent on it. And so we get to then allocate Compute! onto the entire web, all the time on behalf of all the customers. And that's really exciting. That's awesome. Really ambitious vision. You're clearly extremely passionate about this and exciting to see you building. Thank you so much for joining us today, Parag. And thank you for joining us, Andrew. You're very welcome, Sunya. Thanks, Parag. Thank you, Andrew, for your debut. So [55:00] Thank you.