Stripe's Collison Says AI Agents Will End E-Commerce Keyword Search

Stripe创始人Collison：AI智能体将终结电商关键词搜索 自主AI购物软件将消费者的精准限制条件直接与商家匹配，并通过一次性支付凭证实现自动化结账，从而取代传统搜索引擎。 John Collison深入剖析了代理商务在消费端（B2C）与开发者端（B2B）应用场景中的兴起，阐述了传统关键词搜索为何已经过时，以及AI智能体如何擅长将用户的精细化限制条件与利基产品相匹配。他详细介绍了正在成型的新兴技术基础设施，涵盖机器可读的产品数据、安全的一次性支付凭据，以及基于稳定币、用于按使用量付费的数据和API微交易。尽管存在对失控机器人或传统广告模式遭破坏的担忧，Collison认为人类仍将保留在决策环节中，AI将减少繁琐的交易摩擦、降低创业成本，并推动新创业浪潮的兴起。 Stripe 联合创始人 John Collison 表示，代理商务（Agentic Commerce）正引领电商范式转移；其真正价值不在于取代时尚浏览等人类享受的体验，而是消除繁琐、高摩擦的购买与开发者配置流程。 (5:00) 传统的关键词搜索在复杂购物场景中效率低下，商家正从依赖传统 SEO 转向针对大模型进行可读性优化与实时 API 对接，这也让小众利基品牌得以绕过巨头聚合平台脱颖而出。 (10:00) Stripe 正在构建包含一次性支付凭证的代理结账基础设施与稳定币轨道，使 AI Agent 能够安全完成交易，无需反复暴露用户凭证或触发防 Bot 机制。 (20:00) AI 在处理微小决策时不会产生认知疲劳，这让长期受阻的单篇内容付费等微交易与基于使用量的按需计费（Pay-per-use）商业模式成为现实。 (25:00) 受益于 AI 降低协作成本与创业门槛，Stripe 平台第一季度新增企业创建量同比激增 71%，更小、更敏捷的企业生态正在快速涌现。 (45:00)

Stripe co-founder John Collison argues that agentic commerce—autonomous AI purchasing digital resources and consumer goods—will transform internet infrastructure by replacing keyword search with machine-readable specifications and unlocking programmatic microtransactions. While reducing transactional friction across consumer and B2B workflows, the paradigm shifts web discovery while preserving core human decision-making and advertising models. Consumer and Developer Applications of Agentic Purchasing (5:00) Collison divides agentic commerce into consumer (B2C) and developer (B2B) applications. Rather than displacing subjective, leisure-driven experiences like fashion browsing or vacation daydreaming, AI shopping agents primarily eliminate friction in mundane or highly specific tasks. In B2B contexts, this includes software autonomously provisioning domain names and hosting configurations—as seen in Stripe's partnership with Cloudflare—or purchasing API keys. In Collison's view, technology historically trends toward lower-friction alternatives. At minimum, AI agents streamline transactions by handling form auto-fills and checkout flows, while B2C agents add value by parsing user constraints—such as ordering recipe ingredients while accounting for pantry items already on hand—rather than trying to automate taste-driven human experiences. Replacing Keyword Search with Machine-Readable Discovery (10:00) Collison characterizes traditional keyword search boxes as an outmoded method for navigating complex commercial purchases, arguing that human commerce relies on dialogue, physical constraints, and context. As AI agents navigate purchasing on behalf of users, retail visibility is shifting from traditional search engine optimization (SEO) toward machine readability, where LLMs evaluate structured specifications, technical parameters, and third-party editorial reviews. This shift challenges traditional aggregators and heavy advertising spenders, creating opportunities for niche merchants whose products precisely match granular user prompts. However, Collison rejects the notion of a post-advertising internet, asserting that brand equity and advertising will continue to guide human choice in both directed commerce (where sponsored responses integrate into prompt outputs) and undirected commerce (such as visual social media feeds). Checkout Infrastructure and Guardrails for Autonomous Agents (20:00) Executing real-time agentic transactions requires new payment plumbing, including machine-readable APIs for live inventory and pricing, alongside three-way checkout protocols connecting consumers, AI platforms, and merchants. Stripe is developing infrastructure alongside companies like OpenAI, Microsoft, Meta, and Google to issue single-use payment credentials, providing an authenticated checkout path akin to a "TSA PreCheck for bots" without exposing consumer payment details across multiple domains. Addressing liabilities surrounding rogue AI purchasing, Collison considers runaway spending risks overstated. In the near term, most workflows retain a human in the loop for final financial sign-off. For broader autonomy, Collison argues systems will replicate corporate expense management, enforcing programmatic guardrails, spending caps, and tiered delegated authority rather than withholding autonomy entirely. Token-Driven Economics, Granular Billing, and Stablecoin Rails (25:00) Because frontier AI models incur direct marginal costs per generated token, usage-based billing has become the foundational economic model for the AI era. While human cognitive fatigue historically crippled digital microtransactions—such as paying 40 to 50 cents to read single articles—Collison notes that AI agents can evaluate micro-decisions continuously without friction, enabling viable "pay-by-the-sip" data feeds and programmatic research. To accommodate these continuous computational transactions, Stripe acquired billing tracker Metronome. Collison also points to stablecoins as critical underlying rails, offering low-cost, instant settlement that allows software agents to pay fractions of a cent per API request without requiring dedicated user accounts or stored credit cards across dozens of disparate services. Coasean Firm Dynamics and Emerging Ecosystem Frictions (45:00) Collison highlights a broader surge in commercial dynamism, noting that new business creation on Stripe grew 71% year-over-year in Q1 alongside a sharp rebound in app store launches. Referencing Ronald Coase's theory of the firm, Collison suggests that as coordination costs collapse through AI integration, the market will favor an expanding ecosystem of smaller, highly agile firms coordinating programmatically. However, this transition introduces distinct ecosystem frictions. While high-value content providers (such as Reddit and financial feeds) increasingly negotiate formal AI licensing agreements, low-quality ad-arbitrage websites face obsolescence. Host reflections also highlight risks of algorithmic filter bubbles and the rise of adversarial bot-on-bot dynamics—such as AI systems issuing bulk healthcare claim denials that are contested by counter-bots. Collison forecasts an internet populated by nimble, highly automated firms transacting via programmatic rails, but significant questions remain regarding how platforms will balance open human browsing against data scrapers, how retailers will adapt to bot intermediaries, and whether microtransactions will ultimately replace ad-supported web models.

Stripe联合创始人兼总裁John Collison指出，AI代理商业（agentic commerce）正通过将交易摩擦降至最低重塑互联网经济格局，其核心在于让人机协同处理复杂的定向采购与微额结算，而非完全取代人类的消费意愿与决策主导权。 代理商业的范围界定与交易摩擦消除 (5:00) John Collison将代理商业广义界定为“AI替用户执行购买”，并划分为两大主要场景：面向消费者的B2C模式（如用户在ChatGPT或Claude等AI应用内完成产品调研并直接由代理执行支付），以及面向开发者与企业的B2B模式（如软件编码代理或Cloudflare等工具自主采购API访问权限、算力资源或完成域名配置）。Collison认为，回顾网络商业演进历程，代理商业不仅延续了信用卡、PayPal、Apple Pay及Stripe Link消除交易摩擦的技术演化路径，更带来了根本性跨越——直接免除交易终点繁琐的网页表单输入。 针对批评者认为AI购物无法取代度假规划或时尚浏览的论点，Collison反驳称这些批评选错了参照物。他强调，挑选时装或憧憬假期本身具有很高的休闲愉悦感，用户并无外包意愿；代理商业的核心价值恰恰在于接管乏味、琐碎且重复的日常采购痛点，例如根据特定烹饪食谱自动比对家中现有调料并补齐缺失食材，或是让开发者无需手动配置DNS记录和切换开关即可一键部署网络资源。在技术演进史上，低摩擦方案通常具有不可逆的替代优势。 检索范式重塑与面向AI代理的商家优化 (10:00) Collison直言传统的文本框关键词检索在电商领域已显得十分荒谬，除了查找特定书名或影音光盘外，传统检索无法满足服装、家具等复杂品类的购买需求，因为人类真实购物高度依赖上下文、物理尺寸及个性化约束。相比之下，AI对话范式支持消费者精确输入结构化条件，并展现出超越预期的产品匹配“品味”，能够越过传统的巨型电商聚合平台和高额广告投放者，精准发掘并推荐符合条件的小众专业商户（例如特定规格的自行车托运箱或集成电源插头的便携适配器制造商），从而显著提升市场匹配效率。 这一演进迫使商家优化策略发生根本转变：零售商必须从讨好人类眼球转向增强对大语言模型的“机器可读性”。商家不仅需要在全网沉淀详尽且规范的产品规格与真实评测数据，还需在技术层对接实时API接口，向AI代理开放动态库存、尺码变体与即时价格查询，避免模型依赖陈旧的训练集。与此同时，网站必须重新评估反爬虫策略，针对经过身份验证的AI代理构建类似“机场预检通道（TSA PreCheck）”的通行机制，以便合法代理顺畅完成交互。 结算基础设施演进与微额支付商业化 (25:00) 实现实时的代理交易依赖全新的底层安全架构。Stripe正与Google、Meta、微软及OpenAI等平台协作，充当用户、AI平台与商户之间的安全中介，通过签发单次有效支付凭证，使用户无需在多个分散站点重复输入或暴露敏感支付数据。率先采用代理商业标准的商家，将如同早期支持亚马逊Prime一键下单的服务商一样，在AI搜索交互界面中斩获显著的转化率优势。 此外，代理商业从根本上化解了困扰互联网数十年的微额支付困境。以往按篇收取40至50美分的数字内容付费尝试屡屡破产，主因并非价格本身，而是人类对于频繁评估微小支出存在极高的心理认知负荷；相比之下，AI代理天然擅长细致审视每一项微观决策且不知疲倦，使按需、按次计费模式首次具备大规模商业可行性。伴随大语言模型按Token产出计算边际成本的现实，基于用量的计费已成为AI时代的核心引擎，Stripe亦借此收购了Metronome。未来若引入低成本的稳定币（stablecoin）支付网络，AI代理甚至能以极小碎片化的单位自动支付API单次调用费用，无需在各服务商处预先注册充值。 广告模式延续、授权生态与授权边界控制 (15:00) 针对AI将摧毁互联网广告与免费内容生态的担忧，Collison持审慎怀疑态度。他认为人类不会被排除在商业决策链条之外：AI代理负责完成广泛的信息筛选并列出权衡利弊的备选方案，而最终拍板权仍属于人类，因此品牌声誉与广告依然在潜移默化中发挥主导作用。此外，以定向搜索为驱动的需求可能会融入AI平台生成的原生赞助信息，而依靠社交媒体算法驱动的被动发现类消费广告仍将蓬勃发展。 在数据层面，优质专有内容从非彻底“免费”。目前Reddit等平台以及FactSet、FMP等财经数据提供商已在通过向大模型出售训练授权开辟可观的变现渠道，唯有缺乏独特价值、依靠诱导点击和弹窗广告的低质SEO垃圾站点将面临洗牌。关于AI代理“失控乱花钱”的顾虑，Collison认为该风险被严重夸大；短期内人机协同（Human-in-the-loop）将是主流架构，长期看则可借鉴成熟的企业财务治理体系，通过设置分级授权、硬性额度上限与审计监控，在释放代理自动化效能的同时防范越权交易。 宏观创业繁荣与技术落地的现实局限 (45:00) Collison对当前的“智能爆炸”抱有极高热情，并指出AI正在激发出显著的宏观经济活力。Stripe平台数据显示，第一季度新设立企业数量同比激增71%，移动应用商店的新应用发布量亦突破了长达数年的平缓期。引用科斯定理（Coasean economic theory），AI大幅降低了市场协调与技术构建门槛，使得单枪匹马或规模极小的团队能够敏捷地对传统大企业的僵化流程发起挑战，促成更加动态分散的企业组织生态。 然而，Collison同时告诫技术管理者在拥抱变化时需保持谦逊，避免过度自信的妄断，需在深信技术潜能与清醒审视其边界之间保持平衡（他称之为适度的“AI精神官能症”）。以内部落地为例，AI在Stripe软件工程中的普及迅速而自然（许多Bug已能全自动修复），在销售端也能带来量化的高产出；但在法务、风控尤其是财务核算等领域，大模型虽精通公开法条，却在处理专有数值分析时屡屡受挫——往往能生成视觉工整的财务模型，却在表格中段出现严重的数学运算错误。 节目最终揭示出代理商业带来的深层矛盾：尽管自动化消除了传统流程中的琐碎摩擦，但也可能催生新型系统内耗，例如医疗保险领域中理赔审核AI与申诉AI之间的“机器人对抗互搏”；同时，当网页体验彻底转向API化与无感微支付，如何在算法个性化推荐带来的信息茧房之外保留纯粹的探索乐趣，仍是尚未解决的行业命题。

## Background 本期节目由彭博社（Bloomberg）旗下的《Odd Lots》播客于 2026 年 5 月 16 日在 RSS 及网络平台发布。节目主要探讨随着能够代表消费者进行购物的 AI 代理（Agents）崛起，“代理型商业”（Agentic Commerce）将如何彻底颠覆传统的在线购物体验、零售商的应对策略以及品牌向机器而非人类买家进行营销的新范式。 ## TL;DR Stripe 联合创始人 John Collison 深入探讨了“代理型商业”在消费者端与开发者端的落地图景，指出关键词搜索正在被能够理解复杂多维约束的 AI 深度调研所取代。节目剖析了消除支付摩擦、代币化凭证结账以及针对智能体优化（AEO）的底层基础设施，并阐明了广告价值为何不会消亡的“两端 AI、人在中间”决策模型。此外，对谈还展望了微交易（Microtransactions）的真正落地、数据按需付费模式以及小规模高敏捷企业崛起的全新互联网商业生态。 ## Atmosphere & style 整场对话充满极客幽默与敏锐商业洞察的碰撞，主持人 Joe 与 Tracy 凭借对“氛围编程”（Vibe Coding）和网购癖好的自嘲，为看似枯燥的底层支付与 API 架构注入了浓厚的生活气息。嘉宾 John Collison 展现出深厚的行业透视与务实克制，在解构前沿技术兴奋点的同时理性厘清“人机回环”的现实边界。听感上节奏明快、思辨犀利，兼具硅谷前沿创新的狂热想象与华尔街式的冷静审视。 ## Bulletin - `01:45` Stripe 联合 Cloudflare 推出由 AI 代理直接采购域名与完成托管配置的功能，标志着“代理型商业”（Agentic Commerce）从概念步入现实，大幅削减传统云资源配置的繁琐摩擦。 - `10:00` Stripe 联合创始人 John Collison 直言传统关键词搜索在电商中已走向荒谬，AI 驱动的参数与多维约束调研正在打破大型聚合商垄断，显著提升长尾细分品牌的曝光与转化。 - `26:15` 微交易（Microtransactions）受制于人类“认知负荷”的历史难题将被 AI 终结，结合稳定币基础设施，专业数据与内容按单次网络请求（By the sip）计费变现成为可能。 - `45:00` 得益于 AI 大幅降低创业与智力调用门槛，Stripe 第一季度平台上新创建的业务数量同比激增 71%，验证了科斯定理下小型化、高敏捷协同团队的井喷趋势。 ## Key takeaways - `05:08` 代理型商业的核心是“AI 代替人类购物”，涵盖消费者端通过自然语言交互完成商品比选与结账，以及开发者/B2B 端自主采购域名、算力与 API 密钥等数字资源两大支柱。 - `06:17` AI 代理应当接管毫无乐趣可言的流程琐事（如补齐食谱缺失食材或配置网络记录），而非剥夺人类在度假攻略规划或时尚穿搭挑选等消费过程中本身的乐趣。 - `10:00` 电商搜索范式正从关键词输入转向自然语言上下文探索，支持用户在复杂空间与多维尺寸约束下交互，促进了长尾利基品牌的崛起。 - `12:45` 面向智能体优化（AEO）成为新命题，商家不仅需确保数据结构化机器可读，AI 本身在匹配特定任务工具与商品时也展现出极高的“品味”和筛选效率。 - `14:38` 广告业不会因 AI 消亡：未来商业决策呈现“两端是 AI（调研与执行）、中间是人类（拍板）”的格局，长期广告建立的品牌偏好依然是人类最终拍板的核心驱动力。 - `20:00` Stripe 正在构建智能体商务底层交易管道，通过实时 API 获取商家最新库存与 SKU，并借助类似机场预检的“Bot 预检通道”与一次性支付凭证化解安全与防爬虫矛盾。 - `22:48` 智能体落地的路线之争在于“端到端模拟人类点击 UI”还是“构建原生结构化 API”，率先提供原生无缝结算接口的商家将享受类似亚马逊 Prime 标识的转化红利。 - `26:15` 传统信用卡固定费率阻碍了按次小额付费，而对成本计算极其理性的 AI 使得“按口啜饮式”（By the sip）调用数据与稳定币微支付在经济模型上变得切实可行。 - `30:00` 开放互联网的“免费漫游”正在重塑，优质数据所有者转向内容许可或按需付费接入，而严重依赖垃圾 SEO 和点击弹窗的低质内容农场将被彻底清洗。 - `37:37` 代理自主权无须走向完全失控，“人机回环”（Human-in-the-loop）模式通过智能体跑完比选与配置流程、人类最终一键敲回车确认扣款，兼顾了极致效率与风控安全。 - `40:00` 企业应用需警惕“AI 精神狂热”：管理者应清晰洞察模型能力边界，清醒认识到模型在软件编码上极具生产力，但在处理专有复杂财务数据时可能仍是“数盲”。 - `42:45` AI 在企业内部的消化吸收极不均衡：软件工程部门拥抱最快并形成自动修复 Bug 闭环；销售拓展端人均产能显著增长；而法务与财务等后台职能因数据隔离改造成本高推进最慢。 - `45:00` 在科斯企业理论框架下，AI 使外部协调与交易成本大幅降低，企业组织形态将趋向于更小规模、更精简人力的敏捷协同体。 - `50:00` 未来网络可能演变为“Bot 对战 Bot”的高频算法博弈（如保险公司批量拒赔 AI 对抗自动化维权初创 AI），持续拉动算力、芯片及底层硬件基础设施的指数级消耗。 ## Detailed walkthrough ### `01:45` 播客开篇：从 Vibe-Coding 的痛点到 Stripe 与 Cloudflare 的新合作 Joe Weisenthal 与 Tracy Alloway 开启了本期对谈。Joe 兴致勃勃地提到自己最近又在沉迷“氛围编程”（vibe-coding）——只需用自然语言将想法打在屏幕上，AI 就能把应用代码瞬间搭好。然而，整个流程中最让人抓狂的一环，却完全与代码无关：那就是“注册域名”。Joe 忍不住疯狂吐槽自己在 GoDaddy 上的折磨经历：好不容易想出了绝妙的 idea 并生成了应用，一到域名注册商那里，就得挨个配置 A 记录、各种字段、切换 Workers 和 Pages，整个过程繁琐且令人烦躁。Tracy 也深有感触地附和，回忆起早年折腾 Bluehost 时的糟糕体验，尽管域名的复杂配置在防止恶意劫持上有其合理性，但该领域的体验演进实在慢得惊人。 这一痛点在不久前迎来了转机。正当两人出差在马德里期间，支付巨头 Stripe 宣布与云基础设施服务商 Cloudflare 达成合作：用户只需向 AI Agent（智能代理）下达指令，Agent 便能自动完成域名购买并直接部署托管。Joe 激动地在 Newsletter 里大书特书，直呼如果这个功能早一个月上线，自己就能省去天大的麻烦。这不仅让“配置域名”加入了未来再也不需要人类亲力亲为的长长清单，也让“代理型商业”（Agentic Commerce）这个新兴概念迅速引爆了科技与创投圈的关注。 - **Vibe-Coding 的现实卡点**：用自然语言生成软件极度流畅，但将服务推向公网时的传统域名注册与 DNS 解析配置（如 GoDaddy、Bluehost 的字段与页面设置）构成了巨大的操作摩擦。 - **Stripe 与 Cloudflare 合作的意义**：赋予 AI Agent 代替人类直接调取支付与基础设施的能力，实现“动动嘴皮子就买下域名并完成托管”。 --- ### `03:30` 商业与互联网的底层重塑：当“有品位的买家”遇上 AI 代理 Tracy 坦言，作为一名热衷网购且自诩“极具品味”的消费者（Joe 调侃她经常沉浸在各类高端壁纸网站中反复挑选），这是她第一次真切地感受到自己可能被 AI “去中介化”（disintermediated）。这引发了一个极具深意的哲学与商业问题：如果一两年后的 AI 伴侣对用户的个人审美、生活习惯了如指掌，甚至在用户察觉之前就主动判断“Tracy 你正在翻修房间，需要这款新壁纸，我已经帮你下单并安排明天送达”，网购世界的“可发现性”（discoverability）和“品味”（taste）又该何去何从？ 两人进一步剖析了代理型商业对互联网根基的颠覆。互联网从诞生起很大程度上就是为了卖东西而构建的——无论是实体商品、信息还是数字内容；而驱动整个网络运转的底层商业模式，正是基于流量与点击的广告投放体系。过去播客曾深入探讨过互联网如何依赖每秒数百万次的实时广告竞价运转。如果未来的采购决策权从刷手机的人类转向了冷静计算的 AI Agent，不仅传统以“吸引眼球”为核心的广告模式将被颠覆，连带着广告所塑造的大众文化与审美范式都将面临根本性的洗牌。 ```mermaid flowchart LR accTitle: Agentic Commerce Impact on Internet Ecosystem accDescr: 展示代理型商业如何通过绕过传统注意力经济而改变互联网的运作模式。 subgraph Traditional["传统电商模式 (以人类为核心)"] A["人类注意力/浏览"] --> B["广告竞价与SEO"] B --> C["手动表单结账"] end subgraph Agentic["代理型商业 (以Agent为核心)"] D["用户指令/偏好"] --> E["AI Agent 自主决策与对比"] E --> F["API/协议直接结算 (Stripe Link等)"] end Traditional -.->|"范式转移 / 去中介化"| Agentic ``` --- ### `05:08` 专访 John Collison：什么是“代理型商业”的两大支柱？ 随后，播客迎来了重磅嘉宾——Stripe 联合创始人兼总裁 John Collison。当被问及究竟什么是“代理型商业”（Agentic Commerce）这一最新科技热词时，John 给出了返璞归真的定义：“广义而言，它就是 AI 代替你买东西”（broadly it is AI buying something for you）。为了更清晰地拆解其演进脉络，Stripe 将该领域划分为截然不同的两大类别： 1. **消费者端（To C）代理商业**：消费者在 ChatGPT、Claude 或专门的 AI 助手里反复交流、调研某款产品，最后直接让 AI 完成结算购买。 2. **B2B / 开发者端代理商业**：类似于 Cloudflare 与 Claude Code 等工具的联动场景，当开发者指挥 Agent 构建系统时，Agent 会自主采购运行所需的各类数字资源（如调用各类微服务的 API 密钥、域名、算力与服务器配置等）。 Joe 对此深表赞同，并补充道，以前做开发时最痛苦的环节之一就是四处寻找并购买 API Key，中途常常需要打断思路去网页端填写信用卡卡号；如今这一层由 Agent 代劳的自动化闭环，正让开发者体验发生质的飞跃。 --- ### `06:17` 消除交易摩擦的历史演进与 AI 自主决策的合理边界 面对 Tracy 提出的核心质问——“在线支付的历史本质上就是一部消除摩擦的历史（从手输卡号、到登录 PayPal、再到 Apple Pay/Google Pay 的一键支付），代理型商业究竟只是消除摩擦的最新一环，还是底层逻辑的根本改变？”——John 回答道：“两者兼有。” 首先在消除摩擦层面，即便完全抛开 AI 的自主决策不谈，仅就交易的最后一步而言，AI 也能带来质变。John 以自己最近购买装自行车的旅行箱（bike case）为例：他在 AI 应用中深入探讨各种细节——软壳与硬壳的优劣、是否必须拆卸车把、特定车型的兼容性等。在此过程中，AI 帮他发掘了许多极具竞争力的细分小众品牌（John 特别指出，AI 对长尾中小商家的曝光极有助益）。然而，“当你最终敲定了心仪的产品时，你真的还愿意去填写那些繁琐的网页表单字段吗？还是你只想对 AI 说一句：‘听起来不错，帮我买这个尺寸’？”技术的历史反复证明，低摩擦方案永远会胜出，AI 代替用户自动填表结账已经在发生。 然而，一旦涉及“让 Agent 拥有自主决策权”，科技行业却常常陷入啼笑皆非的误区。John 犀利地批评道，业内许多人在鼓吹 AI 代理时总是选一些极其糟糕的例子： > “他们总爱说：‘AI 会把你度假期间的所有活动全包办预订好。’但现实是，人们一整年都在憧憬并幻想他们的假期，研究攻略和预订本身才是最享受的乐趣所在！或者在女性时尚消费中，无休止的翻看滑动（scrolling）本身才是核心意义，买单只是随附的结果。这就好比机器人来抢工作，但大家其实特别想保留这些‘刷屏幕的工作’。” John 强调，真正能发挥 AI 自主决策价值的，是那些人类毫无乐趣可言的琐事与杂务： - **日常生活场景**：给 AI 一份今晚的食谱，让它“把做这顿饭缺的食材全买齐”。AI 能够合理推断家里常备有食盐和大蒜，但缺少碎猪肉（ground pork），因此自主下单补充缺失食材。 - **B2B / 开发者场景**：人类没有任何兴趣去重复填写配置域名的繁琐表单；让开发 Agent 协同 Cloudflare 搞定底层网络与域名资产的购买及绑定，不仅极大释放了生产力，也彻底将人类从无聊的流程中解脱出来。 ### `10:00` 告别“荒谬”的关键词搜索：AI 驱动的产品调研与长尾品牌的崛起 在探讨 B2C 领域的智能体商务（Agentic Commerce）演进时，存在着明显的阶段跨度：从最基础的减少表单填写等摩擦层，到未来 AI 真正具备前瞻性、为你自动购买你甚至不知道自己需要的物品。主持人 Joe Weisenthal 提及近期社交网络上的一张截图——某鞋类零售商的网站上不再展示 Nike Pegasus 或 Air Jordan 等具体品牌名称，而是将商品描述直接写成“适合中高阶徒步者的登山靴”，以此询问目前零售商是否已经在针对 AI 聊天机器人的偏好而非人类品牌认知来调整商品陈列。 John Collison 回应指出，这种现象就像过去为了 SEO 优化而把餐厅直接命名为“泰国菜”（Thai Food）一样，虽然抓人眼球但并非最核心的趋势。当前真正发生、且意义深远的变革，是人们在 AI 应用中进行商品调研（Product Research）的方式彻底重构了传统搜索体验。Collison 直言，到了 2026 年我们居然还在依赖“关键词搜索”简直是“荒唐可笑”（ridiculous）： - 关键词搜索只适用于买书或买 DVD 这种用户已经明确知道具体书名或片名的场景。 - 一旦用户试图购买家具或衣服，要求他们在网页顶部的文本框里敲入几个关键词，根本不符合人类在现实世界中的购物习惯。 相反，AI 驱动的调研允许用户在产品空间中自由探索并施加复杂的现实约束。例如，用户可以说：“我房间里有块空地，想找一件家具，长最多这么宽、高最多这么高，必须能严丝合缝放进去。”这种高度结构化且带有上下文的多维度约束，与 AI 的交互模态天然契合。更重要的是，这打破了大型聚合商（Aggregators）的长期垄断，为中小利基品牌提供了突围机会。以往被埋没的小品牌更容易被 AI 发掘并推荐给消费者，为整个电商生态注入了前所未有的动态活力。 ```mermaid flowchart TD accTitle: 购物搜索范式转移 accDescr: 从传统的单一关键词搜索向基于自然语言、约束条件探索的 AI 智能体搜索的范式转移对比。 subgraph Traditional["传统电商搜索范式"] A["用户已知具体名称"] --> B["输入框关键词匹配"] B --> C["聚合平台与头部品牌垄断"] end subgraph Agentic["智能体电商调研范式"] D["用户复杂现实需求与尺寸限制"] --> E["AI 交互式产品空间探索"] E --> F["精准推荐长尾小品牌与优质匹配物"] end ``` --- ### `12:45` 面向智能体优化（AEO）与 AI 的“好品味” 主持人 Tracy Alloway 追问：如果以往商家是在迎合人类消费者（并辅以针对 Google 的搜索引擎优化），那么在未来，商家该如何面向“智能体”（Agents）进行优化？面向智能体优化究竟意味着什么？ Collison 认为，一方面确实存在类似传统 SEO 的底层诉求，即商家数据必须对智能体保持清晰的“可读性”（legible）与可被发现性；但更关键、也更出人意料的一点在于：当前的 AI 已经展现出了相当敏锐的“品味”（taste）。 现场由此引发了一段关于“氛围编程”（Vibe Coding）的轻松调侃。Collison 调侃 Joe：“你怎么知道一个人在做 Vibe Coding 呢？因为他们一定会忍不住到处跟别人说！”Joe 也笑着连连认领：“我认罪、我认罪。”Collison 接着分析道，正如用 AI 编程时模型在挑选第三方开源代码库和特定底层组件时表现出极强的审美与适配度一样，在电商场景中，AI 也能以极高的品味找到最适合特定任务的工具与商品： - AI 能够发掘出那些人类用户从未听闻、却完美契合工作场景的细分库与工具，使市场配置效率大幅提升。 - Collison 分享了自己的亲身购物经历：他当时想买一个一体化的便携旅行插头转换器兼充电头（integrated travel adapter and power brick），AI 便为他精准推荐了一家名不见经传的小公司产品，体验极佳，而这种小众品牌在传统搜索渠道中几乎不可能被发现。 --- ### `14:38` 广告业会消亡吗？“两端 AI、人在中间”的决策模型 针对智能体带来的去中介化，主持人提出进一步追问：广告一直以来都在创造和形塑文化，经济学界长期争论广告投入究竟是纯粹的无谓损失（deadweight loss）还是积极的信息供给。随着智能体接管选品，未来是否会出现一个“后广告时代”（post-advertising future）？ Collison 明确表示自己对“广告消亡论”持怀疑态度，原因主要有两点： 1. **“人在回路中”（Human in the loop）依然是交易的核心**： 智能体商务并不意味着人类完全退出决策流程。以播客演播室购买麦克风为例，实际的流程并非 AI 全权包办，而是“两端是 AI，中间是人类”： - **前端调研**：AI 执行海量调研，梳理出 3 至 4 种各自存在取舍（Trade-offs）的麦克风方案； - **中端决策**：麦克风没有绝对的最优解，最终拍板决定买哪个的仍然是人类；而在人类做出选择的瞬间，其潜意识里的品牌偏好（例如“这牌子录音音质好”）就会发挥决定性作用，而这种偏好正是由长期广告塑造的； - **后端执行**：选定后，再交由 AI 自动完成下单和结算流程。 因此，只要人类依然是最终决策者，品牌认知度（Brand Affinity）和广告价值就不可替代。 2. **主动搜索与被动消费的区分（Directed vs Undirected Commerce）**： - **定向消费（Directed Commerce）**：用户有明确任务（例如“我要买个自行车装车箱”），这部分过去依赖 Google 搜索广告的流量确实会被 AI 搜索结果部分分流，但未来 AI 应用本身也可能引入广告变现机制。 - **非定向消费（Undirected Commerce）**：人类大量购物行为属于无意识的闲逛，比如刷 Instagram。正如业内调侃的“如今 Instagram 上的广告甚至比原生内容更好看”，精准推荐的广告能不断激发用户的潜在物欲。因此，绝不能在未来的商业世界里“做空广告”。 --- ### `18:40` 机器可读性：商品目录的结构化与品牌个性的平衡 播客在随后的环节切入技术落地细节。Tracy 针对前面提到的“机器可读性”（Machine Readability）提出核心疑问：零售商将自家的产品目录（Catalog）转化成智能体能够顺畅读取的标准格式，其具体流程是怎样的？当前行业内这种机器可读性的标准化程度如何，未来又将走向何方？更重要的是，如果一切参数与描述都被高度标准化以迎合算法解析，品牌和产品那些微妙、独特且吸引人的特质（What makes a product interesting），是否会面临难以被机器精准捕捉并表达出来的困境？ ### `20:00` 让 AI 真正理解与购买商品：从网络声誉到三方交易管道 让大语言模型（LLM）理解并捕捉一款商品的真实“调性”（vibes）极为困难。约翰·科利森指出，要让 AI 智能体全面掌握在售商品，实际上包含两个层次。第一层是“在互联网上建立知名度”，这与面向人类消费者的逻辑完全相同。由于各类大模型在训练阶段就已经“读完了整个互联网”，并且具备实时网络搜索能力，因此详细且信息充沛的产品详情页、真实用户评价以及来自 Wirecutter 等权威机构的评测，都会直接成为 AI 的参考依据。如果一款产品在网上声誉良好、讨论度极高，AI 自然会在推荐时得出“大家都喜欢这款产品”的结论。 第二层则是 Stripe 深度介入的领域——构建让 AI 进行程序化交互的底层机械连接管道（mechanical wiring up）。Stripe 正在与 Google、微软、Meta 以及 OpenAI 等巨头合作推进智能体商务（Agentic Commerce）。这种底层连接主要解决两件事：第一，必须有一种机器兼容的机制，能够实时向卖家发起接口调用，以获取最鲜活准确的信息。AI 仅仅在一年前的模型预训练中知道“某款商品很好”是远远不够的，它必须掌握当前的实时数据——该商品是否仍在促销？具体有哪些 SKU 有现货？目前尺码库存如何？第二，则是打通“智能体结账”（Agentic Checkout），让 AI 能够代表用户真正完成支付闭环。 当科利森提到这里涉及到很多“枯燥的支付细节”（boring payments details）时，节目主持人立刻打断并开玩笑称：“在《Odd Lots》播客里，绝不允许把细节称作枯燥，这里根本没有所谓枯燥的细节，我们就是‘枯燥细节播客’！”科利森笑着顺着话题展开：智能体商务实际上在撮合一场三方交易（Three-way transaction）——即置身于 ChatGPT 等前端应用中的用户、商家以及提供底层协调的 Stripe。在此过程中，既不能强迫用户每次访问新网站都要重新输入一遍支付信息，也不能将用户的核心支付数据随意扩散、分发到各大陌生网站。此外，传统电商网站为了防范黑客和抓取，普遍部署了严密的防爬虫与防 Bot 机制；而现在的变革在于，商家需要主动解绑部分防 Bot 策略，为受信任的 AI 智能体设立类似机场安检的“Bot 预检通道”（TSA PreCheck for bots），通过安全传输一次性支付凭证（One-time use payment credential）来平稳放行交易。 ```mermaid flowchart LR accTitle: 智能体商务三方交易管道 accDescr: 展示用户、AI智能体平台、Stripe管道与商家之间的交互、Bot认证及一次性支付凭证流转。 subgraph Frontend["前端交互层"] U["消费者 (User)"] --> A["AI 智能体 / ChatGPT"] end subgraph Infra["Stripe 交易连接层"] A <--> S["Stripe Agentic Commerce"] S -- "实时库存与SKU查询" --> M S -- "一次性支付凭证" --> M end subgraph Merchant["商家结算生态"] M["商家系统 (Merchant)"] M -. "开放 Bot 快速通道<br/>(TSA PreCheck 机制)".-> S end ``` - **数据新鲜度鸿沟**：预训练模型只拥有过去互联网的静态快照，必须依靠实时 API 才能获取商品现货、尺码与促销状态。 - **安全交易闭环**：通过一次性代币化凭证，既消除了用户在无数独立商户重复绑卡的繁琐，又规避了明文金融信息泄露的风险。 - **爬虫风控逆转**：过去电商视所有自动化 Bot 为安全威胁，如今需要建立白名单准入机制，主动迎合高价值采购智能体。 --- ### `22:48` 计算机控制 vs. 智能体原生：未来网页的“用户”到底是谁？ 既然很多 AI 模型训练的数据集直接包含人类的鼠标轨迹和点击行为，理论上 AI 可以像真人一样在网页上寻找按钮并点击，但主持人提出质疑：如果人类辛辛苦苦做出的网页最终只是被 Bot 抓取和模拟点击，为什么不干脆从一开始就专为 Bot 设计网页？这是否意味着传统意义上的用户体验（UX）正在发生本体论上的位移——所谓“用户”（User）不再是人类，而是智能体？互联网的视觉和交互外观是否会被彻底重塑？ 科利森认为，这触及了当前 AI 领域前沿最核心的一场路线之争：在推动 AI 落地执行现实任务时，到底应该“改造世界以迎合 AI”（让网站对 AI 极其友好、全面文本化/API 化），还是“等待 AI 进化出完美的人机交互能力”（只需给 AI 一个虚拟桌面、鼠标和键盘，即便人类不做任何向下兼容的适配，AI 也能凭借非结构化的计算机使用能力搞定一切）？科利森预计这场辩论将在未来两到三年内见分晓。目前包括 Stripe 在内的许多团队正在全力改造商家管理后台，使其“可被 AI 消费”（consumable by AIs），这主要意味着将以往需要人类点击各种菜单完成的复杂操作转化为纯文本与结构化接口，因为 AI 处理文本的效率远高于视觉图形。 面对主持人关于“这是否类似移动互联网早期拟物化（Skeuomorphism）与扁平化设计之争”的追问，科利森解释道，计算机使用能力（Computer Use）确实提供了一条全自动兼容的底线路径。例如当下极客圈利用运行在 Mac mini 上的 Claude 智能体与无头浏览器（Headless browser），即便商家毫无专门的智能体接口，AI 也能硬生生把商品买下来。然而，企业之所以抢着上线原生智能体商务接口，是因为这是一场争夺用户需求的竞速赛。正如亚马逊平台上带有“Amazon Prime”配送标签的第三方卖家转化率远超普通卖家一样，未来在 AI 对话界面中，如果某些商品能够一键“立即购买”并两天送达，而另一些商品则要求跳转到外部网站让用户自己去摸索点击，消费者无疑会压倒性地流向前者。 - **路线抉择**：端到端计算机控制（UI 自动化）提供开箱即用的旧网络兼容性，而专用 API/文本界面则提供更高的确定性与执行速度。 - **商业紧迫感**：率先接入原生智能体结账的品牌将获得极大的流量与转化优势，重演类似亚马逊 Prime 标识带来的转化率红利。 --- ### `26:15` 终结人类认知负荷：微交易（Microtransactions）与数据按口计费 当智能体主导了商业流程，价格机制是否会被彻底改写？例如消费者只需提示“给我找符合标准的、最便宜的球鞋”，大量需求就会无情流向最低价商品，进而逼迫商家祭出更复杂的动态定价包装，演变成两端机器人之间无休止的算法博弈。科利森对此表示，现在断言博弈演化的最终形态还为时尚早，因为目前的 AI 应用主要被高收入群体率先尝鲜，这并不代表长期稳态下的全社会均衡。 然而，科利森明确指出，AI 商业将带来一个确定性的重大变革：**微交易（Microtransactions）终于要在技术上成为现实**。在此之前，按篇或按次付费在互联网历史上屡屡受挫，其根本症结在于人类的“认知负荷”（Mental Load）过重。如果让一个人去思考是否为了看一篇彭博社的文章支付 40 美分，这种权衡决定的心理阻力极大，大家宁可去买全天通票或包月订阅。但 AI 完全不同——“AI 天生喜欢过度思考，它们热爱严格审视每一个细微的决定”。当决策交给算法，微交易的摩擦力便降为了零。 科利森顺势调侃起彭博社自身的数据体系。他坦言自己非常希望彭博能提供一套面向现代开发者的标准化数据 API，因为在如今风靡的“氛围编程”（Vibe coding）中，一旦涉及金融分析，彭博的终端界面根本不是给 AI 准备的。如果未来能将高质量数据“按口售卖”（selling data by the sip）——比如当用户让 AI 调研霍尔木兹海峡的航运动态时，AI 会在后台自动调用一次彭博数据并支付几分钱的微额费用——这将彻底释放过去无法变现的内容与数据市场。主持人随即幽默地插话宣称，彭博管理层早就把这些重大战略决策全权委托给了他们两位播客主播，科利森当即打趣道：“那请你们赶紧跟迈克（彭博创始人迈克尔·布隆伯格）和团队带个话，我只想让我手下的编程智能体能直接跟彭博的数据对话！” - **微交易的认知瓶颈**：人类不愿意为了几十美分的单次消费反复做心理决策，而 AI 可以无感、极速地权衡微型开销。 - **数据按需付费（By the sip）**：高质量专业数据不必再受限于昂贵的打包月租或硬件终端，AI 驱动的即时微支付能够按查询维度实现颗粒度极细的变现。 --- ### `28:34` “免费漫游”的终结：AI 时代的数据价值与新型互联网税 围绕微交易无摩擦化的讨论，主持人提出了一个更加深远的结构性问题：如果内容的微额变现变得轻而易举，是否会导致整个网络上的各类信息都被明码标价，最终演变成一种人类或智能体在网上浏览时必须沿途缴纳的“数字漫游税”？ 主持人援引了软件工程师圈子内部长久以来的经典隐喻——“免费”（Free）历来包含两种维度：一种是“像免费啤酒一样的免费”（Free as in beer，即完全不收费、零成本获取），另一种则是“像言论自由一样的自由”（Free as in speech，即代码开源、可以任意修改与分发）。而在此基础上，面对生成式 AI 和智能体经济的崛起，主持人认为或许需要提出第三种概念：“像自由漫游一样的自由”（Free as in roaming）。 在当下的技术范式中，任何一段数据对于大模型训练和推理而言都具有潜在价值；甚至用户或 Bot 与数据的每一次交互痕迹（即元数据 Metadata）同样蕴含着巨大的商业利益。当万物皆具训练与引用价值、且微支付基础设施完备之时，那个曾经允许公众自由穿梭、随意抓取与漫游的开放互联网（Free-roaming internet），是否正在无可挽回地走向黄昏？未来的网络生态是否注定会变成一个对每次交互都扣除几亿分之一美分微交易的全新围墙花园？这一关于互联网底层哲学的反思，精准构成了智能体商务落地过程中的深层张力。 ### `30:00` 数据围墙、内容变现与 AI 对免费互联网的冲击 主持人和嘉宾讨论了互联网内容生态的演变。主持人提出疑问：随着有价值的信息逐渐暴露给抓取工具，AI 是否正在终结那个能够“自由漫游”（free-roaming）的公开互联网？ Stripe 联合创始人 John Collison 对此持否定态度。他指出，早在 AI 爆发之前，互联网就从未完全开放过，付费墙（Paywall）和专有数据一直广泛存在。他风趣地打趣道：“我可不记得在 AI 出现之前，有什么‘免费漫游’的彭博（Bloomberg）内容是可以不付钱就看的。”主持人随即笑着接话：“老兄，去买个订阅吧，或者买台彭博终端（Bloomberg Terminal）！”Collison 表示，无论是消费端的优质媒体，还是像 Bloomberg、FactSet、FMP 这样的金融数据提供商，数据本身一直都锁在门后。如今数据所有者并非要关闭网络，而是思考如何在不免费赠送的前提下向外界开放。典型的出路是内容许可协议（Content Licensing Deals）——比如 Reddit 凭借其极具价值的人类社区讨论数据，通过出售数据授权赚得盆满钵满；各大传统媒体出版商也纷纷采取行动，要么起诉 AI 公司，要么与其达成商业合作，甚至两者并存。 ```mermaid flowchart LR accTitle: 数据价值变现路径 accDescr: 专有数据所有者通过内容许可协议或智能体付费调用实现变现的流程图。 subgraph DataOwners["数据所有者"] A["专业金融数据\n(Bloomberg / FactSet)"] B["人类社区论坛\n(Reddit / X)"] C["新闻与媒体机构\n(新闻出版商)"] end subgraph Monetization["变现与接入方式"] D["内容许可协议\n(批量模型训练)"] E["智能体微支付按需采购\n(按需查询与生成报告)"] end DataOwners --> Monetization ``` Collison 还提到了 Stripe 内部在开发者大会（Sessions）上做过的演示：一个编程智能体（Coding Agent）在执行金融研究任务时，会主动出资购买所需数据，最终拼装成一份完整的 PDF 研报。对于主持人提出的“广告衰退导致免费网站不得不立起付费墙”的担忧，Collison 认为人类想浏览的内容与 AI 想要抓取的内容并不相同。优质媒体依然能依靠服务人类获取广告和增量 AI 收入，而真正会死掉的，是那些充斥着“十大排行”、塞满 Taboola 恶心弹窗和垃圾 SEO 页面的低质聚合网站，“没有人会为它们的消亡感到难过”。 - Reddit 通过向 AI 大模型训练商出售数据授权获得了巨额收入。 - 主持人提到曾经开放的 API 生态（如早期 Twitter）如今大幅收紧，如果不用高昂费用订阅企业接口，开发者借助“氛围编码”（Vibe Coding）打造第三方工具的门槛已被大幅提高。 - Collison 澄清道：人类用户目前依然可以在 Reddit、X（原 Twitter）等平台免费漫游并阅读内容，各大平台真正封堵和收费的是用于 AI 训练的大规模抓取行为（Scraping），人类的核心“精神公地”（Intellectual Commons）并未真正被抹去。 --- ### `35:46` 微支付的困境、按量计费与稳定币（Stablecoins）的潜力 对话转向微支付（Microtransactions）的历史与未来。几十年来，业内一直在畅想“花 50 美分阅读单篇文章”的模式，但受限于传统信用卡刷卡手续费，这种小额单笔交易在经济模型上根本无法成立。主持人询问，伴随着近乎零费率的稳定币（Stablecoins）的兴起，微支付是否会在 B2B 或 Agent 场景中真正落地？ Collison 确认了这种“按口啜饮”（Pay-by-the-sip）模式面临的支付障碍：传统信用卡按笔收取固定加百分比手续费，显然不可能支撑每一次 API 请求或单次使用。AI 时代的本质就是按使用量计费（Marginal cost per token），几乎没有哪个大模型提供真正的无上限自助餐服务——无论是 ChatGPT、Claude 的订阅版（有调用频次上限），还是按 API 充值烧 Token，底层都必须按量核算。 为了解决这一问题，行业目前主流的做法是账户充值与额度扣减。Stripe 为此专门收购了 Metronome，以支持这种精细化的用量计费系统（Usage-based billing）。而谈到稳定币，Collison 表达了极大的热情： > “当你想去氛围编码（Vibe Code）写个小网站时……你得去 Cloudflare 买域名，再去 Vercel 搞托管。大家都有过这种到处注册账户、绑定信用卡的痛苦体验。我们认为这种必须到每个服务商处逐一注册账号的繁琐流程理应消失。通过稳定币，你可以直接按单次网络请求（Web Request）或单次使用支付极小金额，彻底免去注册海量服务的麻烦。” - 尽管当前稳定币在日常 B2B 支付中的实际占比还不高，但 Stripe 正在全面铺设对应的基础设施。 - Stripe 收购 Metronome 旨在强化类似电信计费模式（Telco Billing）的额度追踪与用量扣费能力。 --- ### `37:37` 智能体商务的自主权边界与“人机回环”（Human-in-the-Loop） 智能体商务（Agentic Commerce）涵盖了极宽的光谱：从最基础的代填表单，到全权接管数字钱包、完全自主代劳生活消费。主持人抛出许多人担忧的核心问题：如果给智能体赋予消费权限，它“失控跑偏”（Bots going wild/rogue）或者买了不满意的商品，责任到底归谁？法律和赔付责任该如何厘清？ Collison 认为这种对失控的恐慌在现阶段被严重夸大了。他指出，智能体根本不需要完全脱离监管才能创造巨大价值，“人机回环”（Human-in-the-Loop）能够完美兼顾效率与安全： ```mermaid flowchart TD accTitle: 人机回环智能体决策流 accDescr: 智能体负责跨平台繁杂工作并向人类提请最终确认的交互流程。 subgraph AgentAction["智能体处理阶段"] A["理解用户意图"] --> B["跨平台比对方案与价格"] B --> C["自动填充表单与配置依赖"] end subgraph HumanApproval["人类确认阶段"] C --> D{"向人类发起最终确认\n'是否确认花费 $X？'"} D -- "一键按回车授权" --> E["触发底层自动扣款并完成购买"] D -- "拒绝/修改" --> B end ``` 如果 Cloud Code 或代购 Agent 在跨平台比对好一切后，弹出一句提示：“我准备为此花费特定金额，你同意吗？”，用户只需要敲一下回车（Enter）就能完成交易。相比于过去在各个网页间跳转、手动注册账号并重复输入信息，即使用户在最后关头担任审查者，也已经节省了海量的时间与精力。因此，在可预见的消费级场景中，由人类守住最终支付额度和购买标的的闸门，是最稳健且具实用性的主流推进路径。 ### `40:00` 代理授权与企业决策：AI 时代的管理与“AI 精神病”的剂量把控 面对 AI 代理在购物过程中拥有自主决策权可能带来的失控风险，John Collison 借用企业内部治理逻辑给出了一个生动的类比：在 Stripe 这样每年资金流动数以十亿美元计的大型企业中，日常运转本就高度依赖“授权体系（delegated authority）”。企业不可避免会偶尔遇到员工做出荒谬支出决定的情况，而解决之道并不是因噎废食地收回所有权力，而是设立合理的审批流程、支出上限，并在出现离谱行为时给予“敲警钟式的约束”。AI 代理的自主消费逻辑完全可以照此办理——通过清晰的授权额度与人工复核机制，让用户既能享受代理带来的巨大便利，又不必承担失控后果。 主持人和 John 顺势探讨起在技术裂变期管理企业的终极挑战：技术演进速度极端迅速，去年令人兴奋的热点可能转瞬平淡，而前几个月才冒出的概念却可能瞬间成为下一个爆发点。企业究竟该如何在“把握当下的红利（Exploit）”与“面向未来两年的探索（Explore）”之间调配资源？John 强调，最核心的原则是保持极高的灵活性。在当下，如果认知落后三到六个月，信息就已彻底过时。与此同时，必须警惕一种行业通病——AI 似乎格外容易激发人们发表宏大行业预测的“思想领袖癖（thought leadership）”，但 Stripe 的态度是保持敬畏与谦逊，坦承“我们心中的疑问远多于现成的答案”。 更重要的是在组织中把握好对技术的心理预期。John 提出了一个耐人寻味的词——“AI 精神狂热（AI psychosis）”，并引用名言“剂量的多少决定了它是良药还是毒药（The dose makes the poison）”。企业领导者既需要对模型底层的飞跃式能力抱有坚定信念，又必须极其清醒地洞察其局限性。例如，大模型在编写代码方面展现出了无与伦比的生产力跃迁；但如果试图让它们处理复杂的财务分析，就会发现模型在数字运算上简直是“数盲（innumerate）”。它们往往能凭借华丽的语言结构生成一张排版极度精美的电子表格，但在表格填到一半时，数据逻辑就会彻底崩塌——管理者必须拥有敏锐的“指尖感知”，准确分清模型到底在哪些领域靠谱，在哪些领域无能为力。 --- ### `42:45` AI 在企业三大职能部门的落地：研发、销售与职能后勤的分野 谈及企业内部对 AI 的实际消化吸收，John Collison 将企业划分为三大传统职能板块——研发（R&D）、市场开拓（Go-to-Market, GTM）与通用行政支持（G&A），并详细拆解了模型在不同领域的落地生态： ```mermaid flowchart TD accTitle: AI 在企业三大核心职能中的落地差异 accDescr: 展示软件研发、销售拓展与后台职能应用 AI 时的核心驱动与面临挑战。 subgraph RD ["研发与工程 (R&D)"] RD1["工程师热爱先进工具"] --> RD2["自我改进的正向闭环"] RD2 --> RD3["点击'修复 Bug'代理自动解决"] end subgraph GTM ["市场销售 (Go-to-Market)"] GTM1["业务产出高度可量化"] --> GTM2["规避古德哈特定律陷阱"] GTM2 --> GTM3["人均产能指标显著提升"] end subgraph GA ["后台职能 (G&A：法务/财务/风控)"] GA1["重构底层数据基础设施"] --> GA2["兼顾数据调用与安全隔离"] GA2 --> GA3["公开法条表现优异 / 专有数据表现棘手"] end ``` 在软件工程（R&D）部门，AI 的渗透进行得最为顺畅。工程师群体天生对能提升效率的工具充满狂热，任何稍有自我追求的程序员都会主动拥抱 AI，因此极少出现抵触或落后者。整个技术团队迅速进入了一种自我优化的正向循环：日常写代码的工程师也在时刻思考如何用 AI 自动化研发流程。在 Stripe，如今许多 Bug 的修复甚至不需要任何人打开本地代码编辑器，工程师只需在界面中点击“修复此 Bug”按钮，代理便会在后台端到端自主完成修复。 销售与市场端（GTM）同样进展迅猛，其核心优势在于这是科技企业中最容易被精准量化的领域。在研发或法务部门，衡量个人生产力的指标极易沦为“古德哈特定律（Goodhart's Law）”的牺牲品（即一旦某项度量变成目标，它就不再是一个好的度量）；但销售团队不同，每个人的销售业绩与产能数字一目了然。Stripe 的销售团队对 AI 工具展现出了极高热情，尽管目前还无法在因果关系上完全归功于模型，但整体人均产出数字非常亮眼。 真正推进最缓慢、最棘手的则是法务、财务和风控等后台职能（G&A）。阻碍首先来自繁重的数据基础设施改造：过去这些敏感财务与风控数据都是处于严密的物理与权限隔离之下，现在要让 AI 能够消化使用，必须投入海量工程进行清洗与权限架构重塑，在开放给模型的同时确保数据不泄露。此外，模型能力的表现也极其不均衡：若任务能被抽象为编程逻辑，或是处理公开的法律与税收条文（这些已在海量公开语料中充分训练过），模型表现非常出色；但面对 Stripe 内部专有的财务明细与特定业务逻辑时，目前的模型依然很难稳定输出高质量的结果。 --- ### `45:00` 智能爆炸与科斯定理：小团队时代与商业活力的迸发 当话题回到最初的核心——“互联网从根本上建立在商业之上，商业模式的剧变将如何重塑整个互联网版图”时，John Collison 明确表达了他对“智能爆炸（intelligence explosion）”的乐观预期。他指出，Stripe 作为底层支付枢纽，已经在数据前沿感受到了前所未有的创业活力井喷。 - **创业注册量暴增**：在宏观经济整体增长平缓的背景下，Stripe 数据显示第一季度在其平台上创建的新业务数量同比激增了 71%。 - **应用生态复苏**：此前多年处于平缓停滞状态的应用商店（App Store）新应用发布数量，目前也迎来了同步的高速反弹。 - **创业门槛骤降**：过去在大型企业中，员工即便发现某些业务流程极其愚蠢、自己能做得更好，也往往受限于人力资本而无法单干；如今个人能随时从外部低成本调用海量智力资源，辞职下海制造更好竞品的门槛被大幅拉低。 - **科斯定理与企业规模**：在近期的 Stripe Sessions 大会上曾深入探讨过这一命题——在科斯（Ronald Coase）企业理论的视角下，随着智能代理使外部协调与交易成本急剧下降，未来可能会出现更多规模更小、但数量更加庞大、相互敏捷协同的高动态型企业。 在热烈的讨论中，主持人 Joe 对谈话意犹未尽地鸣谢收尾，并邀请 John 后续再次做客播客。 --- ### `46:15` 主持人复盘：算法困境、LLM 记忆的双刃剑与互联网“免费时代”的终结 采访结束后，Joe Weisenthal 与 Tracy Alloway 针对访谈展开了深度复盘。Tracy 首先谈到了代理搜索在可发现性（Discoverability）上的潜在陷阱。她以流媒体算法为例：用户在 Netflix 上看的第一部电影往往具有决定性意义，若不小心点了一部烂片，推荐系统就会永久将用户圈禁在特定的刻板标签中；算法往往倾向于聚类在既定模式里，使人们更难发现真正新颖的产品（不过她也认可 YouTube 的算法在挖掘新事物方面表现相对突出）。 Joe 随即指出，大语言模型的“长期记忆功能”也是一把双刃剑。适度了解上下文固然能提升交互体验，但有时模型会执拗地把用户半年前提出的无关查询硬塞进当下的对话中，反而造成了严重干扰。 两位主持人敏锐地察觉到，一个不可忽视的商业大变局正在发生——当商家不再只需要取悦人类，而是必须研究如何向 AI 代理“做营销”时，零售业将迎来全新维度的 SEO 革命。而在宏观层面上，无缝的小额微交易（seamless microtransactions）基础设施的成熟，很可能促使所有内容创作者与 API 服务商全面开启收费模式。那张长期以来依靠注意力和无意识滑动所维持的“免费互联网”，或许正不可逆转地走向终结。 ### `50:00` API 直付与“Bot 对战 Bot”的荒诞未来 主持人回顾了近期 Stripe 针对新兴代理生态所发布的一系列功能更新。在当下这轮 AI 变革中，存在许多极其具体却令人烦躁的技术痛点——例如代理在互联网上自主操作时涉及的域名解析问题，以及直接向 API 接口发起支付而不再需要人类打开网页、手动输入信用卡来充值积分（credits）的能力。这些在传统视角下看似细枝末节的琐事，在当下的现代网络环境中却成了实打实的巨大阻碍。主持人感叹，很高兴看到 Stripe 敏锐捕捉并出手解决这些新出现的“当代烦恼”。 由此，讨论延伸到了对未来经济形态的幽默讽刺：未来的很大一部分网络活动，很可能演变成纯粹的“机器代理互相对轰与谈判”（bots negotiating with each other）。最典型的现实雏形已经出现在保险行业——保险公司正在部署 AI 模型来“批量拒赔”（bulk deny claims）；与此针锋相对，市面上涌现出一批 AI 初创公司，专门利用算法帮助投保人自动化对抗并申诉这些拒赔决定。在双方代理永无休止的算法博弈中，真正躺着赚钱的赢家似乎只有三类人：提供强大算力与存储支持的韩国半导体芯片巨头、坐收 OpenAI 与 Anthropic 高薪员工房租的旧金山婴儿潮一代房东，以及庞大的医疗健康产业本身。 ```mermaid flowchart TD accTitle: Bot-to-Bot Conflict and Real Beneficiaries accDescr: 展现保险公司AI与消费者维权AI相互博弈的闭环，以及真正获利的底层硬件商与旧金山房东。 subgraph Battle["算法对轰战场"] InsurerBot["保险公司 AI<br/>（批量自动化拒赔）"] ConsumerBot["维权初创 AI<br/>（自动化反驳与申诉）"] InsurerBot <-->|"高频博弈与消耗"| ConsumerBot end subgraph Winners["背后的真正获利者"] KoreaChips["韩国半导体芯片厂商<br/>（供应高性能存储与算力芯片）"] SFLandlords["旧金山老牌房东<br/>（坐收 AI 独角兽工程师高昂租金）"] Healthcare["医疗健康产业"] end Battle -.->|"持续拉动硬件需求"| KoreaChips Battle -.->|"支撑高估值企业用人"| SFLandlords ``` - 核心痛点消除：Stripe 推出的机器直连支付方案，使 AI 代理能够通过 API 直接完成资金结算与点数充值，彻底告别了“人类打开浏览器跳转、手动输入信用卡”的原始繁琐流程。 - 代理博弈闭环：AI 自动化拒赔与 AI 自动化维权申诉形成对抗性博弈，直接带动了算力与基础设施消耗的指数级膨胀。 --- ### `50:45` 播客结语、制作团队致谢与尾部广播 两位主持人 Tracy Alloway 与 Joe Weisenthal 结束了本期对谈，并向观众致谢。随后节目播报了嘉宾与制作团队的社交平台联络方式以及彭博社相关资源的订阅指引，最后串接了关于定制周边服务与《华尔街周刊》（Wall Street Week）的赞助推广信息。 - **社交媒体与订阅信息**： - 主持人：Tracy Alloway（社交平台账号：@tracyalloway）与 Joe Weisenthal（账号：@TheStalwart）。 - 本期嘉宾：Stripe 联合创始人 John Collison（账号：@collision）。 - 制作团队成员：Carmen Rodriguez（@CarmenArmin）、Dashiell Bennett（@dashbot）、Kael Brooks 以及 Kevin Lozano（@kevinlloydlozano）。 - 互动渠道：彭博官网专区（Bloomberg.com/oddLots）提供每日新闻通讯及往期所有节目归档；官方 Discord 社区支持 24 小时全天候话题探讨；彭博订阅用户可通过 Apple Podcasts 彭博专属频道收听无广告版本。 - **赞助商推广**： - 宣传品与品牌定制平台 4imprint 提供服装、饮具、科技产品及托特包等全品类定制，支持“360度满意保障”。 - 彭博旗下由 David Weston 主持的《华尔街周刊》（Wall Street Week）播客推介，节目每周聚焦地缘政治、央行决策、AI 技术革新与能源基础设施，深度访谈顶级 CEO、经济学家与政策制定者。 ## Notable quotes > `05:08` broadly it is AI buying something for you — John Collison > `06:17` They love to say, 'AI will book all your activities on vacation.' But the reality is people spend all year anticipating and fantasizing about their vacation, researching and booking it is half the fun! ... It’s like the robots came for the jobs, but people actually really wanted to keep the job of scrolling. — John Collison > `10:00` ridiculous — John Collison > `12:45` How do you know someone is vibe-coding? Because they can't stop telling you about it! — John Collison > `20:00` TSA PreCheck for bots — John Collison > `26:15` AI天生喜欢过度思考，它们热爱严格审视每一个细微的决定 — John Collison > `26:15` selling data by the sip — John Collison > `30:00` 我可不记得在 AI 出现之前，有什么‘免费漫游’的彭博（Bloomberg）内容是可以不付钱就看的。 — John Collison > `35:46` 当你想去氛围编码（Vibe Code）写个小网站时……你得去 Cloudflare 买域名，再去 Vercel 搞托管。大家都有过这种到处注册账户、绑定信用卡的痛苦体验。我们认为这种必须到每个服务商处逐一注册账号的繁琐流程理应消失。通过稳定币，你可以直接按单次网络请求（Web Request）或单次使用支付极小金额，彻底免去注册海量服务的麻烦。 — John Collison > `40:00` The dose makes the poison — John Collison ## Mentioned John Collison, Joe Weisenthal, Tracy Alloway, Stripe, Cloudflare, GoDaddy, Bluehost, ChatGPT, Claude, Claude Code, OpenAI, Anthropic, Google, Microsoft, Meta, Bloomberg, FactSet, FMP, Reddit, X, Wirecutter, Metronome, Vercel, Apple Pay, Google Pay, PayPal, Netflix, YouTube, Instagram, Nike, Ronald Coase, Michael Bloomberg, David Weston, Odd Lots, Wall Street Week, 4imprint ## Video description <p>The internet is made for shopping. For years, the main inputs for e-commerce transactions involved targeted ads, algorithmic recommendations, SEO, and lots of mindless scrolling. But agentic commerce might represent a sea change for e-commerce: With the rise of AI agents doing shopping on behalf of consumers, how are retailers going to adapt? John Collison, co-founder of the financial services and payment processing company Stripe, has first-hand experience with all the ways e-commerce has changed in the last decade, and he thinks agentic commerce is going to completely transform the online shopping experience. On this episode, we speak to Collison about how AI has already changed the way consumers make purchasing decisions, why keyword search is a "ridiculous" way to find things to buy, what it means when brands will have to appeal to AI agents as opposed to human buyers, and if AI agents can truly mimic human taste.</p> <p><a href="http://bloomberg.com/subscriptions/oddlots" data-link-cookie="1" data-destination="http://bloomberg.com/subscriptions/oddlots">Subscribe to the Odd Lots Newsletter</a><br>Join the conversation: <a href="https://discord.gg/oddlots" data-destination="https://discord.gg/oddlots">discord.gg/oddlots</a></p><p>See <a href="https://omnystudio.com/listener">omnystudio.com/listener</a> for privacy information.</p> [Source](https://podtrac.com/pts/redirect.mp3/tracking.swap.fm/track/UVBrz8bN8aM2Xe47PEPu/traffic.omny.fm/d/clips/e73c998e-6e60-432f-8610-ae210140c5b1/8a94442e-5a74-4fa2-8b8d-ae27003a8d6b/ba6e5213-7720-46c6-87a5-b44c0032e6bc/audio.mp3?in_playlist=982f5071-765c-403d-969d-ae27003a8d83)

[5:08] John Collison: broadly it is AI buying something for you [6:17] John Collison: They love to say, 'AI will book all your activities on vacation.' But the reality is people spend all year anticipating and fantasizing about their vacation, researching and booking it is half the fun! ... It’s like the robots came for the jobs, but people actually really wanted to keep the job of scrolling. [10:00] John Collison: ridiculous [12:45] John Collison: How do you know someone is vibe-coding? Because they can't stop telling you about it! [20:00] John Collison: TSA PreCheck for bots [26:15] John Collison: AI天生喜欢过度思考，它们热爱严格审视每一个细微的决定 [26:15] John Collison: selling data by the sip [30:00] John Collison: 我可不记得在 AI 出现之前，有什么‘免费漫游’的彭博（Bloomberg）内容是可以不付钱就看的。 [35:46] John Collison: 当你想去氛围编码（Vibe Code）写个小网站时……你得去 Cloudflare 买域名，再去 Vercel 搞托管。大家都有过这种到处注册账户、绑定信用卡的痛苦体验。我们认为这种必须到每个服务商处逐一注册账号的繁琐流程理应消失。通过稳定币，你可以直接按单次网络请求（Web Request）或单次使用支付极小金额，彻底免去注册海量服务的麻烦。 [40:00] John Collison: The dose makes the poison

[5:00] John Collison: I think in the history of technology, the lower friction option tends to win out. [25:00] John Collison: The mental load has traditionally been too high for humans, but AIs love overthinking things. They love scrutinizing every little decision. [38:15] John Collison: You still have a human in the loop deciding, making the final decision of both what thing I buy and am I okay spending this much money, but the AIs can still be hugely beneficial as part of your shopping journey... [40:15] John Collison: You need to really believe in the capabilities of the models, but you also have to be willing to reason about the limits... They can put together something, they can build a really nice spreadsheet, but the numbers just go completely wrong halfway through the spreadsheet. [50:00] Joe Weisenthal: It just feels like it's just going to be bots. The only money makers are going to be chip companies in Korea... and old boomer landlords in San Francisco because the employees of OpenAI and Anthropic need a place to live.

[0:00] Whatever your goal, trade show giveaways, client gifts, or team gear, 4imprint has the promo products to match. With thousands of options, from apparel and drinkware to tech and totes, it's easy to find the right fit for your brand and budget, with standout choices at every price point. And with their 360-degree guarantee, you can be 4imprint certain your order will show up just right, right on time. Explore more at 4imprint.com. 4imprint. 4certain. The thing about AI for business? It may not automatically fit the way your business works. At IBM, we've seen this firsthand. But by embedding AI across HR, IT, and procurement processes, we've reduced costs by millions, slash repetitive tasks, and freed thousands of hours for strategic work. Now we're helping companies get smarter by putting AI where it actually pays off, deep in the work that moves the business. Let's create smarter business, IBM. Healthcare doesn always work great If you ever waited on a refill or couldn schedule an appointment you get it That the kind of stuff Optum is changing They using data and technology to integrate patient care pharmacy and everything else So healthcare is connected not complicated What that look like Cheaper prescriptions that are easier to get and care that looks at the whole person how you need it Optum is helping make healthcare work as one for everyone. Learn more at business.optum.com. Bloomberg Audio Studios, podcasts, radio, news. Hello and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal. And I'm Tracy Alloway. So, Tracy, another episode where I can mention that I've been vibe-coding. But you know what? Here's something that I may have mentioned before. I'm sure you have, Joe. But I'm not 100% sure. Okay. I think I've mentioned it. It's actually very easy. You just come up with an idea and you type it in English and then it's there. You know the hardest part? registering a domain name. Yeah, you've complained about this. Everything else is so smooth, right? You just type it in English and then you have your little app or whatever and then it's like, oh, I want to put this somewhere on the web and then you go to a domain name registrar. I had been using GoDaddy for a long time and then it like then set this field for this field and this field for this field and toggle this and this worker and this pages and the A field and the app field and the B field And I swear that like it so maddening I just hated it so much. It's so annoying. It's the hardest part of the whole aspect. I have terrible memories as well about dealing with Bluehost and things like that. Is this the first time you've ever set up a domain? No, that's the thing. This is the other thing. I've done it before. Like if I've done it before and it's still incredibly annoying. Progress comes slow to domain registration. I mean, there's probably good reason for it to be complicated. You don't want it to switch in the wrong direction. Anyway, I was very excited. It's a couple of weeks ago. It was actually while the two of us were in Madrid, the company Stripe, the payments company, announced that there was going to be some agreement with Cloudflare, which does a lot of web hosting stuff, where you could just have the agent buy a domain name for you. And so you think of one and you just type it and buy this domain name and host it there. And I mean, oh, my God, I wrote about it. The newsletter is like this would be so this would have been so nice a month. Yet another thing that you will never have to do again. Exactly. Let another. So I mean I saw the Stripe news and it generated a lot of buzz and it all comes under the umbrella I guess of agentic commerce So this idea that you can have agents that are actually directly transacting on your behalf with other agents or other websites And I have to say as someone who shops online a lot this is the first time I feel disintermediated by AI I'm actually particularly excited about this episode because I'm very curious, basically, what you specifically will ask as someone, as a person. As a professional shopper on the web. I was going to say as a person of taste. As a person of taste, I have seen you say scrolling very nice paper, like websites for wallpapers and things like that. And I'm curious, what does it mean for that? If your AI bot two years from now knows you so well, it's like, Tracy, I thought you needed this new wallpaper because you're doing this. I just went ahead and ordered it to be delivered tomorrow. This changes the world. It brings up so many interesting questions about discoverability and taste, as you mentioned, And also just the nature of the internet, because when I think about the internet, like a lot of other things, it's basically built to sell you things, whether that's products or information, content. And we know that it also runs on those sales in the sense that a lot of the internet is driven by advertising, which is related to traffic. Obviously, we actually did that episode. [5:00] way back about how the internet runs on millions of like ad auctions. That's right. And so this topic of agentic commerce and the payment system, obviously it's of interest to anyone who shops and does stuff online, but it's also really relevant to the actual evolution of the internet. Couldn't agree more. And I would go further and say it's really important for like culture itself. Advertising is a creation of culture. Ads are influential. Ads do all this stuff. What if there's a changing advertising, all this kind of stuff. Anyway, let's stop hearing from ourselves. We really do have the perfect guest, someone we've wanted to talk to for a long time, someone who's right in the thick of it and all this stuff with agentic commerce and payments. We're going to be speaking with John Collison, the president and co-founder of Stripe. So, John, thank you so much for coming on AdLots. Great to be here. I'm very excited. Yeah, buzzwords, et cetera. What is, quote, agentic commerce, unquote? Yeah, so there's a bit of a definitional question, but broadly it is AI buying something for you. And we think about it in two categories. One is on the consumer side of things. This is where you were skeptical whether Tracy will ever be an adopter where you research something in chat GPT or Claude or what have you and then go buy it there or have it buy it for you But then the second is your Cloudflare example And so it B2B or developer agentic commerce where you're just trying to have CloudCo do something for you. And as part of it, it needs to buy resources. And we can talk about that because they're quite different. Yeah, Tracy, and we'll get to this, but the resources is another thing I've whined about, which is that when you do an API key, you have an idea and you have to go to the web. to buy the API. Sometimes you have to enter a credit card number. That's really annoying, et cetera. But we'll get into that. But that is another big element of it that I feel like is getting closer to being solved. Okay. Well, on this credit card point, I mean, the history of online payments is basically a continuous elimination of the friction, right? And going from, I have to enter my credit card details to I have to log into PayPal to now I can just basically click a button, whether it's Apple Pay or something else. Is agentic commerce Is this just another leg of the frictionless process or is something fundamental in your view changing here? Both. So part of agentic commerce is just reducing the friction. And you right that the general direction of travel over time you go back to the introduction of credit cards when the earliest adopters of computer networks and then the original e and now like you saying wallets like Apple Pay and Google Pay and we have a protocol link and things like that There's generally been this elimination of friction. And our view is that at a very minimum, we're going to see, and Tracy, I think this is where you can maybe get behind us, is leave aside any agent decision making that's happening. just the final step of buying something you know you might have researched a product in an ai app before where you're going back and forth i was trying to buy a bike case recently to travel with my bike and you're asking all these questions hard versus soft and do you need to take off the handlebars and what's the right one it's finding all these cool little niche brands i think it's actually quite good for smaller merchants by the way because you have the ability for them to be discovered that they mightn't have had otherwise but anyway when you find the product at the very end, do you really then want to go and be filling out all these web form fields and things like that? Or do you want to just say, yeah, that sounds good. Buy it for me in this size. And our bet is that people will want the lower friction option. I think in the history of technology, the lower friction option tends to win out. And so we think that clearly will happen and is happening to some degree already It just the agent filling out the forms for you I think where you get this interesting question is and when people hear agentic commerce I think their mind jumps to the agent actually having some autonomy and doing some decision making for you. I think there the problem is people somehow jump to the examples. Anyone who talks about this phenomenon, they pick terrible examples. They're like, oh, the agent will book, you know, all the activities on your vacation for you. It's like people think about, they daydream about their vacation all year. Yeah, that's the fun part. Researching and booking is the fun part. That's the fun part. Or similarly with women's fashion, it's like the whole point is the scrolling, not just the buying. And so people pick these, you know, super enjoy, like it's the robots taking the jobs and people want to keep those scrolling jobs. And so when you think about robots having autonomy, I would instead proffer some examples where you really don't need to be. You give the AI a recipe and say, buy the things that we need to make this tonight. And yeah, the AI can probably guess that you have salt and you have garlic at home, but you need the ground pork. Or again, then migrating to the B2B example, no one needs to configure yet another domain and go through all those forms. If Cloud Codes can work with Cloudflare to buy the domain for you and set it up, I think Joe will be much happier and not feel like he's missed out on a [10:00] life experience. I want to talk more about the B2B stuff, but let's for a moment stick around on the B2C element because I think there are gradations of this. So one is just another elimination of the friction layer. At the far end of the spectrum, conceivably, there will be things where the AI truly does anticipatory buy you things that you didn't know you need, etc. And then there might be like some middle. I saw this screenshot recently and I couldn't find it before the episode, someone tweeted it, but some retailer, I think it was a shoe retailer, had put a bunch of shoes on their website. But instead of putting any brand names on the shoe, like we would say like a Nike Pegasus or a Nike Air Jordan, all of the shoes were like hiking boot that's good for semi-serious hiker. Do you see this currently, where we are on the agentic commerce trajectory. Do you see this currently where there's already a change in how retailers are presenting or displaying or promoting their wares such that they're more attuned for what the chatbot would identify or the AI would identify as cool versus what a human influenced by brands would pick up as cool Yes and no I mean there definitely changes i think some of the stuff that most notable is often not the most relevant it a little bit sounds like what you're describing is a little bit like the thai restaurant calls thai food that's right yeah you know do best in the seo but i think what we see as a kind of maybe a more interesting direction of travel is again what's happening already some amount of the form fields being filled out for you and kind of the friction reduction. And a huge amount of, again, I presume you guys have both done this, is research in AI apps. And the way I would frame that is just keyword search is ridiculous. It's ridiculous that we got to the year 2026 relying on keyword search, where that makes sense for buying a book or a DVD where you know the title of the book or DVD that you're trying to buy. But that's about the limit of keyword search. And so the fact that anyone was trying to buy furniture or clothes or anything like that and there was this text box at the top of the website where you put in your keywords it's like that's just that's not how anyone shops in the real world things like that and so we're seeing this ai powered research where people can explore product space and start to give constraints and you have this textual search interface but you end up with kind of much higher quality search where it like okay i have a spot in this room where I looking for a piece of furniture that this wide max and this high because I need to fit it in there That the kind of thing that just much better suited to the AI modality So you see different kinds of product research possible. And like we said, it's very helpful to smaller brands because people aren't necessarily stuck behind the traditional aggregators anymore. They can break out. And again, I have found in my little research projects that I've done, you often get these small brands surfaced that you might not have heard of before. And so we think it's actually probably quite good for the dynamism. Well, let me ask Joe's question in a slightly different way, which is if we assume that in the future, if I'm a business selling something online, previously, I was selling to humans more or less, like maybe I'd optimize for Google search results or whatever. But like mostly I'm trying to appeal to humans. And now I'm trying to appeal more to agents. What is optimizing for agents actually look like? Part of it is similar to traditional SEO, right? So you want to be legible to those agents. You want to be discovered by them. But part of what at least I seen is that the AIs have pretty good taste Joe I don know if you found this in your vibe coding By the way Joe I reminded of it It sort of like the old joke how do you know if someone vibe codes They tell you about it Exactly Guilty guilty guilty I guessing you found that the AIs are very tasteful at picking libraries and picking specific subcomponents to use. And similarly, in our kind of early experience of the agent of commerce, they're quite tasteful in finding tools that suit the job pretty well. And so I think you may get this more efficient market phenomenon where at least in picking coding libraries or something like that, the AIs have quite, like I find when I do stuff, they're discovering libraries that are perfect for the job that I hadn't even heard of or something like that. I think you're starting to see the same phenomenon with products where again, I have had the experience of, you know, I was trying to find a travel adapter and I found it from this tiny company that I had never heard of, but doing exactly what I want, where it's like an integrated travel adapter in Powerbrick. It's really good. and I would never have come across that company otherwise. This is incredibly important because, as I mentioned in the beginning, advertising creates culture. It's part of culture, especially iconic advertising. And maybe it'll be a little while before the Nikes of the world, the Nikes and Apples of the world, they're going to be advertising for a long time. But do you see this eventually as sort of like one of the things I think that the [15:00] internet appreciates about Stripe as a company is you and your brother's commitment to interesting discussions of economics and so forth. And you talk about econ history. There are great debates about whether ad spending is a huge deadweight loss on the economy or whether it's something positive, et cetera. And I'm curious, could you see a sort of post-advertising future, you know, at some point on the horizon? No, I think we're quite skeptical of that for two reasons one again with agentic commerce we don't think the human goes away and the human in the loop goes away and there's a huge amount of agentic commerce that can happen like we're describing where you still have the human as the end decider and so it will likely be the case that if you are searching for something if you have a new microphone for the odd lots studio you might ask it to go research and it'll present three or four different options with different trade-offs There's no right microphone that's going to be kind of concentrated on things like that. And then ultimately, you will make a call and it'll go buy it for you. And when you make that call, some brand preference you have. Oh, sure. I just know that they like the ones with the really good sound quality Not clear why you have that association Maybe there some advertising in there somewhere But so we think brand preference and as part of that advertising will clearly stick around Because again, it's definitely not going to be the case that agent of commerce means no humans make buying decisions. That's absolutely not our case. But again, this may be a good example. You want to buy new microphones for the studio. You do a lot of AI research. You then have a human decide. And then you do a lot of AI execution. And so it's AI on both sides and then human decision-making in the middle. brand affinity really matters in that world the other thing of course is commerce can be directed or undirected where in the directed side like i was saying i'm trying to buy a bike case it's like this is a thing that i know i need to go do i'm going to go do some research previously you know a google ad might have been relevant to me whereas now it's what appears in the uh the ai search results obviously that does remove some advertising but for now at least maybe some of the ai apps at least choose to bring back some advertising in those cases but then a lot of shopping is undirected where you're scrolling Instagram. And the joke goes that these days on Instagram, the ads are better than the organic content because they've gotten so good with the targeting and there's nice reassuring things that you might want to buy. And there that is advertising and people are going to stay scrolling social apps and finding cool products they want to buy. So I wouldn be short advertising in this world Whether you're planning a big tech event, launching a new campaign, or just stocking up on team gear, finding the right promotional products makes all the difference. 4imprint offers thousands of options. From on-trend apparel and premium drinkware to tech, totes, and giveaways so you can find the right fit for any audience, purpose, or budget. You can customize it all. Your logo, your message, your look. And many items come with no setup charge to help you save. And if you're really watching the bottom line, you'll find standout choices at every price point so you can make a real impact while staying on budget. Plus, you'll get expert help, fast turnaround times, and their 360-degree guarantee so you can be four-imprint certain your order will arrive on time and look exactly right. Whatever your goal, 4imprint makes it easy to find your perfect promo match. Explore the possibilities today at 4imprint.com. 4imprint, for certain. OnDeck is built to back small businesses like yours. Whether you're buying equipment, expanding your team or bridging cash flow gaps OnDeck loans up to help make it happen fast Rated A by the Better Business Bureau and earning thousands of five trust pilot reviews OnDeck delivers funding you can count on. Apply in minutes at OnDeck.com. Depending on certain loan attributes, your business loan may be issued by OnDeck or Celtic Bank. OnDeck does not lend in North Dakota. All loans and amounts subject to lender approval. Get the news you need in just 15 minutes. Start your day with Bloomberg Daybreak, the podcast with a global view on the stories that matter. I'm Nathan Hager. And I'm Karen Moscow. Join us each morning for curated stories on current events, politics, business and foreign relations. Plus one conversation on the day's biggest developments all in just 15 minutes. Subscribe to Bloomberg Daybreak for a precise, thoughtful take on the stories that matter. Listen to Bloomberg Daybreak each morning on Apple, Spotify, or anywhere you listen. You mentioned machine readability earlier. Talk to us about the actual process of converting your catalog into something that an agent can actually read and how standardized that is at the moment and how standardized do you expect it to be in the future? And then also, if you're standardizing for machine readability, I feel like it's difficult to capture some of what might make a product interesting. [20:00] difficult to like code the actual vibes of a product, even with LLMs. I just find there's something difficult to capture. So talk to us about how you make it so that agents fully understand what is available to them. Yeah, so there's two levels here. The first is just being known in the same way that you'd be known to humans, because obviously all the LLMs have read the whole internet. That's what they're trained on. And they can do web searches. And so if you have good product pages that are informative and lay out the details of what the product does, if you have reviews, you know, if there's like a bunch of Wirecutter reviews, things like that, you've probably seen this in research searches where, you know, the AI app will say, you know, people love this product, and they say XYZ, but they've read the whole internet. And so being known, you know, I'm a big deal on the internet helps with the AI apps as it does for humans. That's kind of step one. Step two and where Stripe gets more involved is you need a lot of the mechanical wiring up so that the AIs can do stuff that are more programmatic. And this is where our agent of commerce speech that we had an announcement about. And this is what we working on with Google and Microsoft and Meta and OpenAI and all those folks to actually do some of the commerce wiring up And there I say it two things One is you need a machine compatible way to get the very latest information about the product details. So it's one thing for the AI to have in its training run, maybe a year ago, read the whole internet and know that this product is good. You need to be able to tell the AI, is it still on sale? What SKUs are in stock? Which sizes we have right now and things like that. And so that kind of up-to-date being able to make real-time calls to the seller so that the ai has fresh correct information that it can present that's thing one and then thing two is wiring up the agentic checkout so the ai can actually buy it and here you get into a bunch of boring like payments details but for example there's no boring no no no you're not allowed to you don't want to be you're not allowed to describe oh that's true details as boring on adlots on adlots there's There's no such thing as boring details. This is the Boring Details Podcast. Yeah, so start. Yeah, keep going. Start. Okay, so then for actually kind of handling a three-way transaction where there's you in ChatGPT buying a product from this merchant or something like that, we're facilitating that transaction. And there you want to be, you don't want to be re-entering your payment details on each website. At the same time you don want to be flinging your payment details around across different websites And so we built all the wiring up And also don forget a lot of websites have historically bots used to be a bad thing They used to discriminate against bots They would have bot protections and things like that. Whereas now it's like, OK, no, we actually want agents to be able to do this. And so you're unwinding some of the bot protections and having a little TSA pre for the bots that we work with, a way to kind of securely transfer a one-time use payment credential and things like this. There's a lot of wiring up beneath the scenes that businesses have to do. This is actually maybe like a good spot to slightly seg over into the B2B element of agentic commerce, because this gets to another thing that I think about. You know, I know one of the things that available training data is literally just looking at user mouse behavior, right? Because then theoretically, the AI model can go to a website and click the buttons just the way a human wants. But it all, it seems silly to me. Like ultimately, if we're building a human, if we're building websites for human use and then a bot's going to actually scrape them, why not just actually build it for bots in the first place? What are we seeing there? What are you seeing from retailers where the fundamental experience is it going to be reshaped such that what we call the UX the user experience is not really the you is not the person but the you is the bot And is that changing how the Internet is just going to look So you're getting to a very core debate that's happening in AI right now, like some of these kind of questions at the frontier of AI as to where things go for practical purposes. But one of them is when you think about AI adoption for real world tasks, there's almost a race between maybe we make the world consumable by AI and we make the websites friendly for AIs to navigate and things like that, or maybe just the AIs get good enough at unstructured computer use that you give them a desktop and you give them a mouse and a keyboard and there's no backwards compatibility that we built, but they just end up figuring it out anyway. And it feels like over the next year or two or maybe three, this question will be resolved. right now a lot of people including us are working on making the stripe dashboard merchant website things like that consumable by the ais and what that means basically is text where the ais are very good at working with text and so you need a textual interface for a bunch of things that previously might have been clicking around on a website but sorry just to be clear this reminds me a [25:00] a bit about the debate whether like apps in the app store should be skeuomorphic or not. But just to be clear, Stripe is kind of taking a bet on the other side that actually the websites that look like they're designed for humans will not be the way to place your where to place your chips. Well, it's an option, right? Because if the AIs get really good at computer use quickly, then it's good. You know, then we have that answer. And you're seeing this a little bit when people get their open claw to go buy something and it's running on their Mac mini at home, The merchant might say, oh, but I don't have an agentic interface. It's like, well, we bought it anyway. The claw just went and did it and used a headless browser. And so you don't need to do anything to enable the computer use world because the whole point is that it's fully backwards compatible with the current world. But in the meantime, I think people want to have these really powerful agentic experiences. It's a bit of a race, right? I think businesses correctly want to be the first ones to capture all this demand. because I think it'd be the case if you start getting presented results in an AI app and it's like, okay, here are a bunch of options. This one, you can just click buy now and it'll arrive in two days. Or this one, you have to go to the website and click around and blah, blah, blah, blah. I think you'll start seeing some consumer preference for the ones that click one, kind of like on Amazon when there like a seller that has Amazon Prime in the third sellers you way more likely to pick that And Amazon has done studies on this You can imagine an analogous world within the AI apps where the businesses that support agentic commerce sooner outperform the ones that don't. Can you talk to us about how agentic commerce could impact prices? And we all know that already when people are researching for products, you might put in a prompt saying like, find me the cheapest pair of, I don't know, sneakers that fit the following standard. And so you could imagine a world where lots and lots of consumers are being driven to the cheapest possible option for their needs. And you could also imagine a world where maybe shops start to respond with more sophisticated price pack architecture, maybe dynamic pricing or something like that, which makes me wonder if the future is just going to be bots negotiating with each other over price endlessly? I think it's a bit, I mean, we can all speculate. I think it's a bit too early to have realized data there because one, it's an early phenomenon. Obviously, secondly, the leading edge of AI apps is quite high income, which is not necessarily representative of what the durable equilibrium will be. The one that we think about a lot is microtransactions are probably going to be now possible And so you probably have this whole segment of commerce that previously wasn open Microtransactions they never work or never have worked historically because of the kind of mental load of deciding. There's no point at selling Bloomberg articles for, you know, 40 cents each, because by the time you've decided, you might as well just have bought a day pass or a subscription or something like that. However, if Bloomberg, and actually I would love if Bloomberg had a proper API for their data, because if you want to do anything involving finance, when you're Vibe coding, it'd be really handy to have the Bloomberg data and the terminal is not exactly the right interface for AI. But selling Bloomberg data by the SIP, so it's like, okay, we're doing a query about the Strait of Hormuz, let's pull in the Bloomberg data for that. And you pay whatever small amount for that one little data request. That's a microtransaction that could actually happen. So that's the big effect that we think the time is finally there, because the mental load has traditionally been too high for humans, but AIs love overthinking things. They love scrutinizing every little decision. And so we might as well do that. This is something I've been thinking about as well. If microtransactions become frictionless, right? Do people try to monetize more and more content such that basically like you're kind of paying a tax to browse the internet By the way John I just want to let you know I don know if you know this but bloomberg has actually delegated some of these big strategic questions to me and tracy so we could they actually said no the podcasters should get to decide so that we could we could actually talk about this if you guys could put word in with mike and the crew that just i would like my my coding agents to be able to talk to bloomberg yeah i bet you'd like that we're messaging him right now but no this actually once again perfectly segs to my next question one of the things for i I've heard for years from my software developer friends. They use this binary free as in beer or free as in speech, right? So there's like software that you can download for free and that's free as in beer. But then free can also mean sort of like you can do anything you want with it. Free as in speech. I'm wondering on this point whether we need a third sort of one, free as in roaming, free roaming. Because every piece of data is theoretically valuable to a model maker. And every interaction with a piece of data, which maybe is called metadata, that interaction activity is also very valuable, etc. And I'm curious, given that every interaction and any version of that is valuable, are we going to be at the end of this sort of free-roaming internet? Or will we always just be paying these billionth of a cent microtransactions, which theoretically – [30:00] become doable and reasonable as we just sort of go along the internet, surf the web, because this is all very valuable information that's theoretically being exposed. I don't think AI is causing the end of the free roaming internet because there's always been a lot of proprietary stuff, newspaper paywalls. I don't remember a free roaming Bloomberg that I could read without paying even pre the AI era. Get the subscription, man. you can import get the terminal okay sorry yes i feel like i know you really want our terminal anyway sorry keep going yes so there's always been a lot of proprietary stuff on the kind of consumer side where you know this content is only available behind a paywall and then again to the data point this bloomberg and fact set and fmp and you know all these uh different providers and so i think what all the providers are thinking about is how do i make this stuff available to the without just necessarily making it available for free. And so what that looks like with the data owners is content licensing deals, right? Reddit now makes a lot of money from selling the Reddit data because you have all this like super interesting human discussion. You have the various media publishers either kind of suing or doing partnerships with or both the AI companies We are very interested in data providers generally like we did a demo of sessions as to how, you know, you could do a financial research project with your coding agent buying the information it needs to put together the PDF answer. But we're quite early in that journey. But no, I think there's been lots of proprietary data for a long time. And so it's more about how do we open that up to AI agents? I'm not sure you'd see that much of a constriction on what's available on the free internet. But just to push back on that a little bit, there's also been a lot of free stuff on the internet. And one of the reasons it's been free is because of traffic and advertising. And so if that's not going to get as much advertising or traffic anymore, because everything is going through an LLM platform and people just aren't linking back directly, the temptation would be to try to make up that revenue by charging for the actual content. Yeah, but that assumes that what the AIs want to browse and what the humans want to browse will be the same thing. And I'm not sure that'll necessarily be the case where it'll still be the case, let's say, for media properties, that is aimed at human consumption. And so they do fine with the advertising revenue they have Maybe they have some incremental revenue that they can get from the AI stuff And then I mean maybe you right that the really crummy aggregator pages where it like top 10 lists you know you Google something and you get some SEO page And there's like a million tabula ads on it that you're trying to click through and the popover takes over your screen. Maybe those sites have a harder time, but I don't think anyone's going to really be too sad about that. But you've talked about this yourself, I think, in a recent speech you gave. It's actually depressing to me because now, thanks to the Vibe coding apps, there's a bunch of stuff I would have liked to build, say, around Twitter data, right? And there was an actual era in which a lot of these social platforms are very open with their APIs, et cetera. And now everything's sort of shut down, so you can't even really, you know, unless I want to spend thousands of dollars, you couldn't really build like an app that does something with someone's Twitter account, et cetera. And I think you talked about this in your recent speech. Doesn't this further accelerate the move of everyone to just sort of clamp down and allow people to roam their site and slurp up data? Yes. But just to clarify, the human roaming has, you know, you can still browse Reddit for free. You can still browse Twitter for free What they clamping down on in some cases is ai training scraping where they want to put a toll on that and maybe they didn kind of intend for all that information to be kind of freely available to the ai apps and there you can argue whether that should be freely available to open ai and google and everyone like that or they should have to pay for it but it's not the case that are very important intellectual commonses of reddit and x and all the other asylums that we like to frequent are going away for us as you move. Whether you're planning a big tech event, launching a new campaign, or just stocking up on team gear, finding the right promotional products makes all the difference. 4imprint offers thousands of options from on-trend apparel and premium drinkware to tech, totes, and giveaways so you can find the right fit for any audience, purpose, or budget. You can customize it all your logo, your message, your look and many items come with no setup charge to help you save. And if you're really watching the bottom line you'll find standout choices at every price point so you can make a real impact while staying on budget. [35:00] Plus, you'll get expert help, fast turnaround times, and their 360-degree guarantee. So you can be 4imprint certain your order will arrive on time and look exactly right. Whatever your goal, 4imprint makes it easy to find your perfect promo match. Explore the possibilities today at 4imprint.com. 4imprint. 4certain. OnDeck is built to back small businesses like yours. Whether you're buying equipment, expanding your team, or bridging cash flow gaps, on decks loans up to $400,000 help make it happen fast rated a plus by the better business bureau and earning thousands of five-star trust pilot reviews on deck delivers funding you can count on apply in minutes at on deck.com depending on certain loan attributes your business loan may be issued by on deck or celtic bank on deck does not lend in north dakota all loans and amounts subject to lender approval get essential news on the people and companies pushing the tech sector to new frontiers. Hi, I'm Ed Ludlow. Join me for Bloomberg Tech, a daily podcast focused exclusively on technology, innovation, and the future of business. Every weekday, we bring you the latest insights on Silicon Valley's top companies and conversations with tech's biggest decision makers Listen to Bloomberg Tech on your commute home and stay ahead of the news cycle Subscribe today on Apple Spotify or anywhere you listen You know you talk about microtransactions and people have been talking about microtransactions for literally decades, probably. And the example you gave, oh, why can't we just buy an article for 50 cents? I think there's various reasons why that never took off. But one aspect is just the cost of transaction and with in theoretically particularly with like stable coins you could have almost free transactions are you seeing actual use of stable coins in sort of b2b transactions because the sheer cost is so low like where do stable coins come into this yeah you're right that when you have this pay by the sip model you obviously just can't economically speaking have kind of a an individual card transaction per usage. And by the way, this is the economic model of the AI economy because you have marginal costs for each token. And so no one really offers an all you can eat plan. You can subscribe to ChatGPT or Claude or something like that. And you'll have a plan and you'll eventually hit usage limits. Or you can pay on an API basis and you top up with some amount of credits and then burn down the tokens But everyone charges per usage at some level And so then what do you do And the kind of common mode, which we power a lot of is kind of this usage-based billing. We actually bought a company called Metronome to power that, where you track credits, right? And so that kind of telco billing, you have your card on file and then you track credits. We are quite excited about what you're referencing which is stable coins for this you're asking kind of how much are we seeing of it right now not that much yet but we're building out all the infrastructure for it the place where i think stable coins could be really exciting is you don't have to then go have a relationship with all these services and so if you want to vibe code a little website and you're like okay i'm going to buy the domain from cloudflare and host on versell and all these things you know you've had the experience of you have to go and sign up for all these services and do these things. And we think that experience of signing up for all these services should go away. And so we are making it where you don't need to do that. And paying via stable coins could be one way where you pay per web request or per individual use, very small amounts, and you don't have to go sign up for a whole bunch of different services. So I realized that Ajected Commerce is a spectrum of services everything from just simply filling out a form to giving an agent a certain amount of wallet spend and saying go off and figure out my life for me and spend the following on whatever Everyone also likes to talk about bots going wild, right, and going rogue. And so I imagine if you are giving an agent a certain amount of money and telling it to spend in a certain way, instruction specificity can vary. But what happens if things go bad or what happens if you're simply dissatisfied with the choice that the agent makes? Whose fault is that? Like, where does the liability actually lie? I think this concern is quite overstated because, one, the agents can do a huge amount of useful work for you while there is still a human in the loop. And so imagine if Cloud Code asks you, are you OK if I spend this amount and you just have to hit enter? that's much better than the experience of signing up for a service and clicking around the web face and things like that so you've still saved a huge amount of time even if you're kind of reviewing the transaction similarly again the consumer use case the way we suggest people visualize agentic commerce is you still have a human in the loop deciding making the final decision of both what thing i buy and am i okay spending this much money but the ais can still [40:00] be hugely beneficial as part of your shopping journey, even if you're as a human signing off on that transaction. And then you get a bit into delegated authority. It's kind of like, you know, we run a, you know, at Stripe, a large company and we spend billions of dollars a year and there's lots of delegated authority. You might say, what if you have one crazy person who makes a bad spending decision? And we occasionally have that. And, you know, how do you solve it? It's like, I don't know, give people spending limits on how much they can spend and you, you know, whack them on the knuckles if they do something really crazy. And you just solve it with kind of sensible limits and delegation of authority and approval process. It's kind of the same with an agent. Well, this gets me to a question I wanted to ask about running a big company in the age of AI. Things are changing insanely fastly. Fast. Fastly? Fast. Things are changing extremely fast. And there are certain things that people were excited about last year that didn't fizzle out in the same way or certain things that people got really excited about even in December 2025 that became the next hot thing. For Stripe specifically, or for a company, how do you think about saying, okay, here is a thing that we want to exploit. Here is a moment and we want to take advantage of it and we going to build the thing for this moment versus we want to allocate resources for anticipating what going to be the thing 24 months from now explore where this is going How do you think about the sort of resource allocation between exploring and exploiting in such a fast moving environment from the corporate standpoint? I would say you need to be very flexible because as you were saying, if your information is three months out of date or six months out of date, it's just woefully old. And so we tried to be pretty flexible. And I know there's AI somehow brings out a desire for thought leadership in people. Yes. Oh my God, everyone. Very brand pronouncements. Yeah. And just, I don't know, like we have more questions than answers and we're trying to figure it out as we go through. So that's one is kind of having the right level of humility. The second thing is you kind of need to have the similarly the right level of AI psychosis. The dose makes the poison where you need to really believe in the capabilities of the models. but you also have to be willing to reason about the limits. And so to give you an example, really amazing at writing codes and huge productivity uplift there. If you try to do anything, I don't know if you guys have tried to do anything like a financial analysis with the models, they're just horrible at numbers, innumerous. And they kind of fool you because they can put together something that they can build a really nice spreadsheet but you like but the numbers just go completely wrong halfway through the spreadsheet Oh great point You absolutely right And so you need to have kind of a bit of a fingertip sense for what the models are good at versus bad at. I kind of think of this in you split up the company into the traditional, you know, this R&D, go to market and G&A. And I think about the different domains. They work quite differently. within engineering engineers have always been passionate about tools and better tools and any kind of a half self-respecting engineer is very eagerly adopting ai you don't get too many laggards and so we have seen a huge amount of adoption within software engineering at stripe that's going really quickly i don't think we're at the end state but we're pretty like that machine i would say is pretty well set up to be in this self-improving loop because the people who are kind of doing the engineering day in day out also think about how can i better automate this process how can i use ai to to make it work faster and so a lot of bugs that are fixed at stripe no one ever opens an editor for them they just hit the fix this bug button in the interface and that's done agentically so that that's software engineering on sales and kind of go to market what helps you is that it is the most measurable domain inside of tech companies and companies broadly and so So you can reason about the specific productivity of a specific salesperson in a way that you can with a software engineer or a lawyer or something where any kind of metric there is very subject to Goodhart's law. Whereas with go-to-market, it's measurable. And so we have seen at Stripe, at least, our sellers are super into AI. They're adopting it very eagerly. And you can't ascribe causality in the productivity numbers, but the productivity numbers are very good. The place where I think it's trickiest is within the other general functions. You think about legal and finance and risk and things like that. And there, there is a lot of AI happening, but it's just tricky because one, there's a huge amount of infrastructure work that needs to happen. It's like, okay, we actually have to get this financial data into a form where it's usable by the AIs. And we used to keep it under lock and key, and now we need to kind of load it in a way that the AI can use. So we had a big infrastructure build at Stripe to make the data usable to the AIs, but also kind of keep it locked down appropriately and things like that. And then, like I said, the AIs are just uneven in which domains they are good at. And so if it is a coding problem or if it could be turned into a coding problem, very good at that. The public tax code and the public legal code, they're very good at that because it's techs that they've been trained on. But working with, I would say, [45:00] specific financial data at Stripe or kind of specific internal stuff that we have, that's just kind of harder to get good results out. It's not impossible and the models will improve a lot, but I would say that is kind of the trickiest domain to get great results. So just going back to the beginning of this conversation, we talked about how the internet basically runs on commerce. And if the nature of that commerce is changing, you would expect some impact on the internet. We've been asking you various questions about that, but I'm just going to ask you very directly, What is your sort of broad vision of the Internet in the world of agentic commerce? We're very excited for the intelligence explosion. And so in our pocket of the Internet, we are seeing much more entrepreneurship happening. That shows up in the numbers. So to give you a sense, new business creation on Stripe in Q1 was up 71 percent year over year. that business generally is not growing that quickly. It's like the leading edge numbers are seeing much more business creation. You see this in other numbers too. So the app store, like number of app launches was quite flat for many, many years. And now it's seeing kind of similar growth to that And so there are going to be many many fruits of the intelligence explosion But we get quite excited about the idea that you have more people creating companies and maybe you get more dynamism, right? We talked about some of the effects where, you know, you'll maybe see more price competition. If you work at a big company and you think, hey, the way we are doing this is really silly. I could do it so much better. The bar to you being able to quit the company and develop a competing better offering is now lower because you have more power available to you. You can recruit the intelligence that's available to you. And so our prediction is that you get more economic dynamism, which we're already seeing in our leading edge numbers. You know, we're seeing more startup formation. But we talked about this at Stripe Sessions, you know, kind of in a Kosian sense, maybe you see smaller firms and more of them because it's easier for them to coordinate with each other. John, I have like a million more questions for you, but I'm aware of the time. So let's just leave it there. This was such a treat. Really enjoyed getting the chance to talk and would love to have you back sometime. Absolutely. We'd love it. Tracy I really thought that conversation was super interesting I mean just to start there is a lot here This is a big deal. I think where it's going to go, the question of whether just commerce directly and then also your question of the Internet more broadly, there are some pretty big questions and pretty big implications from which way some of this stuff unfolds. Yeah, I'm glad that John appreciates the enjoyability. of scrolling. Yeah, well, yes. I do have questions about, I guess, discoverability when it comes to agents. So everyone by now knows the example of you go on Netflix, the first video or movie TV series you ever watch is incredibly important because the algorithm sort of builds itself around that one choice. And if you watch something stupid, then forever and ever it's going to pinpoint you as this one demographic. So there is a tendency for algos to sort of cluster among like recognizable patterns. So it becomes harder to discover new things, new products. I will say some algos obviously do it better than others. YouTube's algorithm I find like actually really phenomenal for resurfacing or resurfacing new things Well within LLMs itself I find memory to be a very double sword which is on some level I do think it more useful when it knows a little bit about the context of the user But there'll be times when it'll shoehorn any, like you did a query six months ago and it'll try to bring it back around to that. I think there's a lot there. I think there's a lot there with the future of advertising. Just setting aside, what is the equivalent of SEO in the LLM? we should probably do an episode from the perspective of retailers that are really having to think about how they surface within. Totally. So that in itself is a huge shift if you're a retailer and now you're trying to target bots as well as humans. And then the overall question of what happens to the Internet in this new world, like I tend to be very cynical about the overall direction of the Internet and just assume that every new advance in technology is going to make it worse, but you could see a future where the ability to do seamless microtransactions basically means that everyone starts charging for their content, for their APIs, and so you actually have to pay a fee to be on the internet. This is what I mean. The end of free as enrolling, I think, is an interesting question. I also just personally, I find it very validating that I'm not crazy that actually being able to do something at a... [50:00] a domain name is something that other people thought needed to be solved. You know what I'm saying? So some of these things seem very specific that Stripe announced in the last couple of weeks related to both the domain name question, related to being able to pay an API directly rather than go to a website and enter your credit card and stuff like that for credits or whatever. I found these things to be very annoying and big deals. And I'm sort of glad that I feel that's crazy. Like, actually, these are annoying things within the new contemporary landscape. Absolutely. And speaking of annoying things, I also feel like a big chunk of the future is going to be bots just, like, negotiating with each other. And I've already heard in the insurance sphere, you know, you've heard of insurers using AI to, like, bulk deny claims, right? And then I've heard of new AI startups or existing companies using AI to contest the insurers pushback. And so it just feels like it's just going to be bots. The only money makers are going to be chip companies in Korea because everyone's going to be having to get more powerful chips and old boomer landlords in San Francisco because the employees of OpenAI and Anthropoc need a place to live. And health care. Yeah. All right. Shall we leave it there? Let's leave it there. This has been another episode of the All Thoughts Podcast. I'm Tracy Alloway. You can follow me at Tracy Alloway. And I'm Joe Weisenthal. You can follow me at The Stalwart. Follow our guest, John Collison. He's at Collision. Follow our producers, Carmen Rodriguez at Carmen Armin. Dashiell Bennett at Dashbot. Kale Brooks at Kale Brooks. And Kevin Lozano at Kevin Lloyd Lozano. And for more OddLots content, go to Bloomberg.com slash OddLots for the daily newsletter and all of our episodes. And you can chat about all these topics 24 in our Discord discord slash OddLots And if you enjoy OddLots if you like it when we talk about the future of the internet then please leave us a positive review on your favorite podcast platform And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad-free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening. Thank you. promo products to match. With thousands of options, from apparel and drinkware to tech and totes, it's easy to find the right fit for your brand and budget, with standout choices at every price point. And with their 360-degree guarantee, you can be 4imprint certain your order will show up just right, right on time. Explore more at 4imprint.com. 4imprint. 4certain. Hi, I'm David Weston. Join me every week for the Wall Street Week podcast to hear stories of capitalism from around the world, from geopolitical tensions and central bank decisions to artificial intelligence, energy, and infrastructure. We sit down with the CEOs, economists, policymakers, and thought leaders whose decisions are shaping markets everywhere we find them. Subscribe to the Wall Street Week podcast on Apple, Spotify, or anywhere you listen.