Miro Executives Warn AI Speed Erases Organizational Intelligence

Miro高管警告：AI速度正在抹杀组织智慧 不受约束的AI速度剥离了达成共识所需的有效人际摩擦，迫使团队陷入更快速的遗忘。 该演讲概述了五项可落地的策略，以确保快速执行能够转化为真正的组织学习，而非流于表面的速度。演讲强调，必须保留与客户的直接接触以建立人类直觉，迫使AI模型在共享画布上将其推理过程可视化，并转变公司文化，奖励有见地的提问而非即时的答案。此外，演讲主张进行规范的回顾反思以防止“更快遗忘”，并总结指出，速度最终应当用于换取更深层次的理解。 保护与客户的直接接触对维系人类真实直觉至关重要，过度依赖 Gong 录音或整理摘要会脱离实时交流中的情绪脉搏与关键上下文。 (16:14) 高级 AI 模型往往隐藏其内部推理逻辑，团队应借助 Miro 画布将中间思考步骤可视化，像审查数据模型公式一样对推导过程进行审视与协作核查。 (17:22) 随着生成式 AI 使现成答案商品化，能够彻底改变问题视角的深刻提问变得更具价值，组织文化应向鼓励和奖励高质量提问转变。 (18:21) 高频交付若缺少复盘机制只会加速组织的遗忘而非积累认知，保持高质量复盘能倒逼团队在前期确立清晰假设并规范衡量标准。 (18:57)

Brendan McCook argues that organizations must ensure technological speed serves to cultivate deep understanding rather than merely maximizing output volume, emphasizing practices that protect productive friction and collective intelligence. Preserving Direct Human Contact and Intuition (16:14) While automated digests and transcripts streamline data access, McCook emphasizes that direct human contact remains essential for developing authentic product intuition. Experiencing live interactions—such as observing a user visibly hesitate while navigating an interface or reading an unedited complaint—provides vital emotional context that synthesized summaries eliminate. McCook notes that revenue intelligence tools like Gong have generated an unintended second-order effect: team members frequently bypass live customer meetings in favor of reviewing recorded sessions later. However, engaging with passive recordings fails to reproduce the real-time dynamics and conversational pulse experienced by actively participating in live discussions. Visualizing Intermediate AI Reasoning (17:22) Advanced models like Claude execute sophisticated, multi-step tasks but often conceal their internal logic, leaving teams with opaque final outputs. Drawing an analogy to management consulting, McCook asserts that unexpected outputs from complex data models should never be addressed with an arbitrary correction factor; teams must inspect the raw inputs, examine the formulas, and trace the workflow. To enforce this rigor with artificial intelligence, Miro teams export intermediate reasoning steps onto visual canvases. Displaying every step allows collaborators to collectively audit the underlying logic, identify potential errors, and engage with the process rather than passively accepting isolated results. Valuing Transformative Questions Over Commodity Answers (18:21) The widespread availability of generative AI tools like ChatGPT has turned answers into a free, instant commodity. McCook claims that intellectual value in an organization is no longer defined by offering quick solutions on demand, but by formulating insightful questions that shift an entire group's perspective. Because the right inquiry can fundamentally reshape problem definitions and project trajectories, McCook contends that workplace cultures must actively recognize, encourage, and reward high-quality questioning rather than defaulting to the celebration of rapid answers. Structuring Disciplined Retrospectives to Prevent Forgetting (18:57) Retrospectives represent a fundamental metric of collective intelligence, yet they are often neglected because they take place after a project has already shipped. McCook warns that prioritizing high shipping velocity without setting aside time for reflection leads to accelerated forgetting rather than genuine organizational learning. Adopting structured retrospectives, a discipline rooted in naval traditions, enhances the entire development cycle. The expectation of a rigorous post-mortem compels teams to articulate clear hypotheses at project onset, structure workflows deliberately, and measure performance metrics accurately. McCook concludes that the ultimate goal of these organizational habits is to convert raw speed into deeper comprehension rather than higher volume alone, positioning Miro as a collaboration layer that preserves productive friction as industry trends actively try to engineer it out.

Brendan McCook在Ai4分享中提出，面对人工智能与自动化工具的普及，团队应致力于用效率换取更深层的组织理解，而非仅追求交付数量的增长。他认为企业必须主动维系与客户的直接感性连接、拆解AI黑箱逻辑、奖励高质量提问，并依托严谨的复盘机制来沉淀集体智能。 保留一线接触以维系真实直觉 (16:14) 自动化分析与转录工具虽然免去了跨部门调取客户记录和提炼洞察的繁琐步骤，但过度依赖二手摘要容易切断团队对真实情绪的感知。以Gong等营收智能工具为例，其产生的次生效应是让团队成员倾向于被动查看录音，而非亲自加入客户的实时沟通。然而，静态录音或综合简报无法替代身处现场所感受到的对话脉搏。 McCook强调，直接的人际接触才是产品直觉的来源。亲历用户交流、观察使用者在面对界面时的犹豫停顿，或是阅读未经加工的原生投诉，能提供结构化摘要往往会过滤掉的复杂情境。团队需要刻意保护这些未经中介的沟通时刻，以维持对真实需求的判断力。 可视化呈现AI中间推理过程 (17:22) 以Claude为代表的高级AI模型擅长处理复杂的多步骤任务，但其往往隐藏了内部的推理步骤，仅向团队输出最终结果，导致执行过程陷入黑箱。McCook将此与管理咨询建模类比：当数据模型推导出的结论出现异常时，分析人员不应草率套用经验性的修正系数掩盖问题，而必须逐层排查原始输入、公式推导与完整计算流向。 组织对AI的应用也应遵从同等标准，要求其展现思考路径。在Miro的内部实践中，团队将AI的多步推理逻辑直接导出至可视化白板上，方便全体成员共同审查底层假设、推敲环节细节并修正错误，避免在孤立状态下盲目采信不透明的自动化产出。 鼓励优质提问而非廉价答案 (18:21) 随着ChatGPT等生成式AI工具的普及，基础事实与常规解答已变成即时且近乎零成本的通用商品。在团队协作中，“全场最聪明的人”不再体现为能够脱口而出既定答案，其核心价值已转移到能否在关键讨论节点提出具有启发性、能彻底重塑问题视角的深刻提问。 McCook指出，一个切中要害的高难度提问足以扭转整个项目的演进方向并抬升交付品质。因此，组织文化应当顺应这一认知模式的转变，主动鼓励、识别并重奖提出突破性问题的成员，而非单纯褒奖快速给出标准答案的惯性行为。 将工作流锚定于复盘机制 (18:57) 复盘是组织沉淀学习成果最有效的工具之一，但由于它通常安排在产品上线之后，在追求前行交付动量的驱动下最容易被团队轻率跳过。McCook警示，如果组织只顾快速发版而缺乏反思机制，高频交付所带来的往往不是真正的认知升级，而是“遗忘得更快”。 借鉴海军的制度化传统，严格的复盘流程能够倒逼前期工作流的规范化。为了在项目收尾时进行有价值的讨论，团队在最初立项阶段就必须明确提出更清晰的假设，以更审慎的方式拆解任务，并设立精准的数据指标，从而检验团队是否真正具备可持续的集体学习能力。 McCook总结指出，上述方法的核心诉求在于让技术带来的高速度服务于认知的深化，而不是陷入单纯的产出膨胀。在当前行业普遍试图通过技术消除一切工作摩擦的背景下，Miro定位于成为协助团队驾驭“良性摩擦”的协作层，借由反思与透明协作巩固组织的集体智能。

## Background 该内容由 Ai4 频道于 2026 年 8 月 18 日发布在 YouTube 平台，主讲人为 Miro 产品营销负责人 Brendan McCook（以及产品总监 Kendra Wilkins）。视频探讨了 Eugene Wei 提出的“隐形渐近线”在 AI 时代的体现——当个体的产出速度飙升 10 倍时，组织的集体智慧却难以跟上 AI 的脚步，进而阐述了为何团队在消除摩擦的同时丧失了深度认知，以及如何在不牺牲速度的前提下通过 AI 培育集体智慧。 ## TL;DR 演讲深入剖析了企业在 AI 时代遭遇的新型“隐形渐近线”：工具与上下文瓶颈打破后，过度消除协作摩擦导致集体智慧沉淀出现断层，6 步“意义建构”过程被压缩为 AI 黑盒提示词，引发思考浅层化危机。针对 MIT 调研中 95% 的 GenAI 试点缺乏实效的现状，演讲指出成功破局的关键在于引入可视化“协作层”并主动设计有益的思维摩擦。为此，团队应践行重视口述推导、死守现场真实接触、将 AI 推理视觉化、奖赏顶级提问以及围绕复盘设计工作流这五大原则，确保以速度换取深层理解而非盲目堆砌交付物。 ## Atmosphere & style 演讲开场以潜艇军官转型的风趣自嘲拉近距离，随后迅速切入极具纵深感的商业与认知理论，节奏从容而条理明晰。在整个科技行业狂热追求极致提速的背景下，主讲人以冷静清醒的批判性视角展开反思，通过白板激辩、严苛反馈等生动案例引发强烈共鸣。整体氛围兼具战略洞察的厚重感与实战操作的启发性，带给观众一种在智能化狂奔中重新找回思考深度与协作温度的笃定感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=4s) “隐形渐近线”（Invisible Asymptote）代表企业增长的隐形天花板，单靠重复既有策略与局部微调无法破局，需要进行重新定义品类级别的战略重构。 - [02:21](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=141s) AI 发展在跨越工具采纳与上下文连接后，正撞上“集体智慧渐近线”：个体效率虽提升 10 倍，组织整体却因认知脱节而难以沉淀协同价值。 - [05:08](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=308s) 团队在盲目追求速度时极力消除人际阻力，例如大量高管全自动审阅报告，彻底切断了战略经验与深度思考向团队传递的关键反馈通道。 - [08:29](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=509s) 试错成本的断崖式下跌导致团队失去了“高质量失败”带来的深刻复盘与认知重构，陷入极速上线众多功能却毫无认知积累的盲目试错陷阱。 - [09:43](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=583s) AI 黑盒压缩并跳过了六步“意义建构”（Sense-making）协作链条，使得交付物脱离了团队思考过程，导致决策依据退化为浅层摘要与空心化直觉。 - [11:48](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=708s) MIT 调研显示 95% 的企业 GenAI 试点无实质成效，根源在于工作流与协作方式脱节；而突围的 5% 成功者均在关键节点主动设计了有益的思维摩擦。 - [11:48](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=708s) 未来十年的数字化核心正从“彻底消除摩擦”转向“智慧地设计摩擦”，亟需引入连接智能体与底层记录系统的协作层，将思考推导过程显性化。 - [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) 日常协作应重视口述演示以替代冷冰冰的 AI 摘要，同时在自动化收集数据时死守真实的业务与用户现场触点，以保护商业直觉。 - [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) 团队应要求 AI 将多步推理过程以视觉化方式在画布上展开推敲，并在现成答案极度廉价的时代，重点奖赏能够颠覆问题视角的优质提问。 - [19:15](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=1155s) 高质量的复盘是沉淀集体智慧的试金石；缺乏反思的高频交付只会导致更快的遗忘，提速的终极目标是用速度换取认知深度而非单纯堆砌产出。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=4s) 潜艇军官转型的产品营销人，以及亚马逊的“隐形渐近线” 来自 Miro 的产品营销负责人 Brendan McCook 在 Ai4 大会上登台开场。他幽默地分享了一个关于自己的趣味背景：他的职业生涯最初其实是一名潜水艇军官。台下的听众或许会好奇“潜艇军官转行做产品营销”究竟是怎么回事，但他调侃道，潜水艇上其实根本没有产品营销这回事，这也让他格外珍惜能够代表 Miro 站在这里探讨 AI 时代协作的难得机会。 随后，Brendan 引出了本次演讲的核心主题——“隐形渐近线”（The Invisible Asymptote）。这一概念最早由 Eugene Wei 在 2018 年的一篇知名博文中提出。Eugene Wei 曾是亚马逊的战略财务分析师，其核心职责就是对业务进行前瞻预测，寻找增长的边界与破局点。“隐形渐近线”指的是企业在沿着现有路径发展时，终将撞上的一堵“隐形天花板”：即便团队还在拼命重复此前奏效的做法，增长却依然不可避免地放缓并逐渐走平。 在亚马逊的发展历程中，最典型的隐形渐近线案例就是提升消费者的钱包份额（Share of Wallet）。为了让用户把亚马逊当作日常购物的首选，团队付出了巨大努力： - 持续优化客户体验，扩充海量商品选择，并不断打磨极具竞争力的价格。 - 尽管如此，用户的消费频次依然停滞不前。 最终，亚马逊团队意识到真正的阻碍在于“运费”。即便运费再便宜，只要结账页面出现运费金额，用户就会产生哪怕一丝一毫的犹豫。正是这细微的犹豫构成了增长的天花板。打破这条隐形渐近线，光靠局部的微小优化是行不通的，必须依靠像推出 Amazon Prime 这样具有“重新定义品类”（Category-defining）级别的战略重构。 ```mermaid flowchart TD accTitle: 业务增长撞上隐形渐近线的演变路径 subgraph NormalGrowth["传统增长路径与瓶颈"] A["现有策略持续发力<br/>（优化体验/价格/选择）"] --> B["业务快速增长<br/>（Up to the Right）"] B --> C["撞上隐形渐近线<br/>（增长停滞/同质化打法失效）"] end subgraph Breakthrough["突破与跃迁"] C --> D["传统思维：原地打转<br/>（重复原有动作）"] C --> E["战略重构：突破天花板<br/>（例如推出 Amazon Prime）"] end ``` --- ### [02:21](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=141s) AI 时代的三大渐近线：从工具使用、上下文连接到集体智慧 在 AI 浪潮中，企业同样在撞上全新的隐形渐近线。许多组织大举投入 AI，试图借此提升交付速度和产出，但在经历初期成效后，往往迅速陷入边际效益递减（Diminishing Returns）的窘境。Brendan 将企业经历的 AI 渐近线划分为三个阶段： ```mermaid flowchart LR accTitle: AI 发展过程中跨越的三大渐近线 subgraph TechPhase["技术维度突破（已基本解决）"] A["1. 工具采纳渐近线<br/>（工具难用/门槛高）"] -->|工具易用化/自动化| B["2. 上下文渐近线<br/>（脱离核心数据与记录系统）"] B -->|MCP与系统打通| C["技术就绪<br/>（个体提速 10x）"] end subgraph BehaviorPhase["行为与组织维度（当前痛点）"] C --> D["3. 集体智慧渐近线<br/>（团队脱节/摩擦被消除）"] end ``` 1. **采纳渐近线（Adoption Asymptote）**：初期工具极其难用。Brendan 回忆自己在 Okta 负责战略项目时，曾和几位技术极客同事在本地终端部署 Ollama，费尽周折只为了从 Zoom 会议录音中提取行动项。那套终端工具体验极差且频繁崩溃，根本无法向全员推广。但随着 Zoom 等产品原生集成了该功能，这一技术易用性门槛已被彻底跨越。 2. **上下文渐近线（Context Asymptote）**：过去 AI 主要用于泛泛的文本摘要和内容生成，脱离了企业的核心数据系统（Systems of Record）。而在最近半年内，随着模型上下文协议（MCP）等工具生态的爆发，AI 已经能够深度连接业务系统，赋能高价值工作流。 3. **集体智慧渐近线（Collective Intelligence Asymptote）**：当技术和上下文瓶颈相继被打破，个体的工作速度暴增 10 倍时，组织却迎来了更严峻的“行为与协作挑战”。 许多团队看似每个人都在疯狂产出，但组织的整体节奏并没有变快，决策的信念感（Conviction）反而更低。其根本原因在于：**AI 极大地加速了个体获取知识与单打独斗的速度，却让“拉齐团队认知”变得更加困难。** 组织正在丧失沉淀“共享集体智慧”（Shared Collective Intelligence）的能力，而这恰恰是团队发挥 1+1 > 2 协同效应的基石。 --- ### [05:08](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=308s) 被抹除的“建设性摩擦”与失落的专家思考 构建集体智慧的过程本质上是极其痛苦且反人性的——它要求团队直面分歧、理解多维视角、改变固有想法，并在激烈的碰撞中达成“坦诚表达异议并全力执行”（Disagree and Commit）。因此，在追求速度的本能驱动下，人们会用 AI 极力消除这种“摩擦”（Friction）。 Brendan 引用了 Anthropic 经济指数（基于 Claude 连接器数据）的一项惊人发现： - 在使用 Claude 审阅下属提交报告的高管中，高达 **69%** 的人选择将该审阅过程**完全自动化**。 高管往往拥有最全面的业务上下文、最深厚的专业知识与战略经验，而他们向员工传授这些高价值认知的主要渠道正是“日常反馈与战略对齐”。当管理层选择将这一过程完全交给 AI 自动化时，关键的认知传递通道便被彻底切断。为了追求速度，团队抹平了所有的协作摩擦，但正是这些摩擦在过去沉淀了不可替代的深度价值。 ```mermaid flowchart TD accTitle: 消除摩擦对团队认知的负面循环 A["过度依赖 AI 自动化协作<br/>（如 69% 高管全自动审阅报告）"] --> B["消除人际沟通中的阻力与摩擦"] B --> C["阻断战略上下文与经验的传递"] C --> D["团队丧失深度思考与共同认知"] D --> E["沦为表面提速，丧失核心竞争力"] ``` 为了具象化这一点，Brendan 回顾了自己在 Reforge（一家打造增长与产品专业进阶课程的公司）早期的经历。当时他的工作是对顶尖行业专家进行数百小时的深度访谈，并将专家的经验提炼成方法论框架： - **过去的思考模式**：访谈后等待数天拿到文本，团队围在白板前把打印出来的访谈稿贴满墙壁，在画满框架草图与便签的白板前激烈争论、相互推翻、持续迭代。 - **痛苦但质变的反思**：将方案提交给 CEO 后，CEO 会留下一份写满数百条尖锐、严厉且极难落地的反馈意见。面对这些令人沮丧的批注，团队在抵触与自我防卫中艰难消化，但正是解决这些棘手批评的过程，使最终的课程产出实现了质的飞跃。 学员购买课程并不是为了那几张现成的框架图，而是为了买单支撑这些框架背后的“深度思考过程”。如果用今天的 AI 将访谈整理与框架总结完全自动化，表面上省去了所有痛苦与时间，但团队也随之失去了在应对不适感中打磨专业认知（Expertise）的唯一机会。 --- ### [08:29](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=509s) 试错成本降低的隐患：从“高质量失败”滑向“快速但盲目的试错” 在传统的产品研发模式中，团队在规划阶段留有充足的缓冲空间来进行深度协作：挑战彼此的前提假设、与客户做扎实的验证、反复推演所有数据。 在过去，产品发布是一件极其昂贵且高风险的事情： - 正因为发布成本高昂，团队才会倾尽全力注入最强大的信念感，确保问题假设和解决方案假设都达到极致精准。 - 一旦产品失败，不仅对业务打击巨大，也会给团队带来真实的阵痛。但正是这种痛苦逼迫团队进行深刻的复盘与认知重构，带来如同“创伤后成长”（Post-traumatic Growth）般的经验跃升。 然而在当下的 AI 时代，团队的交付与上线速度变得极其敏捷，但失败的代价也变得微不足道。由于不再感受到失败的痛感，团队也省去了高质量的反思与复盘： - **过去的模式**：谨慎、深入地经历一次漫长而沉痛的失败，但能为未来的成功建立起坚实的认知基石。 - **当下的危机**：在完全没有反思和认知积累的情况下，极速地上线并搞砸 100 个功能，却始终无法为组织构建面向未来的核心成功基础。 ### [09:43](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=583s) 意义建构的六个阶段与 AI 时代的思考空心化 在探讨组织认知时，演讲者引入了被誉为“实地调研界的詹姆斯·邦德（James Bond）”的学者简·奇普切斯（Jan Chipchase）关于“意义建构”（sense-making）的经典框架。演讲者特别表达了对“意义建构”一词的偏爱：在这个过程中，团队并不是在制造具体的代码产品或交付物，而是在“梳理出意义”，这本身就具有极高的价值。传统的意义建构遵循一个严谨的六步递进链条：从明确假设（identify a hypothesis）、收集数据（collect data）、结构化信息（structure information）、聚类知识（cluster knowledge）、提炼洞察（synthesize insights），最终评估并升华为智慧（evaluate wisdom）。在过去，这六个环节需要团队成员共同投入大量时间，通力协作完成。 然而，随着工作模式向 AI 时代演变，这种认知链条发生了隐秘而危险的断裂。现在的常态往往演变成：单个员工独自向大模型输入提示词（prompt），而上述原本需要深思熟虑的六个阶段全部被压缩并隐匿在 AI 的黑盒逻辑之中。表面上看，这带来了极大的效率飞跃——个人可以瞬间跃升到最终产出，生成一份精美的交付物（artifact）并直接丢给团队。然而，问题恰恰在于：“哪怕团队成员会仔细阅读它，但他们从未参与过这一意义建构的过程；事实上，连生成它的你，也没有真正参与过意义建构。” 这种跳过思考磨砺的捷径导致了深层恶果：团队开始依赖越来越缺乏严密推导的直觉做出决策，而一旦决策做出，又极难在组织内部达成真正的对齐（alignment）。如果寄希望于“只要给 AI 喂入更多上下文、让 AI 代替人类做更多决策，问题就会迎刃而解”，这恰恰是掉入了“隐形渐近线（invisible asymptote）”的陷阱——虽然看似在全力推进技术，但组织的实际产出能力却在无形中触到了天花板。哪怕是在最成熟的企业中，员工每天依然要做出数十甚至上百个细微判断：今天优先做什么、读什么、下一步构建什么、找谁沟通、在何处投入时间。当支撑这些微观决策的依据退化为“执行摘要的执行摘要（executive summaries of executive summaries）”时，组织整体的认知便滑向了无法挽回的浅层化。 ```mermaid flowchart TD accTitle: Traditional Sense-Making vs AI Prompting subgraph Traditional ["传统团队协作的意义建构（高摩擦·高认知）"] A["1. 确定假设<br/>Hypothesis"] --> B["2. 收集数据<br/>Data"] B --> C["3. 结构化信息<br/>Information"] C --> D["4. 聚类知识<br/>Knowledge"] D --> E["5. 提炼洞察<br/>Insights"] E --> F["6. 升华智慧<br/>Wisdom"] end subgraph AIPrompting ["AI 介入后的断裂模式（零摩擦·浅认知）"] P["个人提示词输入<br/>Individual Prompt"] --> G["AI 黑盒内闭环完成 6 步<br/>Hidden AI Processing"] G --> H["快速生成最终成果<br/>Instant Artifact"] H -. 缺乏对齐.-> I["浅层决策与执行断层<br/>Superficial Knowledge"] end ``` ### [11:48](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=708s) 破除隐形渐近线：协作层的缺失与阻力的智慧 针对这一困境，演讲者剖析了当前企业技术架构的核心短板。近几个月来，技术生态取得了长足进步，特别是将 AI 智能体（AI Agents）与记录系统（Systems of Record）深度打通。这两个层级的协同极大拉升了单点工作流的速度，但也加剧了孤岛现象：智能体在各自的烟囱里高速运行，系统只记录最终数据结果而不记录背后的推导逻辑，导致“执行往往发生在真正对齐之前（execution happens before alignment）”。 这两者之间真正缺失的，是一个承上启下的**协作层（collaboration layer）**。协作层的使命并非替代计算，而是为团队提供一个直观的交互场域：在这里，团队可以看清智能体的推理逻辑、沉淀共享的工作范式、设计核心业务流，从而避免个人在各个方向狂奔时累积起巨大的“集体智慧债务”。演讲者援引了其母校麻省理工学院（MIT）的一项重磅调研：高达 95% 的企业生成式 AI 试点项目实际上未带来任何可量化的业务成效。其根本原因绝非大模型能力不足，而在于“学习断层（learning gap）”——企业的工作流与协作方式根本没有跟上技术的迭代。 最耐人寻味的是，那成功突围的 5% 标杆项目，非但没有极力消除协作阻力，反而在工作流中**主动设计了有益的摩擦力（designing friction in the process）**。正如有远见的数字化研究所指出的那样：“如果说过去十年的数字化转型是为了消除摩擦，那么接下来的十年将是关于如何智慧地设计摩擦。”那些懂得精准校准阻力、将其转化为学习动力而非阻碍行动的企业，最终赢得的不仅是投资回报率（ROI），更是组织的韧性、深层意义以及持久的竞争优势。并非所有摩擦都是好摩擦——像等待会议记录这种琐事当然需要自动化；但涉及到核心认知与集体智慧的沉淀时，协作层正是守护和校准这种“关键摩擦”的核心阵地。这正是 Miro 近 15 年来所践行的“智能协作（Intelligent Collaboration）”：将智能画布作为连接智能体与底层记录系统的协作中枢，让团队的思考过程可视化、连贯化，重构集体的思考深度。 - **95% vs 5% 的反差**：MIT 调研显示 95% 的企业 GenAI 试点未能产生实际成果，失败根源在于组织协作方式的滞后；而仅有的 5% 成功者均在关键节点逆向保留了必要的“思维摩擦”。 - **三大核心缺陷**：缺少协作层导致智能体割裂工作、底层系统只存结果不存逻辑、执行跑在共识达成之前。 - **差异化优势的三要素**：真正的市场领跑者超越了单纯的 ROI 考核，更加注重韧性（resilience）、意义（meaning）与持久优势（lasting advantage）。 ### [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) 在日常团队中守护良性摩擦的实践建议 针对如何在日常工作中抵御浅层化侵蚀、重新找回关键思考阻力，演讲者提炼了五条立即可落地的战术准则： **第一，重视口述演示（Value the voiceover）。** 当前办公文化流行把 AI 总结的文档发来发去，但这往往是在“邀请别人快速扫视”，而不是深度思考。当员工主动带团队梳理其推导链条时，才是在“邀请团队深度参与”。过去很多团队对“会议极简化”矫枉过正（必须提前阅读预习材料、开会立即举手表决、会议火速结束），反而抹杀了认知交锋。团队应当预留时间让成员口头讲述推导过程中的卡点与主观抉择；即便是异步工作，利用类似 Miro 的 Talktracks 录制一段带思路讲解的视频，其信息传递效率与沉浸感也远胜于冷冰冰的 AI 摘要。 **第二，自动化数据收集，但绝不自动化真实接触（Automate the collection, not the contact）。** 自动化收集海量数据（如利用工具导入对话录音、抓取客服工单、生成洞见）确实令人兴奋，但团队必须死守与业务现场的“触点”。直觉往往诞生于未被过滤的真实场景之中——比如亲自旁听一场客户访谈、观察用户在使用产品时的微小迟疑，或是亲眼去读一条具体而刺痛的差评，而不是只看一份被美化过的汇总报告。演讲者举了销售情报软件 Gong 的例子：Gong 虽好，但也引发了一种不良次生效应——当你想跟进客户时，团队会说“我把录音丢给你看就行，不用占用日历去旁听了”。但事实证明，事后倍速看录像绝不可能替代“身处现场、捕捉对话暗流与微表情”所激发出的敏锐直觉。 **第三，要求 AI 将推导过程以视觉化方式呈现（Make AI share its thinking visually）。** 诸如 Claude 等前沿模型处理复杂多步任务的能力极其卓越，但它们往往倾向于直接抛出令人惊艳的最终结论，而掩盖整个中间推理过程。演讲者结合曾经作为管理咨询顾问的经历指出：如果搭建一个复杂的财务数据模型，当输出端数字反常时，分析师绝不可能直接乘以一个系数打九折糊弄过去，而是必须倒查底层假设、引用来源与公式逻辑。面对 AI 亦当如此。在 Miro，团队的做法是将 AI 的思考逻辑以流程图等形式展开铺设在画布上，供全员逐个节点审视、推敲与共同推演，使人机协作摆脱盲盒状态。 **第四，奖赏最顶级的提问，而非现成的答案（Reward the best questions, not the best answers）。** 在 ChatGPT 普及之前，一个人可以通过随时抛出标准答案成为“全场最聪明的人”；而在大模型普及的今天，答案已经彻底廉价化，每个人唾手可得。真正具有稀缺价值的，是在讨论陷入僵局时提出一个打破常规的优质问题——它能彻底重塑团队审视问题的角度，从而逆转整个项目的工作轨迹与交付水准。组织应当把资源和赞誉倾斜给那些提出关键问题的人。 **第五，围绕复盘设计工作节奏（Design your work around retros）。** 结合在海军服役时高频复盘的深切体会，演讲者强调复盘是组织学习最强大的利器，但它也是在项目交付后最容易被敷衍或跳过的环节。若想沉淀集体智慧，就必须把复盘牢固地固化为业务流程的必经之路。 - **重视口述与阐释**：发送摘要是在鼓励浅层扫读，口头引导推导过程才能激发全员深度共鸣；善用录屏讲述替代文本摘要。 - **保护现场触点**：不能让录音录像完全隔绝真实的客户接触，商业直觉源于与真实用户的未过滤互动。 - **AI 推理白盒化**：拒绝直接全盘接收大模型的最终产出，通过画布将多步推理可视化，以便团队进行审问（interrogate）与推敲。 - **提问胜过答案**：当现成解答触手可及，能够颠覆认知视角的优质提问成为组织最核心的智慧资产。 - **坚持复盘机制**：复盘是抵抗遗忘、打通组织闭环最核心的学习工具，绝不能在项目上线后被轻易裁撤。 ### [19:15](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=1155s) 复盘反思与沉淀集体智慧：用速度换取深度理解而非单纯产出 要检验一个团队或组织是否真正具备构建“集体智慧”（Collective Intelligence）的能力，最有效的衡量标准就是观察其复盘会议（Retrospectives / Retros）的频率是否稳定、质量是否过硬。复盘从机制上强制团队停下脚步，创造出关键的“学习时刻”。当前在敏捷与 AI 加速的环境下，许多人产生了一种认知偏差，认为如果团队交付和发布的速度极快，复盘的优先级就可以往后放，甚至认为忙于赶进度根本没有时间复盘。然而，如果在高吞吐量、快节奏交付的同时剥离了这种沉淀反思的时刻，团队实际上并没有积累任何认知与经验，带来的结果只是“更快的遗忘”（"there is no learning. There's just faster forgetting."）。 以结构化、严谨的方式践行复盘，能够从根本上重塑整个工作流的正向循环。当团队明确知道后续要进行高质量的复盘检验时，他们在项目启动阶段就会提出更加结构化、可验证的假设，在规划与执行工作流时会更加审慎，在定义和衡量成功指标时也会更为严密，从而确保团队在复盘对话中能够聚焦于真正有价值的核心问题。 回顾演讲中提到的五项核心原则，它们的初衷绝不是为了拖慢团队的交付节奏，而是要确保团队所提升的极限速度能够真正转化为深层的“理解与洞察”，而不仅仅是盲目堆砌“产出”（"make sure that the speed is buying you understanding, not just output"）。在 AI 时代，当整个行业都在试图通过自动化和工具链彻底“消除协作摩擦”（engineer out friction）时，Miro 致力于作为组织的核心协作层，帮助团队以富有建设性、高成效的方式驾驭这些必要的摩擦，从而在 AI 时代突破增长与认知的天花板。 - **复盘是检验集体智慧的试金石**：复盘质量与持续性直接决定了组织是将快速迭代转化为组织资产，还是沦为无意义的快速遗忘。 - **倒逼前期工作质量**：结构化的复盘机制能够反向推动假设构建的严密性、任务拆解的条理性以及成功指标衡量的精确性。 - **核心目标定位**：团队提速的终极意义在于换取认知层面的深度理解（Understanding），而非单纯追求交付工件的数量（Output）。 ## Notable quotes > [09:43](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=583s) 哪怕团队成员会仔细阅读它，但他们从未参与过这一意义建构的过程；事实上，连生成它的你，也没有真正参与过意义建构。 — Brendan McCook > [11:48](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=708s) execution happens before alignment — Brendan McCook > [11:48](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=708s) 如果说过去十年的数字化转型是为了消除摩擦，那么接下来的十年将是关于如何智慧地设计摩擦。 — Brendan McCook > [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) Value the voiceover — Brendan McCook > [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) Automate the collection, not the contact — Brendan McCook > [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) Make AI share its thinking visually — Brendan McCook > [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) Reward the best questions, not the best answers — Brendan McCook > [15:05](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=905s) Design your work around retros — Brendan McCook > [19:15](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=1155s) there is no learning. There's just faster forgetting. — Brendan McCook > [19:15](https://www.youtube.com/watch?v=TkgWq6PUXdI&t=1155s) make sure that the speed is buying you understanding, not just output — Brendan McCook ## Mentioned Kendra Wilkins, Brendan McCook, Miro, Eugene Wei, Amazon, Amazon Prime, Okta, Ollama, Zoom, Model Context Protocol (MCP), Anthropic, Claude, Reforge, Jan Chipchase, Massachusetts Institute of Technology (MIT), Talktracks, Gong, ChatGPT, Ai4 ## Video description Kendra Wilkins - Product Director, Miro Brendan McCook - Head of Product Marketing, Miro Every company eventually runs into an invisible asymptote, a term coined by Eugene Wei describing a hidden ceiling where growth quietly flattens no matter how hard you push. In the AI era, a new one is emerging: the point where an organization's collective intelligence can't keep up with the speed of its AI. Individuals are moving 10x faster, yet that speed rarely translates into 10x more value. In racing to move faster, teams are engineering out the very friction that once built deep, shared knowledge. This talk unpacks why that friction mattered, what the best teams are doing to preserve it without slowing down, and how to use AI to accelerate the development of collective intelligence rather than trying to replace it. [Source](https://www.youtube.com/watch?v=TkgWq6PUXdI)

[9:43] Brendan McCook: 哪怕团队成员会仔细阅读它，但他们从未参与过这一意义建构的过程；事实上，连生成它的你，也没有真正参与过意义建构。 [11:48] Brendan McCook: execution happens before alignment [11:48] Brendan McCook: 如果说过去十年的数字化转型是为了消除摩擦，那么接下来的十年将是关于如何智慧地设计摩擦。 [15:05] Brendan McCook: Value the voiceover [15:05] Brendan McCook: Automate the collection, not the contact [15:05] Brendan McCook: Make AI share its thinking visually [15:05] Brendan McCook: Reward the best questions, not the best answers [15:05] Brendan McCook: Design your work around retros [19:15] Brendan McCook: there is no learning. There's just faster forgetting. [19:15] Brendan McCook: make sure that the speed is buying you understanding, not just output

[16:14] Having data immediately available at your fingertips is fantastic because it eliminates the hassle of jumping through hoops just to gather transcripts, pull customer tickets, or parse customer insights. [16:14] However, it is vital to protect those direct moments of human contact because that is where true intuition lives. [16:14] Colleagues: We'll send you the recording, you don't actually need to sit in. [16:14] Engaging with a passive recording simply cannot replicate the live experience of "feeling the pulse of the conversation" [17:22] Advanced AI models like Claude excel at executing complex, multi-step tasks, but they frequently hide their internal reasoning process. [17:22] Organizations should hold AI models to that exact same standard by forcing them to make their thinking visible. [18:21] Since the arrival of tools like ChatGPT, answers have become a commodity—they are instant, free, and accessible to everyone. [18:21] Being the "smartest person in the room" no longer means having an immediate answer on demand. [18:21] Organizations need to actively lean into, encourage, recognize, and reward high-quality questioning rather than just celebrating quick answers. [18:57] A reliable test of whether a team is built for collective intelligence is the consistency and quality of its retros. [18:57] Shipping at a high volume without dedicated moments for reflection leads to "faster forgetting" rather than actual learning. [20:04] Brendan McCook: The ultimate goal of these practices is to ensure that speed serves to buy understanding rather than merely increasing output volume.

[0:04] Thank you all. Um it's awesome to be here. We're  going to talk about collaboration in the AI era [0:09] today. Um my name is Brian McCollough.  I lead uh product marketing at Miro. [0:14] And a fun fact about me, um I started my  career actually as a submarine officer. [0:19] And I know you're thinking like another submarine  officer turned product marketer, how many of them [0:23] are we going to hear? But actually uh very  little product marketing happening happens [0:27] on a submarine. Um but that makes me very grateful  to be here at AI 4 representing Miro in this role. [0:34] Um the title of this was the next invisible  asymptote. Who here has heard of the invisible [0:39] asymptote? It was a blog post written in about  2018 by Eugene Wei. Um this was something I [0:45] feel like we talked about a ton back when it was  written, and it really changed my perspective on [0:50] growth and strategy. And essentially, Eugene Wei  was a financial strategic financial analyst at [0:55] Amazon, and his goal was really to project  out what's happening with the business, [0:59] where is it going to go, what do we need to look  for? And he described this term of an invisible [1:03] asymptote as this ceiling that companies' growths  would hit um as it went down its current course. [1:09] Um people are trying the same things, and  eventually they stop working, and it flattens out. [1:13] At Amazon, the most obvious example of this um  was they were trying to get more share of wallet, [1:18] more people to turn to Amazon more regularly  for their purchases. Um so, they made a lot [1:24] of customer experience improvements, they had  better selection, they worked on their pricing, [1:28] but still there was this thing holding people  back. And ultimately, as many people were still [1:33] just trying the same things that got them to  where they were, they realized it was shipping. [1:38] That people hated paying for shipping. Even cheap  shipping, it didn't matter. Putting that cost of [1:42] shipping there made people hesitate, ever so  slightly. That hesitation caused a ceiling, [1:47] and they couldn't get through to being more of  a default store until they overcame that. So, [1:51] that's just not an incremental change to overcome.  That takes a category-defining change like [1:56] launching Amazon Prime to do. That's the pattern.  The strategy works for a while and then you're all [2:02] projecting everything's going up to the right,  everything's good. A year later, a quarter later, [2:06] a month later, things change. It stops working.  Met A lot of the companies just spinning their [2:12] wheels trying to do the same things over and over  again. The best leaders look for those invisible [2:16] asymptotes and think about how to overcome them.  So I've been thinking about this a lot in AI cuz I [2:21] see a relevant connection to the world of AI where  companies are trying to transform. They're trying [2:27] to move faster, deliver more output in the A of  AI, but they're running into these ceilings again. [2:33] Um there's a new category happening where the more  people invest in AI, often we're starting to see [2:39] these diminishing returns. People are really  advanced in their AI adoption and still it's [2:43] not working. And every company has an asymptote  at different places and I think the question is, [2:48] do you know where your company's is?  Do you know how to get around it? [2:53] So to bring this to life, let's start with two  common variations of this in the past and these [2:57] are ones I think we all went through but have  largely overcome. At first there was an adoption [3:01] asymptote. Tools just weren't usable enough. It  took a while to figure out how to use AI tools. [3:06] Companies couldn't use them more broadly. Um I  was leading before this role strategic programs [3:10] at Octa. I remember me and some very AI forward  colleagues built these on-prem instances of Ollama [3:17] to do the great grand task of taking a Zoom  call and extracting out action items from it. [3:22] But like we had to build this in a terminal. Uh it  sucked. It was bad. It was really clunky. It broke [3:27] all the time. There was just no way I could get a  large set of the organization to adopt this. But [3:33] luckily, the tools got better. Zoom does this by  itself. Everyone does this by itself. Those simple [3:37] use cases, the tools got better. We've largely  overcome this. Then I think we hit this context [3:43] asymptote. People were working with AI more, but  they were using it for summarization, for content [3:47] generation. It wasn't connected to the systems  of record, the data that's important to do much [3:52] more high-leverage work. And just in the last,  call it 6 months, there's just been an explosion [3:57] of tools connected through MCP that allows you to  do this. People are able to connect, they're able [4:02] to build this. So, organizations are in a place  where they have adoption, they have context, they [4:08] have high-value workflows. They are moving faster  than ever. I think we've all felt that. But, [4:13] as individuals especially are moving 10x faster,  that's where we run into this next asymptote. [4:20] Have you ever looked around and kind of  noticed like we're all doing a lot of stuff, [4:24] but I don't know that we are moving faster as  an organization. I don't know that we're better. [4:28] I don't know that we have more conviction  in what we're doing, maybe we have less. [4:31] This is what I've been feeling, and I think this  asymptote here is so different and so pronounced [4:36] because it's less of a technical challenge. The  first two got overcome with just improvements [4:40] in the technology. The tools got better, we  overcame them. But, I think we're running into [4:44] a behavioral challenge. Whereas AI accelerates  individuals' ability to take in knowledge, [4:50] move faster by themselves, starting to bring  your team along with you is even harder. [4:55] Organizations are struggling to develop this  shared collective intelligence. That's the thing [5:00] that drives alignment. That's the thing that makes  individuals work more than the sum of their parts. [5:08] So, I think the challenge especially is building  this collective intelligence is uncomfortable. [5:14] Like there is every incentive to engineer it out.  We don't like building this, it's uncomfortable. [5:21] Um it involves teasing out differences in  opinions, understanding how other people [5:24] can look at stuff as a problem, changing minds,  disagreeing, committing, all these things. No one [5:29] wants to do that, but there's an important part  there. Um I recently came across this uh Anthropic [5:34] has an economic index. It's a Claude connector. Um  and they they rated the most common use cases um [5:41] of uh for executives, the most common outsourced  tasks. And they found that for executives who use [5:47] Clod to review uh reports submitted by staff, 69%  of them are using to fully automate that process. [5:55] I just think about that for a second. These are  executives. They have the most context about the [5:59] business, the most expertise, the most experience.  How valuable is that information to their staff? [6:06] How valuable is that? And what's the  channel you normally give that through? [6:10] It's through feedback. It's through alignment.  It's through taking your staff's work and [6:14] understanding how does this or doesn't it fit  with the strategy? I'm curious how much less [6:19] effective that is when you are automating that  with AI. Sure, you can argue AI is getting better, [6:23] but I don't think it's getting that much better.  To go faster, we are engineering out all of the [6:29] friction from our work. But it turns out  the friction is doing something important. [6:35] So, this is how I was thinking about this. I used  to work at a company called Reforge. That was a [6:38] company that built professional development  courses for growth and product professionals. [6:42] In my early days, my job was to literally  interview experts in the industry and spend [6:48] hundreds of hours interviewing them and trying  to synthesize their experiences into frameworks [6:52] that were valuable for teams. And to do that,  we'd prep for these interviews, we'd run them, [6:57] I would take notes, I would send off a recording,  I would get a transcription a few days later. We'd [7:01] all look at this as a team, we try to brainstorm,  we pushed against each other, we'd iterate, [7:06] we'd go through stuff. And this photo captures  kind of one of those workshops for one of these [7:09] programs. You can see I have all these printed out  interviews, I have these sketches of frameworks, [7:14] I have all these notes on a board. This is  what thinking used to look like. And I feel [7:18] like even describing this now today feels like  describing a dial-up modem to my kids. Like, [7:24] that's not what work looks like today. So much  of what we can do would automate this. I can get [7:29] a transcription instantly. I'm really glad I can  do that. I don't want to wait three days for that. [7:33] But all of this other work is getting automated as  well. Furthermore, we would send this to our CEO, [7:39] and he was notorious. He She drop in  hundreds of comments, really thoughtful, [7:42] really challenging comments that were hard to  action. And I remember feeling this sense of like [7:48] despair, this sense of defensiveness, this sense  of discomfort trying to action this. I wanted to [7:54] show him, you don't understand all the work we did  to get here. But every time I went through this [7:58] process of actioning these difficult comments,  the work was better. Not just a little better, [8:03] categorically better. We were ultimately building  frameworks and courses, but people weren't paying [8:08] for frameworks. They were paying for the  thinking that went behind those frameworks. [8:11] And I think about if AI were to do this process  today, that job I used to have, I could choose to [8:16] automate essentially all of it. So, where's  the differentiation? Where's the thinking? [8:21] We were building expertise by working through that  discomfort. And when you avoid that, you lose it. [8:29] So, I think there's a parallel to product teams  today. Um product teams used to have this buffer [8:34] in planning where they would collaborate,  they would challenge each other's assumptions, [8:39] validate with customers. Um they would look  at all the data. Launching was expensive. [8:45] Launching was expensive, so you put all  the conviction you could behind it. When [8:48] you launch something, you wanted to have the best  hypothesis of what's going to land in the market, [8:53] the best problem hypothesis, the best solution  hypothesis. You wanted to be sure this was [8:57] going to win because if it didn't, it was really  expensive. And when you failed, it hurt. It hurt [9:03] a lot. It hurt the business, it hurt you, but it  forced this moment of reconciliation. Almost like [9:07] this post-traumatic growth. I believe there was  something about the world that worked this way, [9:12] and I saw something different. I have to figure  out what that difference is. Those painful [9:17] failures actually set up the foundation for future  success. And I'd argue today we can ship a lot [9:22] faster. We all see that. But we're shipping things  uh much faster in a way that we're not learning [9:27] from them. It's less painful when they fail.  We're not doing the same level of reflection. [9:32] So, instead of shipping one thing wrong  slowly, we're going through and shipping [9:36] through a hundred things wrong quickly without  building the foundation for future success. [9:43] I really like this framework from Jan Chipchase.  He is a field researcher. He was called the James [9:48] Bond of field research. He talks about  how sense-making happens. As a side note, [9:53] I love the term sense-making. It's like I'm not  making a product. I'm not making an artifact. I'm [9:57] just making sense, and that's important enough  for me and for others. But he goes through this [10:03] and you can see you start with identifying  a hypothesis. You collect this data. [10:07] You structure the information. You cluster the  knowledge. You synthesize insights, and ultimately [10:12] you evaluate wisdom. Six steps that all used to  take time. And they all used to be done together. [10:20] When I think about work I've seen as we shift  into an AI era, it starts to shift a little bit. [10:25] I don't know if this looks familiar to you all,  but it's a lot of one person prompting with [10:29] their thinking, and all six of those stages  kind of happening inside the thinking of AI. [10:35] And you can say, "Hey, I can jump to  the end of this much quicker." You can, [10:38] absolutely. I can produce an artifact, and I  can even share that artifact with the team, [10:42] and hopefully they'll read it deeply, but  they weren't part of the sense-making process. [10:47] In fact, you weren't part of the sense-making  process. So, what happens from this? You start [10:51] making decisions with less and less rigor of  thought. And then once you make those decisions, [10:56] it's harder and harder to align your teams around  them. Now, maybe you're thinking this is okay. [11:02] We'll just arm AI with better context. We'll  have it make more of the decisions, and now we [11:06] don't even need to make these decisions, so  it's fine. But to me, that is the invisible [11:10] asymptote in motion. Doing the same things,  expecting the outcome to grow in the way it did, [11:16] and hitting the ceiling where organizations  aren't putting out the output that they could. [11:22] Cuz I think ultimately, even in  the most advanced situations, [11:25] in the most advanced organizations rather, uh  people are forced to make not forced, people [11:30] make dozens if not hundreds of small decisions  every day. What to prioritize, what to read, [11:35] what to build next, who to talk to, who to update,  where to spend time. These decisions are being [11:41] made with shallower and shallower knowledge.  Executive summaries of executive summaries. [11:48] All right, so how do we get around this? Well,  [clears throat] let's take a look at where I think [11:51] a lot of organizations are today as they  move through this AI transformation. [11:55] I think over the past few months, especially,  there's been this great advancement in connecting [11:59] AI agents to systems of records. I think these are  two very important layers that people are learning [12:04] how to use closer and closer together. This is  where the speed of work is really accelerated, [12:08] but created this challenge I'm talking  about. Each agent works in a silo. [12:12] Systems store outcomes, not the reasoning behind  them, and execution happens before alignment. [12:20] So, I think what's missing is what we're calling  the collaboration layer. It's a layer in between [12:24] this. It's a layer designed for teams to  interact with the thinking of this layer, [12:28] to capture reasoning, to capture shared ways of  working, and design business-critical workflows. [12:34] It's a place that can make sense of these things  in an intuitive way for teams to come together, [12:38] so you don't build this debt of collective  intelligence by having individuals moving fast [12:43] in all different directions. And this isn't  just my observation. My my alma mater, MIT, [12:48] did a did a study where they found that 95% of  these uh enterprise GenAI pilots were delivering [12:55] essentially no measurable outcome. And the reason  wasn't the models. And so, they call the learning [13:00] gap. Companies weren't evolving their ways  of working um to around this technology. [13:07] The few the 5% of of these projects that actually  succeeded were designing friction in the process. [13:13] The rest of them were trying to engineer it out.  And I like this quote here. I'll just come over [13:17] and read this. I think it represents it well.  If the past decade of digital transformation [13:21] was about removing friction, the next decade  will be about designing it wisely. Those who [13:27] learn how to calibrate resistance enough to drive  learning rather than paralyze progress will be [13:33] the ones to capture not just ROI but resilience,  meaning, and lasting advantage. When I think about [13:39] the future of work, I think a lot of people are  focused on ROI. To me, the teams that have always [13:44] felt differentiated in the market have these later  parts, resilience, meaning, and lasting advantage. [13:50] But I'd argue whether you agree with those  two things being equally important or not, [13:54] they're all dependent on the ability to calibrate  the right level of friction in your processes. [14:00] Not all friction is good. Some of it's bad. I  don't want to wait for transcripts. I want to [14:04] be able to automate. I want data at my disposal.  But where are you engineering out the things that [14:08] matter the most? I believe that that collaboration  layer is the place to design for that friction. [14:16] And so, at Miro, this is what we think of as  intelligent collaboration. We're trying to think [14:19] through how do you build collective intelligence?  How do you create processes that lean into that [14:25] valuable friction versus engineering it away? You  know, for 15 years, Miro has been on the pulse of [14:31] collaboration challenges that have emerged with  new technologies. We helped people connect through [14:36] COVID in new ways. We helped connect to systems  of record to help build out enterprise-level [14:41] workflows. Now, we're seeing this emerge and  we're starting to connect our intelligent canvas [14:46] to those agents and to those systems of record  to help you build workflows that are visible, [14:51] that are coherent, that you can collaborate  on together. If you haven't seen Miro lately, [14:56] I highly recommend check it out. Check out  our booth. Reach out to me. Happy to tell [14:59] you more about it. I think there's a lot  of really exciting things going on there. [15:05] But as individuals, just to close out  here, I think there are a few things [15:07] you can do in your teams today to protect this  uncomfortable or this discomfort in your work. [15:12] To to find the ways to make sure you're not  engineering out the friction in the wrong places. [15:19] So, one, I think of as value the voiceover. Um I  think we have gotten really used to summarizing [15:25] docs, sending them, sharing them, reviewing  them. But when you send someone a summary, [15:30] you are inviting them to scan it. When you  ask them to walk through your thinking, you [15:34] are inviting a team to engage. I feel like we went  through this process where like meetings need to [15:38] be super efficient, send me a pre-read, we're all  going to read it, we're going to meet together, [15:42] we're going to make a decision, meeting over. And  I think we're overcorrecting on that too much. [15:46] Take time in meetings. Have people walk  through how they got to their reasoning, [15:50] where they felt challenges, where they had  subjective decisions to make. You know at [15:54] Miro we use talk tracks a lot. If you have to  be asynchronous, record it. Record that thinking [15:59] through. People will engage with that recording.  They will learn more. They will capture more of [16:03] the understanding that you have um, than if you  send them a summary, especially generated by AI. [16:10] The second one is automate the collection, not  the contact. As I said, information at your [16:14] fingertips is awesome. I love having data right  there. I don't need to go through all these uh, [16:19] loops to get data. Bring in transcript. Bring in  customer tickets. Uh, understand all the insights [16:25] that are coming from them. But protect those  moments of contact where you can. Cuz that's [16:29] where a lot of the intuition lives. Contact can be  sitting in on a customer call. Can be watching a [16:34] user hesitate with your product. Can be reading a  complaint rather than just a digest of complaints [16:39] that are out there. I love the tool the tool Gong.  I know there's lots of tools like that out there. [16:44] It has helped me learn more about customers and  about uh, the go-to-market motion at the company [16:50] more than I ever could without it. But this kind  of second order uh, effect has happened with tools [16:55] like Gong. Where now it's like I ask to sit in  on a customer call and they're just like, "Well, [16:58] we'll send you the recording. You don't  actually need to sit in. I don't want to, [17:01] you know, organize it around your calendar." Like,  that's the moment. That's the moment you see it. [17:06] There's no way I will be engaging with a recording  in the same way as when you are sitting there [17:11] feeling the pulse of the conversation, actively  in it. You all are here today versus listening [17:16] to a summary of what this could be. There's  some value in that that needs to be protected. [17:22] The third one, make AI share its thinking  visually. What I've seen is I use Claude a lot. [17:28] It's so good and getting better and better  at doing these complex multi-step tasks [17:33] while hiding all of its thinking. It will  generate really impressive outputs, but you [17:38] don't always see what's going on in the middle.  You know, I used to be a management consultant. [17:42] We built a lot of data models. If I build a really  complex data model and the output looks weird, [17:48] I'm not just going to take the output and put  like a correction factor on it and like take it [17:52] down 10%. I'm going to go back into the inputs.  I'm going to understand where it's pulling from. [17:56] I'm going to understand if formulas are off, if  there's different data that needs to go in there, [18:00] because I want to understand the whole flow of  this and make sure the answer is right. We should [18:04] ask the same of AI. We do this all the time at  Miro. We just export thinking from AI onto a Miro [18:09] canvas. You can see every step of it and you can  interrogate it. You can collaborate on it. You can [18:14] look at it all together. So, your team is working  through that thinking together versus separate. [18:21] Reward the best questions, not the best answers.  I feel like this has been something that's emerged [18:25] since the start of ChatGPT, where you used  to be able to be the smartest person in the [18:29] room by having the answer on demand. And  now answers are free. Everyone has them. [18:33] It's super easy to get it. But, there are  those moments. You probably have all felt it, [18:37] where you're in a conversation, you're in a  discussion, and someone asks the right question [18:42] that changes your perspective of the problem. It  changes the course of the work you're doing and [18:46] changes the course of the output you're creating.  Those moments you need to lean into. You need [18:51] to reward them. You need to encourage them.  You need to recognize when they're happening. [18:57] And the last one, design your work around  retros. I have always been a proponent of retros, [19:01] maybe back to my Navy days, where we did retros  all the time. Retros, I think, are one of the [19:06] most sing like the one of the single value most  valuable learning tools you can do. They're also [19:10] the easiest thing to skip. They happen inherently  after you've shipped something. You don't [19:15] necessarily need to do it. If you are ever trying  to test if your team or your organization is set [19:21] up to build collective intelligence, look at  how consistent and high quality your retros are. [19:27] Because retros force that moment of learning. And  I think there's some thinking that maybe if you're [19:32] shipping really fast, retros are less important  because there's just not that much time for you to [19:36] retro. But I would say if you are shipping at high  volume and not creating that moment of reflection, [19:42] there is no learning. There's just faster  forgetting. You're not going through that process. [19:47] And when you are doing retros and you  have a structured way of approaching them, [19:50] it changes the whole course of your work. You're  more structured with your hypothesis starting it. [19:54] You're more deliberate about how you structure  your work. And you're more careful about how [19:58] you measure success to be able to have the right  conversation. When I look at these five things, [20:04] none of these things should slow you down. They  should just make sure that the speed is buying you [20:10] understanding, not just output. And that's going  to be critical in this day and age. That's all I [20:17] have. Um as I mentioned, my name is Brendan  McCook. Um if this is a topic that resonates [20:21] with you, reach out to me. I love chatting about  this. I'd love to show you how we're thinking [20:25] about using Miro to really be this collaboration  layer and help teams navigate this friction in a [20:30] productive way in a time where everyone else is  trying to engineer it out. Thank you very much.