Glean Work Lab Chief Links AI Enterprise Success to Org Chart Overhaul
Ai4
Organizations achieve meaningful AI productivity gains by pairing executive mandates with peer influencers and integrating algorithms directly into existing daily workflows rather than deploying standalone tools.
## Background
Published on YouTube by Ai4 on August 26, 2025, this video features a presentation by Rebecca Hinds, Head of the Work Transformation Lab at Glean. According to the page description, Hinds breaks down five strategic shifts that separate high-performing AI organizations from those stuck in isolated pilots, offering a practical path forward to scale enterprise-wide results.
## TL;DR
Rebecca Hinds presents five fundamental organizational shifts required to move from stagnant AI experimentation to true enterprise transformation. She argues that companies must reframe AI from a transactional tool to a collaborative teammate, pair top-down mandates with peer AI influencers, eliminate workflow integration friction, and redirect AI toward improving team collaboration rather than isolated output. Finally, she advocates for restructuring legacy org charts to allow AI cross-departmental agility, using real-world examples like Moderna to demonstrate how aligning human change management with technology yields real impact.
## Atmosphere & style
The presentation is insightful, authoritative, and research-driven, delivered with the clear, structured cadence of an executive masterclass. Hinds balances academic rigor with relatable workplace humor, turning practical empirical data into an engaging narrative about human behavior in the corporate world. Watching it feels like a compelling strategy session that elevates abstract AI concepts into concrete, actionable steps for organizational change.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) Success with AI transformation requires reframing the technology's relationship with employees, which led some organizations to rebrand "artificial intelligence" as "augmented" or "amplified intelligence" to reduce psychological friction.
- [00:20](https://www.youtube.com/watch?v=f-7B300g604&t=20s) Only 9% of workers view AI as a collaborative teammate—and see significant productivity gains—while 53% view it as merely a transactional tool.
- [00:20](https://www.youtube.com/watch?v=f-7B300g604&t=20s) Research shows employees hold an automatic bias against AI, rating identical leadership responses as less helpful simply when they knew an AI generated them.
- [05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) Top-down executive mandates reach only 30%–35% of employees; pairing mandates with internal "AI influencers" nearly doubles reach to 60%–70%.
- [05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) Gamified events like promptathons help surface non-technical AI champions across departments and reward iterative skill growth.
- [08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) The "toggle tax"—switching between apps over 1,200 times a day—causes severe distraction, requiring over 23 minutes to regain focus after a context switch.
- [08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) To drive adoption, organizations must integrate AI directly into existing workflows rather than adding to the growing multitude of standalone generative AI tools.
- [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) Over-indexing on individual AI productivity creates a "tragedy of the commons" with automated noise that burdens teams, whereas high-value AI targets team-level collaboration.
- [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) AI can analyze team dynamics like "discursive diversity" or serve as a meeting moderator to optimize brainstorming, prevent groupthink, and balance participation.
- [16:13](https://www.youtube.com/watch?v=f-7B300g604&t=973s) Traditional 19th-century org charts isolate functional silos, whereas modern AI transformation requires fluid structures like Moderna's merger of HR and IT under a single executive.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) Framing AI Beyond a Threat
Rebecca Hinds reflects on the origins of her fascination with artificial intelligence during her PhD program in 2019, when she studied retail companies adopting the technology. She observed a stark dichotomy: while some businesses achieved tangible, game-changing outcomes, others ended up with expensive technology "collecting digital dust." Diving deeper through dozens of interviews with executives and frontline employees, a distinct pattern emerged. The organizations that succeeded did not merely install new software; they completely reframed what AI meant to their workforce.
To overcome employee resistance, several successful leaders took the radical step of banning the phrase "artificial intelligence" altogether. Etymologically, the word "artificial" felt inherently synthetic, deceptive, and directly opposed to human intelligence, creating an emotional barrier that invited friction rather than adoption. Instead, companies rebranded the technology as "amplified intelligence" or "augmented intelligence" to emphasize human capability rather than human replacement.
- **Origin Story**: Research began in 2019 during Hinds's PhD study analyzing AI implementation across retail enterprises.
- **Career Trajectory**: Hinds previously led the Work Innovation Lab at Asana and now leads thought leadership and the Future of Work Lab at Glean.
- **Core Thesis**: Breakthrough AI transformation requires five fundamental organizational shifts, starting with reframing the relationship between human and machine intelligence.
---
[00:20](https://www.youtube.com/watch?v=f-7B300g604&t=20s) Shift 1: Moving from "Tool" to "Teammate"
The first major shift centers on psychological framing. Research conducted by the Work Innovation Lab reveals that 53% of workers view AI strictly as a transactional "tool"—akin to a hammer or a calculator that is picked up for a specific task and immediately set down. However, a distinct 9% of employees perceive AI as a "teammate." This small cohort reports dramatically higher productivity gains because they interact with the system relationally. Instead of asking "What can the technology do for me?", they ask "What can I do *with* the technology?", pushing past initial prompts to collaborate iteratively.
Mindset dictates adoption because human beings naturally reject or discount algorithmic outputs if they lack a relatable human feeling, regardless of how accurate the technology is. Hinds highlights a study by Raj Choudhury at Harvard, where an AI bot was trained on years of communications—emails, meeting transcripts, and documents—from Zapier CEO Wade Foster. Employees could only distinguish the real CEO from the bot 58% of the time (slightly better than a coin flip). Yet, whenever workers *believed* a response originated from the bot, they consistently rated it as less helpful, proving that perceived artificiality invites automatic bias.
- **53% of Workers**: View AI as a transactional tool.
- **9% of Workers**: View AI as a collaborative teammate and see significantly higher productivity gains.
- **Harvard/Zapier Study**: Employees rated AI-generated leadership responses as less helpful simply because they knew an AI generated them, despite accuracy matching the real CEO.
---
[05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) Shift 2: Balancing Mandates with Grassroots "AI Influencers"
Executive memos and top-down mandates telling employees to use AI "or else" routinely fail to drive meaningful enterprise-wide change. True transformation requires a dual approach that pairs executive direction with organic, bottom-up momentum. Frontline workers do not adopt new habits simply because executives at the top of the organizational totem pole order them to; they adopt them when trusted peers whom they relate to demonstrate the value of the change.
Finding and empowering these internal "AI influencers" is critical, though difficult, because they rarely hold fancy titles or sit in traditional IT roles. Hinds points to learning platform Udemy, which hosted a company-wide "Learns AI Week" structured as a promptathon. The event served two purposes: it acted as a natural selection mechanism to surface passionate AI champions, and it explicitly rewarded continuous improvement—celebrating the best "before and after" prompt iterations and fostering collaborative co-creation rather than just single-output speed.
- **30%–35% Reach**: Maximum employee reach achieved by top-down executive mandates alone, according to research by Rob Cross (Babson College).
- **60%–70% Reach**: Employee reach achieved when executive mandates are paired with activated peer influencers (nearly double the effectiveness).
- **Udemy Promptathon Model**: Gamified internal hackathons that evaluate "before and after" prompt growth to identify non-technical champions across departments.
---
[08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) Shift 3: Eliminating the "Toggle Tax" and Integration Friction
Even with the correct mindset and enthusiastic culture, AI adoption frequently hits a brick wall known as the "toggle tax." In the modern Fortune 500 workplace, employees switch between different application windows over 1,200 times a day. This constant context-switching converts daily workflows into a chaotic game of "digital pinball." Research shows that once a worker is distracted during an app toggle, it takes an average of over 23 minutes to fully regain focus on their original task.
Flooding the workplace with standalone generative AI tools only exacerbates cognitive fatigue. Glean's enterprise research indicates that large companies utilized an average of 150 generative AI applications last year, a number projected to double to over 350 by the end of this year. When using AI requires opening another browser tab, managing another login, and adding to the toggle tax, usage drops off sharply. The solution is not adding *more* AI, but embedding it directly into existing daily workflows so it operates almost invisibly.
- **1,200+ Toggles**: Daily application switches made by average enterprise knowledge workers.
- **23+ Minutes**: Time required to recover focus after getting distracted during a context switch.
- **150 to 350+ AI Tools**: Rapid growth trajectory of distinct generative AI applications deployed within average enterprise environments.
---
[10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) Shift 4: Transforming Collaboration and Meetings
Most executives treat AI exclusively as an individual productivity play, asking how it can help single employees produce more work faster. Over-indexing on individual metrics creates a workplace version of the "tragedy of the commons"—employees use AI to generate floods of summary notes, redundant notifications, and automated content that burden everyone else. While the individual appears hyper-productive on paper, overall team productivity plummets. Real value lies in leveraging AI at the team level to repair broken collaboration, which consumes 85% to 90% of a knowledge worker's workday (a 50% increase over the last two decades).
AI can monitor collaboration mechanics to optimize team performance. In academic studies on "discursive diversity"—the degree to which team members vary their vocabulary during discussions—high-performing teams were found to use diverse language during creative brainstorming, but naturally converged on shared vocabulary during execution phases. Soon, real-time AI will join meetings to gently nudge teams toward linguistic divergence or convergence depending on the project phase.
- **50% Increase**: Growth in collaboration demands placed on knowledge workers over the last 20–30 years (Rob Cross research).
- **85%–90% of Worktime**: Percentage of time knowledge workers spend in collaborative activities.
- **5.5x Multiplier**: High-performing organizations are 5.5 times more likely to reward effective team collaboration over isolated output.
- **AI Meeting Moderators & Twins**: AI used as an objective third party can flag toxic bulldozing, act as a "devil's advocate" to disrupt groupthink, or attend routine updates as a digital twin (though if everyone sends a bot, the meeting should simply be an asynchronous document update).
---
[16:13](https://www.youtube.com/watch?v=f-7B300g604&t=973s) Shift 5: Rewiring Org Charts for Fluid Complexity
It is impossible to drive an AI transformation using organizational structures designed in the 19th century. Traditional org charts—originating around 1896—were intentionally engineered to reduce operational complexity by siloing authority into rigid domains, such as establishing distinct decision-making boundaries for a men's clothing department versus a women's department.
However, modern AI thrives by looking *across* departmental silos to optimize global outcomes rather than local sub-optimums. Hinds highlights research co-conducted with Stanford professor Melissa Valentine studying a digital retailer implementing AI for apparel assortment planning. To let the algorithm evaluate cross-departmental profitability holistic patterns, the company had to make its org chart vastly more flexible. A prominent modern example is Moderna, which merged its Human Resources and IT departments under a single Chief People and Digital Technology Officer (formerly the CHRO) to bridge human adoption expertise with technical implementation.
- **1896 Org Chart Legacy**: Modern enterprise structures still mirror 130-year-old operational models built solely to isolate complexity.
- **Moderna Merger**: Combined HR and IT under one executive leader to seamlessly pair human change management with technical scaling.
- **Actionable Takeaway**: Organizations do not need to tackle all five shifts at once; leaders should select one or two core shifts to execute with full commitment.
## Mentioned
Rebecca Hinds, Glean, Work Transformation Lab, Work Innovation Lab, Asana, Future of Work Lab, Raj Choudhury, Harvard, Zapier, Wade Foster, Rob Cross, Babson College, Udemy, Stanford, Melissa Valentine, Moderna
## Video description
Rebecca Hinds, Head, Work Transformation Lab, Glean
Abstracts
What separates organizations that achieve real AI transformation from those that don’t? Most are investing in AI, but few are seeing meaningful impact. In this talk, I’ll break down five strategic shifts that set high-performing AI organizations apart—from reshaping employee mindsets to rewiring the org chart for scale. I’ll share research-backed practices that help teams move from isolated pilots to enterprise-wide results. If your AI strategy feels siloed, stalled, or stuck in experimentation mode, this session will give you a practical path forward.
Glean Work Lab负责人：企业AI成功离不开组织架构重组
企业若想通过AI实现切实的生产力提升，需将管理层指令与同行倡导者相结合，并将算法直接嵌入现有的日常工作流中，而非部署孤立的工具。
Rebecca Hinds提出了从停滞的AI试探迈向真正企业转型所需的五大根本性组织变革。她指出，企业必须将AI从事务性工具重新定义为协作队友，将自上而下的指令与内部同伴AI倡导者相结合，消除工作流程整合阻力，并引导AI用于改善团队协作而非仅仅提高孤立产出。最后，她主张重组传统组织架构以赋予AI跨部门的敏捷性，并以Moderna等现实案例说明，唯有将人的变革管理与技术相协调，才能产生真正的实效。
仅 9% 的员工将 AI 视为协作“队友”并获得显著的生产力提升，而 53% 的人仅将其视作事务性工具；研究显示员工对 AI 存在固有偏见，知晓内容由算法生成时往往会给出更低的效用评价。 (0:20)
自上而下的管理层指令通常仅能触达 30%–35% 的员工，而将高管推动与基层“AI 影响力者”相结合，可将整体采纳率近乎翻倍至 60%–70%。 (5:13)
知识工作者每天在各类应用间切换超过 1,200 次，所造成的“切换税”使其每次分心需逾 23 分钟才能重新聚焦，因此推动落地的关键是将 AI 深度内嵌于现有工作流，而非堆砌独立工具。 (8:48)
过度追求个体生产力容易引发生成海量无效信息的“公地悲剧”；高价值的 AI 应用应着眼于改善占据员工 85%–90% 工作时间的团队协作体验与会议机制。 (10:35)
源于 19 世纪末的传统筒仓式组织架构限制了 AI 的跨部门全局优化能力，企业需重塑灵活组织结构，如 Moderna 合并人力资源与 IT 部门以打通技术落地与组织变革。 (16:13)
Reading summary
Rebecca Hinds of Glean argues that enterprise AI transformation succeeds not through technology deployment alone, but through fundamental organizational and cultural shifts that reframe human-machine collaboration.
Reframing AI as a Collaborative Teammate (0:20)
To counter employee apprehension, Hinds notes that some enterprises avoid the term "artificial intelligence"—viewing "artificial" as synthetic and oppositional to human capability—in favor of phrases like "augmented" or "amplified" intelligence. Research from the Work Innovation Lab indicates that 53% of workers treat AI strictly as a transactional tool used for discrete tasks and put away. In contrast, the 9% of employees who approach AI relationally as a "teammate" report significantly higher productivity gains by engaging in iterative, co-creative dialogue.
Adoption hurdles are frequently psychological rather than technical. In a Harvard study led by Raj Choudhury, an AI model was trained on communication archives from Zapier CEO Wade Foster. While employees could distinguish the real executive from the model only 58% of the time, they consistently rated identical advice as less helpful whenever they believed it originated from the bot, demonstrating an inherent bias against perceived artificiality.
Activating Peer Influencers Alongside Executive Mandates (5:13)
Top-down executive directives demanding AI usage typically fail to create lasting behavioral change across an organization. Citing research by Babson College's Rob Cross, Hinds highlights that executive mandates reach only 30% to 35% of a workforce. However, pairing leadership direction with trusted, grassroots peer influencers nearly doubles reach to between 60% and 70%.
Because internal champions rarely hold specialized technical titles, identifying them requires deliberate programming. Learning platform Udemy addressed this by hosting an internal "Learns AI Week" structured as a promptathon. The event surfaced non-technical advocates across departments by evaluating employee progress between "before and after" prompt iterations rather than rewarding raw output speed.
Eliminating Workflow Friction and Toggle Tax (8:48)
Even when employees embrace AI, adoption often stalls due to operational friction. Fortune 500 knowledge workers switch between applications over 1,200 times per day, and regaining focus following an interruption takes an average of more than 23 minutes.
Deploying isolated generative AI applications compounds this cognitive fatigue. Glean's enterprise research shows that large organizations used an average of 150 generative AI applications last year, with that figure projected to exceed 350 by the end of the year. Hinds argues that requiring workers to juggle separate browser tabs and logins diminishes usage, making it essential to embed AI directly and unobtrusively into existing daily workflows.
Prioritizing Team Collaboration Over Individual Output (10:35)
Evaluating AI primarily as an individual productivity tool risks creating a "tragedy of the commons," wherein workers generate excessive automated summaries, notifications, and drafts that place administrative burdens on their peers. Collaboration demands on knowledge workers have increased by 50% over the past two decades, now accounting for 85% to 90% of their workday, and high-performing organizations are 5.5 times more likely to reward collaborative success over isolated output.
Instead of optimizing single-user metrics, enterprises can deploy AI to support group dynamics. Research on "discursive diversity" demonstrates that high-performing teams use varied vocabularies during creative brainstorming but converge on shared language during execution. Hinds envisions real-time AI serving as an objective facilitator in meetings to guide linguistic transitions, balance speaking time, and challenge groupthink.
Restructuring Siloed Hierarchies for Cross-Functional AI (16:13)
Enterprise org charts modeled after 1896 organizational frameworks were designed to isolate operational complexity by confining authority to strict departmental silos. However, modern AI algorithms operate by analyzing cross-functional datasets to optimize enterprise-wide outcomes rather than localized department goals.
Research co-conducted by Hinds and Stanford professor Melissa Valentine on apparel assortment planning demonstrated that algorithmic optimization requires flexible, cross-boundary collaboration. Reflecting this structural evolution, Moderna consolidated its Human Resources and IT divisions under a unified Chief People and Digital Technology Officer, aligning change management capabilities directly with digital systems deployment.
Hinds cautions enterprise leaders against attempting all five organizational transformations simultaneously, recommending instead that organizations select one or two core structural or cultural shifts to execute deeply.
精炼阅读摘要
Glean 未来工作实验室负责人 Rebecca Hinds 认为，企业人工智能转型失败的主因不在技术本身，而在组织架构与心理认知；实现突破需要企业完成五项核心组织转变，重塑人机协作与运营模式。
心理认知：从单向工具转向协作伙伴 (0:20)
推动人工智能落地的首要阻碍来自员工的心理抗拒。Hinds 指出，“人工智能”（Artificial Intelligence）一词中的“人工”易引发虚假与取代人类的负面联想，部分成功企业甚至直接将其重新定义为“强化智能”（Amplified Intelligence）或“增强智能”（Augmented Intelligence）。Work Innovation Lab 的调研显示，53% 的员工将 AI 视为像计算器一样用完即弃的事务性“工具”，仅有 9% 的员工将其视为“团队伙伴”。然而，正是这 9% 采取迭代协作态度的群体报告了显著更高的生产力提升。
技术输出即使高度精准，也常常因员工的心理偏见而被贬低。哈佛大学 Raj Choudhury 针对 Zapier 的研究显示，基于首席执行官 Wade Foster 多年沟通记录训练出的 AI 机器人，其生成内容仅有 58% 能被员工与真人区分开；但只要员工知晓或认定回复来自 AI，就会习惯性地给出更低的有益度评价。这表明企业若不重塑员工对人机关系的认知，技术成果便难以被客观接纳。
推广策略：自上而下指令结合基层倡导者 (5:13)
仅凭管理层的高压命令无法促成企业范围内的深层转变。Babson College 教授 Rob Cross 的研究表明，纯粹依靠自上而下的行政指令，其员工触达率通常仅为 30% 到 35%；而若将管理层方向与同僚中的“AI 意见领袖”相结合，触达率可跃升至 60% 到 70%。一线员工往往不会因为高层命令而改变工作习惯，却更容易受到身旁可信赖同事的示范带动。
这些基层推动者往往分散在业务一线，不一定拥有显赫头衔或 IT 背景。在线学习平台 Udemy 通过举办全员性质的“AI 学习周”（以 Prompt 大赛形式展开）来发掘内部人才。该活动通过评估员工在提示词设计上的“前后对比”进步幅度，奖励跨部门的协作共创与持续迭代，从而自然筛选出具备热情的非技术型推广骨干。
工作流整合：消除“切换税”与工具膨胀 (8:48)
即使企业具备积极文化，AI 工具落地也常受制于“切换税”（Toggle Tax）。在现代财富 500 强工作场所中，知识型员工每天在不同应用窗口间切换超过 1,200 次，工作流程被切割得支离破碎。数据显示，员工在切换应用发生分心后，平均需要超过 23 分钟才能重新集中注意力回到原本任务。
独立生成式 AI 工具的泛滥正在加剧认知疲劳。Glean 的企业研究显示，大型企业去年平均使用约 150 款生成式 AI 应用，预计到今年底这一数字将激增至 350 款以上。如果使用 AI 意味着必须额外打开浏览器标签页并管理新账号，其实际使用率便会骤跌。解决之道并非采购更多独立工具，而是将 AI 无缝嵌入员工的日常工作流中，使技术近乎隐形地发挥效用。
协作机制：从个人提速转向团队协同重构 (10:35)
将 AI 仅仅视为个人提速工具容易引发职场“公地悲剧”：单个员工利用 AI 大量生成会议摘要、多余通知和自动化文档，表面上看个人产出激增，实则给其他团队成员带来严重的信息过载。Rob Cross 的研究指出，现代知识工作者有 85% 到 90% 的工作时间用于协作，这一负担在过去二三十年间增加了 50%。高绩效组织优先奖励团队协同的概率是单独个人产出的 5.5 倍。
AI 的更大价值在于作为中立第三方修复团队协同机制。关于“话语多样性”的学术研究表明，高效团队在头脑风暴期使用多元词汇，在执行阶段则自然收敛至统一词汇；AI 可在会议中实时监测语言模式，引导团队在不同阶段实现发散或收敛。此外，AI 还能识别会议中的言语霸凌行为、充当“反面意见提出者”打破群体迷思，或作为数字分身代为出席常规通报会——尽管若所有参会者都派遣机器人出席，该会议理应直接改为异步文档沟通。
组织重组：打破筒仓以适应全局算法优化 (16:13)
传统的企业组织架构雏形可追溯至 1896 年，其初衷是通过划分部门界限（如严格区分男装部与女装部的决策权）来隔离和降低运营复杂性。然而，现代 AI 的优势恰恰在于打破壁垒、实现跨部门的全局最优分析，这使得百年前的静态结构成为算法落地的瓶颈。
Hinds 与斯坦福大学教授 Melissa Valentine 针对一家数字化零售商的研究显示，为了让算法能够跨部门评估整体服装选品组合与利润，企业必须重塑组织架构以赋予其更高流动性。现实中的典型实践如生物制药公司 Moderna，该公司将人力资源部门与 IT 部门整合，设立由原首席人力资源官领导的“首席人力与数字技术官”，将负责组织文化与员工变革的管理经验同底层技术实施直接绑定。
Hinds 强调，企业不必试图同时推进全部五项变革，领导层应挑选一到两个与当下痛点最契合的领域重点突破。然而，如何在消除应用切换成本的同时统一全公司的数据流动，以及传统垂直晋升通道如何适应横向流动的新型组织架构，仍是企业在实践中需要持续权衡的问题。
Detailed notes — zh
## Background
该视频由 Ai4 频道于 2025 年 8 月 26 日发布在 YouTube 平台上。视频主讲人为 Glean 的 Work Transformation Lab 负责人 Rebecca Hinds。根据页面简介，内容主要探讨将企业 AI 战略从孤立试点推进至全公司落地的五大关键变革，涵盖员工心态重塑与组织架构重组等核心议题。
## TL;DR
演讲者结合多年的企业研究与案例分析，阐述了推动 AI 变革成功所需的五项战略转变。企业需要引导员工将 AI 从单纯的“工具”重新定义为合作的“队友”，并通过“AI 影响力者”以自下而上的方式带动全员采纳。同时，必须消除“应用切换税”，将 AI 融入现有工作流，并把焦点从个人生产力转向团队级的深度协作。最后，企业应打破 19 世纪式的传统部门壁垒，重构能够支撑 AI 全局优化的柔性组织架构。
## Atmosphere & style
整场演讲呈现出专业、务实且充满洞察力的学术与商业分享氛围。演讲者将丰富的心理学数据、顶级学府的研究案例与真实的企业转型实践交织在一起，语言严谨而富有条理。内容从剖析员工对 AI 的心理抵制与认知误区切入，逐步升维至对百年传统组织架构的深度反思，为观众带来兼具理论深度与实践落脚点的沉浸式思考体验。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) 组织推动 AI 变革需要实现五大关键转变，从重塑员工心智模型到重构适应 AI 时代的组织架构。
- [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) 转变一：重新定义员工对 AI 的认知定位，从单向使用的“工具”转变为共同协作的“队友”。
- [01:28](https://www.youtube.com/watch?v=f-7B300g604&t=88s) 53% 的员工仅将 AI 视为普通工具，而 9% 将其视为“队友”的群体展现出更高的生产力提升效率。
- [03:42](https://www.youtube.com/watch?v=f-7B300g604&t=222s) 员工对 AI 存在主观心理抵制，只要以为回答来自 AI，即使客观准确度极高也会被认为缺乏帮助。
- [05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) 转变二：拒绝纯自上而下的行政强制指令，转向结合并激活自下而上的“AI 影响力者”。
- [05:39](https://www.youtube.com/watch?v=f-7B300g604&t=339s) 纯行政指令只能覆盖 30%-35% 的员工，结合内部“AI 影响力者”能将采纳率提升至 60%-70%。
- [07:21](https://www.youtube.com/watch?v=f-7B300g604&t=441s) Udemy 举办 Promptathon 竞赛，通过评估“优化提升幅度”而非单看绝对水平来筛选并培养关键变革者。
- [08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) 转变三：消除跨系统切换的“应用切换税”，将 AI 直接无缝无形地融入现有工作流界面中。
- [08:55](https://www.youtube.com/watch?v=f-7B300g604&t=535s) 500 强企业员工每天切换应用达 1200+ 次，被打断后重新聚焦平均需要超过 23 分钟。
- [09:34](https://www.youtube.com/watch?v=f-7B300g604&t=574s) 企业内部 AI 工具激增，核心矛盾已非工具数量匮乏，而是未将 AI 放在正确的嵌入式入口上。
- [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) 转变四：从过度聚焦“个人生产力”转向重塑“团队协作层”的 AI 赋能。
- [11:02](https://www.youtube.com/watch?v=f-7B300g604&t=662s) 过分强调个人效率易引发“公地悲剧”，单点生成的大量冗余内容反而会严重拖累团队整体效率。
- [11:49](https://www.youtube.com/watch?v=f-7B300g604&t=709s) 知识工作者 85%-90% 的时间用于协作，AI 应作为主持人或反方代表参与动态引导，破除群体思维。
- [16:13](https://www.youtube.com/watch?v=f-7B300g604&t=973s) 转变五：打破 19 世纪式按部门隔离的传统组织架构，构建能支持全局最优解的柔性 Org Chart。
- [17:02](https://www.youtube.com/watch?v=f-7B300g604&t=1022s) 传统部门架构旨在降低管理复杂性，而 AI 的优势在于跨部门处理复杂性并寻求全公司级别的“全局最优”。
- [19:08](https://www.youtube.com/watch?v=f-7B300g604&t=1148s) Moderna 将 HR 与 IT 部门合并由单一高管领导，以成功弥合“技术部署”与“组织变革”之间的天然鸿沟。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) 转变一：重塑员工对 AI 的心智模型（从“工具”到“队友”）
Rebecca Hinds 总结了她过去六年针对企业 AI 落地与研究的洞察（先后在 Stanford 读博、在 Asana 领导 Work Innovation Lab，现任 Glean 思想领导力及 Future of Work Lab 负责人）。早在 2019 年研究零售企业部署 AI 时，她就发现成功与失败的企业形成了鲜明对比：成功的公司不会只把 AI 当作又一项新工具，而是对其重新定义（Reframe），甚至有公司禁用“Artificial Intelligence”一词，认为“Artificial”（人工/虚假）天生自带威胁感与对抗意味，因而将其改称为“Amplified Intelligence”（增强智能）或“Augmented Intelligence”（扩增智能）。
根据 Work Innovation Lab 的心理学研究数据，绝大多数员工（53%）仍将 AI 视为一种像锤子或计算器一样的**“工具”（Tool）**——即“用完即用即扔”的事务性关系；然而，仅有 **9%** 的员工将 AI 视为**“队友”（Teammate）**。这 9% 的群体在生产力提升方面的汇报率显著更高。因为当员工将 AI 视为队友时，他们不再被动地问“这技术能帮我做什么？”，而是主动思考“我和它能一起完成什么？”，并在第一次 Prompt 结果不佳时像对待人类同事一样继续与 AI 协作深入探讨。
```
员工对 AI 的认知定位与行为差异：
┌──────────────────────────────────────────┐
│ AI 视作工具 (53%) │
│ 观点：事务性工具（如锤子、计算器） │
│ 行为：用完即弃，首轮 Prompt 不佳即放弃 │
└──────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────┐
│ AI 视作队友 (9%) │
│ 观点：合作共创者 │
│ 行为：持续追问与协作，生产力提升率显著提高 │
└──────────────────────────────────────────┘
```
演讲者援引哈佛大学 Raj Choudhury 教授针对 Zapier CEO Wade Foster 的 AI 替代测试：即使 AI 模型的回复在客观准确度上极高，但如果员工**主观上以为**该回复来自 AI Bot，他们对回答实用性的评估就会大幅打折。这表明：**即便 AI 在技术上完全正确，如果不具备“人类队友”般的情感与协作体验，依然会遭到员工的心理抵制**。
- 53% 的员工将 AI 视为普通工具，仅有 9% 的员工将其视为“队友”。
- 将 AI 视为队友的员工，其生产力改善幅度显著高于普通用户。
- 哈佛大学 Raj Choudhury 团队将 Zapier CEO Wade Foster 的多年沟通数据（邮件、会议纪要等）训练成 AI Bot，员工仅有 58% 的概率能分辨真人与 Bot（接近掷硬币概率）。
- 研究核心结论：只要员工“以为”回答来自 AI，无论事实是否如此，都会主观评价该回答“缺乏帮助”。
---
### [05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) 转变二：自上而下的行政指令 vs 自下而上的“AI 影响力者”
许多公司试图通过行政命令（Top-down mandates）强推 AI 普及，甚至发出“不用 AI 就解雇”的警告，但这种做法往往适得其反，导致员工只是出于敷衍（Going through the motions）而形式化使用。高效的企业深知，真正的变更管理（Change management）需要将自上而下的政策与自下而上的运动相结合。员工不会因为高层老板的命令而心服口服，但他们会跟随身边信任、有同理心的同侪。
在每个组织内部，都存在一群具有超高同行影响力的**“AI 影响力者”（AI Influencer）**。这些关键人物往往不是层级最高的高管，也不是技术最顶尖的技术专家。为了精准识别并激活这些关键人，在线教育平台 Udemy 举办了一场名为“Learns AI Week”的全公司 Prompt 提示词大赛（Promptathon）。该活动巧妙利用两个机制：一是将竞赛作为筛选机制，识别出谁在积极参与并充满热情；二是不单奖励输出质量最高的人，而是特别奖励“优化前后对比最大”（Best before & after）以及在 Prompt 合作共创中表现突出的员工。
- 根据 Babson College 的 Rob Cross 教授的研究：纯粹依靠自上而下的行政变更指令，仅能覆盖大约 30%–35% 的员工。
- 当结合并激活组织内部的“AI 影响力者”后，AI 技术的全员采纳率可直接翻倍，提升至 60%–70%。
- Udemy 通过举办全公司范围的 Promptathon，以“前后提升幅度”和“协作反馈”为评分标准，成功遴选并激活了内部的 AI 影响力者。
---
### [08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) 转变三：消除“应用切换税”，将 AI 融入已有工作流
哪怕员工拥有正确的心智模型，也有自下而上的推动力，阻碍 AI 落地的一大硬伤是**“应用切换税”（Toggle tax）**。在世界 500 强企业中，员工每天在不同系统与软件应用之间来回切换的次数高达 1,200 次以上。每一次切换都会分散注意力，研究显示，员工一旦被迫中断当前工作，平均需要超过 23 分钟才能重新聚焦到原始任务上，工作体验如同“数字弹珠台”（Digital pinball）。
Glean 去年的调查研究显示，目前大型企业内部平均已在使用约 150 种生成式 AI 工具，而到今年年底这一数字预计将翻倍至 **350 种以上**。如果每次使用 AI 都需要员工打开新网页、输入新账号登录并承受更高的切换成本，采纳率必然暴跌。因此，企业面临的问题并非“AI 工具不够”，而是“AI 没有被放在正确的位置”。AI 必须做到嵌入式（Embedded）、一体化（Integrated），甚至在日常工作流中达到“近乎无形”的状态。
- Fortune 500 企业员工平均每天进行切换应用/标签页（Toggle）动作达 1,200+ 次。
- 任务因切换而被打断后，员工重新恢复深度专注平均需要 23 分钟以上。
- 大型企业内部使用的生成式 AI 工具数量正从 150 种激增至 350 种以上。
- 解决 Toggle tax 的关键是减少新入口，将 AI 直接无缝嵌入员工已有的日常工作界面中。
---
### [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) 转变四：从“个人生产力”转向“团队协作层”的 AI 赋能
当前绝大多数高管和开发者都将 AI 局限在“个人生产力工具”（Personal productivity tool）的定位上，试图用它来实现“减员增效”。然而过度聚焦个人生产力极易引发经济学上的**“公地悲剧”（Tragedy of the commons）**：当某个员工为了追求个人效率最大化而用 AI 生成大量通知、邮件和内容分发给他人时，个人看起来效率极高，但团队和组织层面的整体效率却面临断崖式下跌。
知识工作者（Knowledge workers）每天有 85% 到 90% 的时间在进行团队协作（这一数字在过去 20-30 年间增长了 50%），但绝大多数协作都是极低效且混乱的，因为传统的协作过程难以被量化和察觉。研究表明，高绩效团队在创意阶段需要极高的**“语篇多样性”（Discursive diversity，即使用多元的词汇与句式），但在执行和协同阶段则需要快速收敛（Converge）。AI 能够在跨平台协作中扮演“会议主持人”或“反方代表”（Devil's advocate）的角色——实验证明，当一个中立的 AI 在会议中提出“大家是否考虑过其他不同观点？”时，不仅能有效破除群体思维（Groupthink），还能显著提升参会者对会议最终决策的满意度。
- 知识工作者日常 85%–90% 的工作时间用于协作，过去 20 到 30 年间协作时长增加了约 50%。
- 高绩效组织奖励有效协作的概率是普通企业的 5.5 倍。
- 优秀团队的协作规律：创新探索期保持高“语篇多样性”（Discursive diversity），落地执行期则降低多样性以达成快速对齐。
- AI 在会议中的应用不应止步于“自动转写与纪要”，而应参与到动态引导中（如拦截霸道发言、担任反方角色解构群体思维、甚至以 Digital Twin 数字分身形式参会）。
- 演讲者提醒：如果大家都派 AI 数字分身参加某个会议，说明该会议根本不该召开，应直接改为异步邮件或项目卡片更新。
---
### [16:13](https://www.youtube.com/watch?v=f-7B300g604&t=973s) 转变五：打破 19 世纪式组织架构，构建适应 AI 时代的柔性 Org Chart
如果企业的组织架构（Org chart）仍停留在上个时代，AI 变革将寸步难行。当下的很多企业架构图与 1896 年（130 年前）的垂直部门树状图几乎毫无二致。从组织行为学来看，传统组织架构的本质是为了**“降低复杂性”**（通过划清控制跨度、汇报线和决策边界来隔离风险）；然而，AI 具备处理跨部门复杂性的强大能力，不需要也不应该被强行限制在孤立的部门“烟囱”里。
以 Rebecca Hinds 与 Stanford 教授 Melissa Valentine 深入研究的某大型数字零售企业为例，该公司在推行服装选品（Assortment planning）AI 算法时，传统的男装采购部只关注男装部门的局部最优解（Local optimum）。而 AI 模型的逻辑是横跨全公司所有品类，寻找能够最大化全公司整体利润的全局最优解（Global optimum）。这倒逼企业打破传统部门壁垒，赋予组织架构更高的弹性。另一个典型案例是**Moderna（摩德纳）**在今年宣布将其 HR（人力资源）与 IT（信息技术）部门合并，由 Chief People and Digital Technology Officer 统一领导，成功缩短了“技术部署”与“组织变革/人效管理”之间的天然鸿沟。
```
传统部门孤岛 vs AI 全局协同：
【传统组织架构】（降低复杂性 / 局部最优解）
┌─────────────────┐ ┌─────────────────┐
│ 男装采购部门 │ │ 女装采购部门 │
└────────┬────────┘ └────────┬────────┘
│ (各自独立决策) │ (各自独立决策)
▼ ▼
男装部门局部最优 女装部门局部最优
【AI 驱动型架构】（处理复杂性 / 全局最优解）
┌─────────────────────────────────────────┐
│ AI 决策引擎 │
└────────────────────┬────────────────────┘
│ (横向打通数据与流程)
┌─────────────┴─────────────┐
▼ ▼
┌───────────────┐ ┌───────────────┐
│ 男装业务 │ ◄───────► │ 女装业务 │
└───────────────┘ └───────────────┘
│ │
└───────────┬───────────┘
▼
实现全公司整体利润最大化
```
- 传统组织架构诞生于 1896 年前后，核心功能是通过划分部门边界来降低管理复杂性。
- 零售企业的 AI 选品案例证明：AI 的优势在于打破部门孤岛、寻找全公司级别的“全局最优”（Global optimum），而非局限于单一部门的“局部最优”（Local optimum）。
- Moderna 将 HR 部门与 IT 部门合并，由原 CHRO 统管人力与数字技术，将“人类行为变革”与“技术基础设施”深度融合。
- 结尾建议：企业无需一次性推进全部五大转变，但应立即选择其中 1-2 项最关键的转变并全力落实。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) 53% 的员工将 AI 视为普通工具，仅有 9% 的员工将其视为“队友”。 — Rebecca Hinds
> [00:04](https://www.youtube.com/watch?v=f-7B300g604&t=4s) 只要员工“以为”回答来自 AI，无论事实是否如此，都会主观评价该回答“缺乏帮助”。 — Rebecca Hinds
> [05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) 纯粹依靠自上而下的行政变更指令，仅能覆盖大约 30%–35% 的员工。 — Rebecca Hinds
> [05:13](https://www.youtube.com/watch?v=f-7B300g604&t=313s) 当结合并激活组织内部的“AI 影响力者”后，AI 技术的全员采纳率可直接翻倍，提升至 60%–70%。 — Rebecca Hinds
> [08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) Fortune 500 企业员工平均每天进行切换应用/标签页（Toggle）动作达 1,200+ 次。 — Rebecca Hinds
> [08:48](https://www.youtube.com/watch?v=f-7B300g604&t=528s) 任务因切换而被打断后，员工重新恢复深度专注平均需要 23 分钟以上。 — Rebecca Hinds
> [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) 知识工作者日常 85%–90% 的工作时间用于协作，过去 20 到 30 年间协作时长增加了约 50%。 — Rebecca Hinds
> [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) 高绩效组织奖励有效协作的概率是普通企业的 5.5 倍。 — Rebecca Hinds
> [10:35](https://www.youtube.com/watch?v=f-7B300g604&t=635s) 如果大家都派 AI 数字分身参加某个会议，说明该会议根本不该召开，应直接改为异步邮件或项目卡片更新。 — Rebecca Hinds
> [16:13](https://www.youtube.com/watch?v=f-7B300g604&t=973s) 传统组织架构诞生于 1896 年前后，核心功能是通过划分部门边界来降低管理复杂性。 — Rebecca Hinds
> [16:13](https://www.youtube.com/watch?v=f-7B300g604&t=973s) AI 的优势在于打破部门孤岛、寻找全公司级别的“全局最优”（Global optimum），而非局限于单一部门的“局部最优”（Local optimum）。 — Rebecca Hinds
## Mentioned
Rebecca Hinds, Glean, Stanford University, Asana, Work Innovation Lab, Future of Work Lab, Raj Choudhury, Zapier, Wade Foster, Rob Cross, Babson College, Udemy, Learns AI Week, Melissa Valentine, Moderna
## Video description
Rebecca Hinds, Head, Work Transformation Lab, Glean
Abstracts
What separates organizations that achieve real AI transformation from those that don’t? Most are investing in AI, but few are seeing meaningful impact. In this talk, I’ll break down five strategic shifts that set high-performing AI organizations apart—from reshaping employee mindsets to rewiring the org chart for scale. I’ll share research-backed practices that help teams move from isolated pilots to enterprise-wide results. If your AI strategy feels siloed, stalled, or stuck in experimentation mode, this session will give you a practical path forward.
[Source](https://www.youtube.com/watch?v=f-7B300g604)
[0:04] Rebecca Hinds: artificial intelligence
[0:04] Rebecca Hinds: amplified intelligence
[0:04] Rebecca Hinds: augmented intelligence
[0:20] Rebecca Hinds: tool
[0:20] Rebecca Hinds: teammate
[0:20] Rebecca Hinds: What can the technology do for me?
[0:20] Rebecca Hinds: What can I do with the technology?
[5:13] Rebecca Hinds: Learns AI Week
[5:13] Rebecca Hinds: before and after
[8:48] Rebecca Hinds: toggle tax
[8:48] Rebecca Hinds: digital pinball
[10:35] Rebecca Hinds: tragedy of the commons
[10:35] Rebecca Hinds: discursive diversity
[10:35] Rebecca Hinds: devil's advocate
Original transcript
[0:04] I first became fascinated by AI back in
[0:07] 2019. I was kneedeep in the thick of my
[0:11] PhD studying retail companies that had
[0:14] implemented AI.
[0:16] Some had done so successfully using AI
[0:19] to generate real results. But others saw
[0:23] nothing. It was just another piece of
[0:25] technology collecting digital dust.
[0:29] And as I dug deeper, analyzing these
[0:32] companies AI strategies,
[0:34] interviewing dozens and dozens of
[0:36] executives and frontline workers,
[0:39] I started to see a clear pattern.
[0:43] The organizations that had been
[0:45] successful,
[0:46] they didn't just deploy AI,
[0:49] they reframed it.
[0:52] Some went so far as to ban the term
[0:54] artificial intelligence entirely.
[0:57] To them, artificial felt threatening. By
[1:01] its very etmology, it applied something
[1:03] at odds with human intelligence,
[1:05] something to be resisted rather than
[1:08] embraced.
[1:10] And so, some of them renamed it. Some
[1:12] called it amplified intelligence, others
[1:15] called it augmented intelligence. Even
[1:18] when they did use that phrase artificial
[1:20] intelligence, they spoke about the
[1:22] technology and importantly its
[1:25] relationship to us humans in distinct
[1:28] ways.
[1:30] I've spent the past six years studying
[1:32] how AI is adopted, implemented, deployed
[1:36] in organizations. First as part of my
[1:38] PhD, then at ASA where I led their work
[1:41] innovation lab and now at Glean where I
[1:43] lead thought leadership along with their
[1:45] new future of work lab. Here's what I've
[1:48] learned. The most successful
[1:51] organizations, they approach AI in
[1:53] starkly different ways than the least
[1:55] effective companies. And today, I'm
[1:58] going to walk through five critical
[2:00] shifts that these successful companies
[2:02] make to get real results with AI. What
[2:06] are their secrets?
[2:09] Let's start with the first shift. The
[2:12] most successful organizations, they
[2:14] don't start with features,
[2:16] functionality, flashy demos. They start
[2:20] like those retail companies did by
[2:23] refraraming how their employees think
[2:26] about AI.
[2:28] Now, there's some really fascinating
[2:30] psychology here at play. This is
[2:32] research we led at the work innovation
[2:34] lab. Right now, most workers see AI as a
[2:39] tool. If you ask workers, what would you
[2:41] compare AI to? Most people, 53% say
[2:45] fundamentally it's a tool.
[2:48] There's a small group, a group of just
[2:50] 9% of employees that see AI differently.
[2:54] They say AI isn't just a tool to me,
[2:57] it's a teammate.
[2:59] And what's really fascinating is this
[3:01] group of 9% of employees, they're
[3:04] significantly more likely to report
[3:06] productivity gains with the technology.
[3:10] Now, why is that the case?
[3:13] When workers see AI as a tool, they tend
[3:17] to think about it very transactionally.
[3:19] It's like a hammer or a calculator. It's
[3:23] something you pick up, you use, and then
[3:25] you put down.
[3:27] But when workers see it as a teammate,
[3:29] they're less likely to ask, "What can
[3:31] the techn technology do for me?" And
[3:34] they're more likely to ask, "What can I
[3:36] do with the technology?" They're not
[3:39] giving up after the first prompt.
[3:41] They're pushing the technology to think
[3:43] act deeper just like they would a human
[3:46] teammate.
[3:48] Mindset is everything because even if a
[3:51] technology is objectively correct, even
[3:54] if it objectively surpasses a human's
[3:57] intelligence, people will resist it or
[4:00] discount its outputs if it doesn't feel
[4:04] human.
[4:06] That became clear in a study led by Raj
[4:08] Chattery at Harvard. Raj and his team
[4:11] trained an AI bot on years and years of
[4:15] communication from Wade Foster, the CEO
[4:18] of Zapier. Meetings, email, transcripts,
[4:22] documents.
[4:24] Employees could only distinguish between
[4:26] the real human CEO and the bot in 58% of
[4:31] cases.
[4:32] Not all that much better than a coin
[4:34] flip.
[4:36] But here's the most important finding.
[4:39] When workers thought the response was
[4:42] coming from the bot, regardless of
[4:44] whether it actually was, they rated the
[4:46] response as less helpful.
[4:51] Even if AI is objectively giving us, our
[4:54] employees what we need, even if it's
[4:57] objectively correct, people are likely
[5:00] to resist it. They're likely to discount
[5:02] it if it doesn't feel human, if it
[5:04] doesn't feel like that teammate.
[5:07] So that's the first shift. AI is a
[5:10] mindset. It's not just a tool.
[5:13] Shift two,
[5:16] we're all seeing it. So many
[5:17] organizations are trying to drive AI
[5:20] adoption through top-down mandates,
[5:23] memos telling their employees to use AI
[5:25] or else.
[5:27] The most effective organizations, they
[5:30] know better. They know that to spur any
[5:32] meaningful change, you need both a
[5:34] combination of top-down direction, those
[5:37] mandates and memos, and bottom-up
[5:40] change.
[5:42] Employees aren't going to adopt this
[5:43] technology, at least not in earnest,
[5:46] just because the big bosses at the top
[5:48] of the organizational totem pool told
[5:50] them to. They're going to adopt the
[5:53] technology because people who they trust
[5:55] and people who they see themselves in
[5:58] tell them why change matters.
[6:01] And in every organization there are this
[6:03] these these people this group of people
[6:07] who have disproportionately high impact
[6:10] on whether others their peers adopt the
[6:12] technology. And what's fascinating is
[6:15] often these people aren't who we think
[6:17] they are. Often they're not highly
[6:20] highly technical. Often they're not at
[6:23] the very top of the org chart.
[6:26] And so finding these people, sometimes
[6:28] called AI influencers,
[6:30] is both important and very, very
[6:34] difficult. One of my favorite strategies
[6:37] comes from Udemy. A couple months ago,
[6:40] Udemy launched what it calls its learns
[6:42] AI week. Essentially, it was an
[6:45] organizationalwide promptathon.
[6:48] Now, there are two really important
[6:50] characteristics of what they did here.
[6:52] one, they used the promptathon, this
[6:55] initiative as a selection mechanism to
[6:58] find these AI influencers.
[7:01] They looked at who signed up, who leaned
[7:04] in, who left excited and wanting more.
[7:08] And second, they didn't just incentivize
[7:12] the highest impact output, the highest
[7:15] impact prompt.
[7:17] They incentivized improvement, best
[7:19] before and after prompt. And and this
[7:22] one is really important,
[7:25] they emphasized feedback, co-creation,
[7:28] collaboration around building these
[7:30] prompts and AI use cases.
[7:34] What we know is activating these
[7:37] influencers is incredibly important. And
[7:39] let's dive just a little bit deeper into
[7:42] why. This research comes from my good
[7:44] friend Rob Cross at Babson College. What
[7:47] he's found is top-down change efforts,
[7:51] they tend to be successful at reaching
[7:54] about 30 to 35% of employees.
[7:58] But when you activate these influencers
[8:01] alongside people in formal roles, it's
[8:04] nearly twice as effective. Reaches 60 to
[8:08] 70% of employees.
[8:13] People don't like to be forced into
[8:16] change. And when they're forced into
[8:18] change, they just go through the
[8:20] motions. Sometimes going through the
[8:22] motions symbolically rather than in
[8:25] earnest.
[8:27] The key to successful AI adoption at
[8:30] scale, wallto-wall, is a combination of
[8:32] these top-down mandates, memos, and
[8:35] activating these influencers. because
[8:37] while they might not have big titles,
[8:39] they have big big influence in
[8:41] encouraging their peers to adopt the
[8:43] technology.
[8:45] Shift two.
[8:48] So, we've talked about the psychology of
[8:50] AI, why mindsets matter, why movements,
[8:54] bottom-up influence beat mandates.
[8:58] But in so many cases, even if you have
[9:00] the right mindset, even if you're
[9:02] activating bottom-up change, there's one
[9:05] critical blocker in so many
[9:06] organizations, and that is what's
[9:08] sometimes referred to as the toggle tax.
[9:13] Already in Fortune 500 companies,
[9:16] employees can toggle between app to app
[9:19] to app 1,200 times or more. And we know
[9:23] from research that as they toggle, if
[9:27] they become distracted, which they often
[9:29] do, it can take them more than 23
[9:32] minutes to return back to the original
[9:35] task.
[9:37] It starts to feel not so much like work
[9:40] and more like digital pinball.
[9:44] And with AI, it can become even more
[9:46] exhausting, even more challenging. This
[9:49] is research we conducted at Glean last
[9:51] year. We found that in enterprises
[9:53] already they're using 150 generative AI
[9:56] tools and by the end of this year it's
[9:59] expected to double to more than 350
[10:03] generative AI tools.
[10:06] If employees need to navigate to a new
[10:09] app, a new login, if they need to rack
[10:12] up more of this toggle tax, they're not
[10:15] going to adopt the technology.
[10:18] In so many cases, the problem isn't too
[10:21] little AI. It's too little AI in the
[10:24] right places. It does need to be
[10:26] embedded, integrated, in many cases
[10:29] nearly invisible for employees to adopt
[10:31] the technology.
[10:35] Shift four.
[10:37] Right now, what we see overwhelmingly,
[10:39] and we see this in how people are
[10:42] building AI technologies, we see this in
[10:44] how executives are approaching
[10:46] strategies and investments. They're
[10:49] treating AI like a personal productivity
[10:52] tool. How can I use this technology to
[10:54] get my employees to do more with less?
[10:59] But that's not the biggest opportunity
[11:01] here. In fact, we know that if we
[11:04] overindex on individual productivity, it
[11:07] can lead to something that economists
[11:09] call the tragedy of the commons. You can
[11:12] easily imagine a world where an
[11:14] individual seeking to maximize their own
[11:17] individual productivity uses AI to
[11:20] delegate all their work to other humans.
[11:22] AI they trigger a flood of different
[11:25] notifications, AI generated content.
[11:30] Individually they might look productive
[11:33] but at that team and organizational
[11:35] level productivity can easily plummet.
[11:39] The opportunity here is not at the
[11:41] individual level. That's part of it but
[11:43] it's at the team and organizational
[11:44] level and it's how we collaborate.
[11:48] What we know is the most effective
[11:49] organizations
[11:51] for quite some time now they've measured
[11:54] and incentivized healthy collaboration.
[11:56] High performing organizations are 5.5
[12:00] times more likely to reward effective
[12:03] collaboration
[12:05] and that's essential. We know that as
[12:07] knowledge workers 85% sometimes 90% of
[12:11] our time is spent collaborating.
[12:15] It's a shocking statistic that's
[12:17] increased by about 50% according to Rob
[12:20] Cross's research over the past two to
[12:22] three decades.
[12:24] Most of our time is spent collaborating.
[12:28] And yet so much of that collaboration is
[12:30] dysfunctional.
[12:33] It's dysfunctional because for so long
[12:36] collaboration has been invisible,
[12:39] right? It happens through so many
[12:40] different people, places, processes,
[12:43] technologies.
[12:44] It's very hard to optimize something we
[12:47] can't fully see or measure.
[12:50] But AI changes things. And in
[12:53] particular, if we use it responsibly and
[12:56] intentionally, it can look across all
[12:58] these different collaboration tools and
[13:00] platforms and help us understand whether
[13:04] that collaboration is effective and how
[13:07] we can make it more effective.
[13:10] In one of my favorite studies,
[13:12] researchers measured what they called
[13:14] discursive diversity.
[13:17] essentially the extent to which people
[13:19] were using different or similar phrases
[13:22] and words as they were collaborating.
[13:26] What they found is the highest
[13:28] performing teams had high levels of
[13:31] discursive diversity when they needed to
[13:33] be creative.
[13:35] But then when they needed to execute
[13:37] work, when they needed to coordinate,
[13:39] move work forward, their discursive
[13:42] diversity dropped. They needed to
[13:45] converge, align, move forward. And so
[13:48] their word choice aligned to reflect
[13:51] that.
[13:53] It's not going to be long before AI
[13:55] joins our meetings and nudges us to
[13:57] increase our discursive diversity when
[13:59] we need to be creative. Probably not
[14:01] using those words, or decrease it when
[14:04] we need to coordinate and move work
[14:06] forward.
[14:09] Now, speaking of meetings, if you want a
[14:11] gut check to understand how effectively
[14:14] your organization is prioritizing using
[14:17] AI to improve collaboration as opposed
[14:19] to individual productivity, look at your
[14:22] meetings.
[14:24] Your employees are probably using AI for
[14:27] notetaking in meetings. They're probably
[14:28] using AI to transcribe meetings. It's a
[14:32] start, but if that's where it ends,
[14:35] you're missing out on the real
[14:37] opportunities. The biggest opportunities
[14:39] we have when it comes to AI in meetings
[14:41] is having meetings part is having AI
[14:44] participate in meetings. In some
[14:47] organizations I've worked with, we've
[14:49] used AI as a meeting
[14:52] moderator to flag toxic behavior in the
[14:56] meeting to nudge people when they start
[14:58] to bulldoze each other's ideas.
[15:02] AI can also participate as a devil's
[15:04] advocate.
[15:06] And here's what's fascinating. Research
[15:08] shows that just having an objective AI
[15:12] in a meeting say something like, "Have
[15:14] you considered other opinions?"
[15:16] That not only dislodges the group think
[15:19] that so often plagues our meetings and
[15:22] brainstorm sessions, but it also drives
[15:25] people to feel more satisfied with the
[15:28] outcome of the meeting.
[15:31] And AI can go even further, showing up
[15:34] on our behalf as a digital twin in our
[15:37] meetings.
[15:39] But we need to be vigilant. We need to
[15:41] be intentional about all of this.
[15:43] Because just because we can send a
[15:45] digital bot, a digital twin to our
[15:48] meetings, doesn't mean we should. If
[15:50] everyone's sending digital bots to
[15:52] meetings, that meeting probably
[15:53] shouldn't happen. It should probably be
[15:55] an email, an update, and a project
[15:57] tracker. Because the real opportunity
[15:59] here when we think about the
[16:01] collaborative layer is not just helping
[16:03] us fix broken collaboration. It's
[16:06] helping us understand whether we need to
[16:08] collaborate in the first place.
[16:13] The final shift.
[16:16] It's nearly impossible to transform your
[16:18] work with AI if your org chart is built
[16:23] for a different era.
[16:26] This is an org chart from back in 1896,
[16:29] about 130 years ago. And if you're
[16:32] honest, this org chart might not look
[16:34] all that different from your org chart
[16:36] today.
[16:38] Now, to understand how our org charts
[16:40] need to change for the AI era, it's
[16:43] important to first understand what's the
[16:45] purpose of org charts. And if we look at
[16:47] the history of organizational behavior
[16:49] and design and theory, we know that
[16:52] fundamentally the purpose of org charts
[16:54] is to reduce complexity.
[16:57] Org charts help us understand our span
[17:00] of control, our responsibilities, who to
[17:03] escalate work to, who to delegate work
[17:05] to,
[17:06] segments decisions. It helps us reduce
[17:09] complexity.
[17:11] But AI can handle this complexity in a
[17:15] way that org charts can't.
[17:18] A couple years ago, I worked with
[17:19] Stanford professor Melissa Valentine to
[17:22] study a large digital retail company.
[17:24] And we studied this retail company as
[17:26] they were rolling out algorithms and AI
[17:29] to nearly every pocket of their
[17:31] organization.
[17:32] And we studied one specific use case in
[17:35] depth, assortment planning. essentially
[17:38] how the organization was deciding how to
[17:41] stock and sell assortments of clothing
[17:44] to customers.
[17:46] This is what their org chart looked
[17:47] like. And in many ways, it was designed
[17:50] to reduce complexity. The men's buying
[17:53] department only needed to make decisions
[17:56] related to the men's buying department.
[17:59] But then came AI. And AI didn't want to
[18:03] stay in these silos. It didn't just want
[18:06] to optimize what was best for the men's
[18:08] clothing department. It wanted to look
[18:10] across the entire organization and
[18:13] optimize the assortments of clothing
[18:15] that would maximize the organization's
[18:17] profitability. It didn't just want to
[18:20] find a local optimum. It wanted to find
[18:22] a global one. And so they recognized
[18:25] that their org chart would need to
[18:26] become more flexible
[18:29] to let AI look across the organization
[18:32] and harness its true potential.
[18:34] And more and more we're seeing
[18:36] organizations grapple with how they
[18:38] should evolve their org charts and make
[18:41] them more flexible to capitalize on AI.
[18:44] Earlier this year, Maderna made
[18:46] headlines when it merged its IT and HR
[18:50] departments. Historically,
[18:53] these worlds are so separate. But what
[18:55] we know is for AI to work at scale, you
[19:00] need a combination of both types of
[19:01] mindsets. You need the deeply human
[19:03] mindset that understands change
[19:06] management adoption, how to manage a
[19:09] digital workforce,
[19:11] and you need the technical mindset to
[19:14] understand how to develop this
[19:16] technology in a way that's secure and
[19:18] scales.
[19:20] And so you can see their chief people
[19:22] and digital technology officer
[19:24] previously the CHRO
[19:27] recognized that this shift needs to
[19:28] happen not just for consolidation but to
[19:31] deliberately close the gap between these
[19:34] two worlds.
[19:37] So think about does your org chart look
[19:39] the exact same it did preAI as it did
[19:43] right now as it does right now.
[19:47] Now, there's a lot of excitement around
[19:48] AI, especially here this week. I'm sure
[19:51] in this room, everyone has some level of
[19:53] excitement and optimism around AI. But
[19:56] if you're like many organizations, you
[19:58] probably haven't seen transformative
[20:01] benefit across your organizations.
[20:03] You've probably seen transformative
[20:05] benefits just in pockets.
[20:08] And it probably comes down to in terms
[20:11] of that gap between your ambition and
[20:14] execution, it probably comes down to at
[20:16] least one, perhaps all five of these
[20:19] five shifts. So I encourage you as you
[20:21] leave today, think about these five
[20:23] shifts. Think about how you can help
[20:25] your organization navigate them, make
[20:29] these shifts. You don't need to choose
[20:31] all five. In fact, you probably
[20:33] shouldn't. But pick one, pick two, and
[20:36] commit to it. commit to it right now.
[20:38] Thank you so much.
[20:40] [Applause]
