Polyrific Launches Catalyst Studio to Cut Enterprise AI Sabotage
Ai4
Enterprise AI rollouts fail because workers sabotage tools imposed by IT, requiring self-service platforms that give front-line employees design control over internal automation.
## Background
Published on YouTube by the Ai4 channel on August 16, 2026, this presentation features Matt Cashatt, CEO of Polyrific. According to the page description, the session introduces the Enterprise Solutions Engine—a self-service environment where business teams can safely design and deploy AI applications connected to corporate data without sacrificing security or compliance.
## TL;DR
Matt Cashatt addresses the cultural and psychological divide in enterprise AI, highlighting widespread workforce anxiety, sabotage, and top-down executive pressure. He argues that AI should augment rather than replace human intelligence, showcasing the immense processing power and efficiency of the human brain. To solve this, Polyrific introduces the Enterprise Solutions Engine and Catalyst Studio, an architectural paradigm that empowers non-technical subject matter experts to build governed, custom AI applications while maintaining security and human agency.
## Atmosphere & style
The presentation carries a reflective, candid, and urgent tone that balances sobering workplace statistics with a message of optimism. Cashatt speaks with humility, openly acknowledging past industry missteps and shifting from a rigid technical pitch to an empathetic discussion on human resistance. The session feels like an engaging, thought-provoking keynote that closes on a collaborative note during audience Q&A, reframing AI adoption as a human-centered movement.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) The fundamental enterprise AI challenge is a growing cultural and psychological divide between technical AI builders and frontline non-technical workers.
- [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) The narrative that AI is destined to replace human intelligence is a flawed story actively being spread and nurtured by technologists themselves.
- [02:10](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=130s) Worker resistance to AI is rising significantly, demonstrated by 142 AI protest events across 42 US states and high employee anxiety regarding job security.
- [02:10](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=130s) Nearly a third of employees (and 44% of Gen Z workers) admit to actively sabotaging internal AI rollouts, while 60% of executives plan to fire non-users.
- [02:10](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=130s) Employees exhibit a paradox by relying on AI for daily support—preferring AI advice over human managers—and using shadow AI tools while resisting top-down company rollouts.
- [02:10](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=130s) Frontline workers reject AI deployments not because of the technology itself, but because they were never consulted about how it changes their roles or uses their data.
- [05:05](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=305s) A landmark Google/Harvard study mapped 1 mm³ of human cortex (requiring 1.4 petabytes and 150M+ synapses) operating on just 20W, illustrating the unmatched complexity of human intelligence.
- [05:05](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=305s) Most enterprise AI pilots fail at deployment because central IT teams build tools in isolation without involving the end-users who actually execute the work.
- [08:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=502s) Prompt-to-app and code generation tools fail to reach production because enterprise readiness requires governance, security, UX refinement, and domain expert involvement.
- [08:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=502s) Polyrific introduced Catalyst Studio and predicted the emergence of the Enterprise Solutions Engine (ESE) as a core software category alongside CRM and ERP.
- [08:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=502s) The Enterprise Solutions Engine pairs a self-service build environment for domain experts with a governed agentic runtime that handles identity, security, and data access.
- [08:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=502s) Keeping non-technical domain experts in the driver's seat prevents 'AI slop' and shifts AI from a threatening force into an empowering tool for workers.
- [13:43](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=823s) Enterprise environments now feature high agent density, with approximately 75 synthetic AI agents operating for every single human employee.
- [13:43](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=823s) Giving frontline workers direct control over AI application design restores individual agency and ensures synthetic agents operate with explicit human authorization.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) Resolving the Human and Social Friction in Enterprise AI
Matt Cashatt, CEO of Polyrific, opens his presentation at the Ai4 conference by addressing the preceding panel discussion. He notes that Polyrific has engineered a platform designed to solve every one of the enterprise AI challenges highlighted by the previous speakers. While his talk was scheduled to focus purely on technical execution and "when your business builds its own AI," Cashatt explains that he made last-minute adjustments to his deck to address a broader, pressing social issue: the growing cultural and psychological divide between AI developers and the non-technical workforce. He emphasizes that while the presentation touches on difficult topics, it concludes on an optimistic note regarding human capability and collaborative technology.
Cashatt stresses that the transition to agentic AI must intentionally include every stakeholder, especially employees, family members, and community neighbors who currently fear AI automation. Rather than allowing tech leaders to impose decisions unilaterally, organizations must provide every worker with a seat at the table. He argues that the mainstream narrative—that artificial intelligence is destined to replace human intelligence—is a fundamentally flawed story that technologists themselves are inadvertently nurturing and spreading.
```
+-----------------------------------------------------------------+
| THE ENTERPRISE AI DIVIDE |
| |
| TECHNOLOGY LEADERS FRONT-LINE WORKERS |
| - High optimism & enthusiasm - Fear of job displacement|
| - Unilateral rollout decisions - Excluded from design |
| - Focus on speed & automation - Active/passive pushback |
+-----------------------------------------------------------------+
```
- **Optimism Gap:** A room poll shows strong optimism among AI conference attendees, but Cashatt warns that AI is currently not benefiting everyone equally, exposing society to darker outcomes if unaddressed.
- **Crossroads in Las Vegas:** While AI developers enjoy conferences in Vegas and celebrate complex builds, many of their colleagues back at the office are quietly terrified of what those applications mean for their livelihoods.
---
[02:10](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=130s) The Resistance Reality: Protests, Sabotage, and Workplace Anxiety
Backing up to examine ground-truth data, Cashatt presents stark metrics demonstrating widespread worker anxiety and pushback against enterprise AI deployments. In July, Reuters documented 142 distinct AI protest events across 42 US states. Public sentiment surveys confirm that over half of American workers fear the technology's impact on their careers. Furthermore, workplace studies reveal an alarming trend: nearly a third of employees admit to actively sabotaging internal AI rollouts, with Gen Z workers showing the highest rates of resistance. Conversely, executive management is taking an increasingly coercive stance, with 60% of executives stating they plan to terminate employees who refuse to adopt AI tools—a policy dynamic Cashatt likens to harsh 19th-century labor disputes.
```
+-----------------------------------------------------------------+
| WORKFORCE SENTIMENT METRICS |
| |
| 52% Workers worried about job impact (Pew Research) |
| 29% Workers admit sabotaging internal AI rollouts (Writer) |
| 44% Sabotage rate among Gen Z participants (Writer) |
| 60% Executives planning layoffs for non-users (Writer) |
+-----------------------------------------------------------------+
```
Despite this overt pushback, worker behavior exhibits a profound paradox: employees simultaneously turn to AI for daily support. A overwhelming majority of workers report that they would rather seek professional guidance from an AI model than from their human manager, highlighting systemic leadership deficiencies. Additionally, 90% of employees regularly leverage personal, unauthorized AI tools for work tasks, even though only 40% of enterprises officially provide approved AI resources. Cashatt points out the irony: "They're marching against AI in the street and then they're leaning on it when they go back home."
- **Lack of Consultation:** The root cause of backlash is that workers were never consulted about how AI alters their day-to-day duties, how their personal work outputs train the underlying models, or how local infrastructure (data centers and power plants) impacts their communities.
- **The Core Issue:** Workers are not rejecting the technological capabilities of AI; they are rejecting the terms under which it is being forced upon them without their consent or input.
---
[05:05](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=305s) Human Intelligence vs. Artificial Neural Networks
Reframing the problem, Cashatt asserts that true enterprise value lies in pairing human intelligence with artificial execution rather than aiming for human replacement. He shares a personal reflection on the unmatched complexity of the biological human brain, referencing a landmark 2024 study by Google and Harvard. Researchers mapped a single cubic millimeter of the human cortex—a minuscule slice of tissue that required 1.4 petabytes of data to render, encompassing tens of millions of cells and over 150 million synaptic connections. Remarkable above all, the entire human brain operates on just 20 watts of power, equivalent to a dim household light bulb.
```
+-----------------------------------------------------------------+
| GOOGLE/HARVARD CORTEX MAP DATA |
| |
| Sample Size: 1 cubic millimeter of human cortex |
| Data Required: 1.4 Petabytes |
| Cell Count: Tens of millions of cells |
| Synapses: >150,000,000 synaptic connections |
| Power Consumption: ~20 Watts (Entire brain) |
+-----------------------------------------------------------------+
```
Cashatt makes a candid confession about his own past messaging when presenting to business owners in New York. He recalls telling audiences that if they didn't "get on the train," they would be left behind. He realized after receiving a cold response that these business leaders and workers "never asked to be on a train or go anywhere in the first place." Diffusing this tension requires bringing the controls of AI directly to the end-users who execute the work, letting them dictate how and where automation is applied.
- **Failing Enterprise Pilots:** The majority of enterprise AI pilots fail at the deployment stage specifically because central IT teams build solutions in isolation without involving the end-users in the initial design process.
- **Centralized IT Bottleneck:** Traditional AI stacks rely on a single central development queue and generic tools bolted onto legacy workflows, creating severe adoption bottlenecks and zero ROI.
---
[08:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=502s) Introducing the Enterprise Solutions Engine (ESE) and Catalyst Studio
Turning to developer paradigms, Cashatt discusses prompt-to-app and code generation tools like Codex, Claude Code, Lovable, and Bolt.new (misheard in transcript as "base 44"). While these tools excel at generating rapid mockups and functional concepts, they fall far short of shipping production-ready enterprise applications. Real-world deployment requires UX refinement, rigorous QA, security reviews, domain research, user acceptance testing (UAT), and deep integration with subject matter experts (SMEs). To bridge this gap between concept and enterprise governance, Polyrific developed a governed agentic runtime alongside a new product tier currently in their lab: **Catalyst Studio**.
Cashatt predicts the emergence of a new software category: the **Enterprise Solutions Engine (ESE)**. Just as enterprise software history established dominant anchors like Salesforce for CRM and SAP/Oracle for ERP, the ESE category will become a standard enterprise pillar. By placing a secure, governed runtime underneath user-facing build studios, organizations solve identity and security risks—such as agent impersonation—via dedicated service accounts. This allows non-technical business users to rapidly construct and deploy targeted solutions (even single-day applications) without going through IT.
- **AI Thinks Fast, Humans Think Forward:** Cashatt highlights a key distinction: AI delivers speed and broad processing power, but human judgment is required to determine what problems are actually worth solving and which stakeholders to engage.
- **Avoiding "AI Slop":** Keeping human domain experts in the driver's seat prevents the generation of "AI slop"—bloated emails, low-quality software applications, and unhelpful automation.
- **Sustainable AI Adoption:** Involving frontline employees directly in application design transforms AI from a threatening tool wielded *against* workers into an empowering tool leveraged *by* them.
---
[13:43](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=823s) Open Q&A and Audience Reflections on Individual Agency
Cashatt opens the floor to audience questions and comments. Kristen, an attendee working at an AI attestation infrastructure company called Orin based in Boston, Massachusetts, steps up to the microphone. She reinforces Cashatt’s thesis, emphasizing that the original internet ("inter-net") was designed to connect human beings sitting in chairs. Today, however, organizations face an explosion of synthetic agents—noting a statistic heard during the conference that enterprise environments now host roughly 75 AI agents for every single human employee.
```
+-----------------------------------------------------------------+
| ENTERPRISE AGENT DENSITY METRIC |
| |
| 1 Human Employee <---> 75 Synthetic AI Agents |
+-----------------------------------------------------------------+
```
Kristen argues that empowering workers with building tools restores personal agency over their data, identity, demographic information, and APIs. By giving frontline workers control over how these synthetic agents operate on their intellectual property, enterprises flip the incentive model so that automation cannot proceed without explicit human authorization. Cashatt agrees with her observations, concluding the session by urging the audience to visit the Polyrific booth and reminding attendees that AI is a collective movement that must empower every individual it touches.
- **Locked and Shackled IP:** Kristen raises the challenge of managing enterprise data and intellectual property in environments dominated by autonomous, synthetic agents.
- **Closing Remarks:** Cashatt thanks the audience for participating and encourages tech leaders to build inclusive, human-centered AI ecosystems.
## Notable quotes
> [02:00](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=120s) They're marching against AI in the street and then they're leaning on it when they go back home. — Matt Cashatt
> [05:05](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=305s) never asked to be on a train or go anywhere in the first place. — Matt Cashatt
## Mentioned
Polyrific, Matt Cashatt, Ai4, Reuters, Pew Research, Writer, Google, Harvard, Codex, Claude Code, Lovable, Bolt.new, Catalyst Studio, Enterprise Solutions Engine, Salesforce, SAP, Oracle, Kristen, Orin
## Video description
Matt Cashatt - CEO, Polyrific
The next wave of enterprise AI will not be built by engineering alone. As agentic AI matures, business teams can increasingly build their own problem-solving applications, but most platforms force a choice between speed and control. This session introduces a new category, the Enterprise Solutions Engine, designed to resolve that tradeoff: a self-service environment where teams design and ship AI applications connected safely to corporate data, on infrastructure the enterprise governs. Matt Cashatt walks through how this reshapes the build-versus-buy calculus, why the traditional AI platform stack falls short, and what it takes to give business users real building power without sacrificing security, observability, or compliance. Executives will leave with a clear lens for evaluating platforms and a playbook for unlocking AI velocity across the organization.
Polyrific推出Catalyst Studio以减少企业AI遭抵制破坏
企业AI落地之所以失败，是因为员工会抵制IT部门强推的工具，因此需要能让一线员工自主掌控内部自动化设计的自助服务平台。
Matt Cashatt探讨了企业AI领域的文化与心理鸿沟，重点指出了员工普遍存在的焦虑、抵制破坏行为以及管理层自上而下的压力。他认为AI应当增强而非取代人类智能，并阐述了人脑巨大的处理能力与极高效率。为解决这一问题，Polyrific推出了Enterprise Solutions Engine与Catalyst Studio，这一架构范式使非技术领域的业务专家能够构建受治理的定制AI应用，同时兼顾了系统安全与人类自主权。
Polyrific 首席执行官 Matt Cashatt 指出，技术构建者与一线员工之间存在严重的文化裂痕；全美 42 个州已记录 142 起 AI 抗议活动，近三分之一的员工（Z 世代高达 44%）承认曾暗中破坏内部 AI 的推行。 (2:10)
职场呈现出矛盾现象：90% 的员工在日常工作中依赖非官方的 AI 工具，甚至更倾向于向 AI 寻求指导而非管理者，但由于自上而下的部署从未征询员工意见且涉及数据使用，引发了强烈的抵触情绪。 (2:10)
借由哈佛与 Google 对 1 立方毫米人脑皮层（耗费 1.4PB 数据且包含逾 1.5 亿突触，整脑能耗仅 20W）的研究，说明人类智慧的不可替代性；多数企业 AI 试点夭折正是由于中央 IT 团队脱离一线业务孤立研发。 (5:05)
Polyrific 提出“企业解决方案引擎”（Enterprise Solutions Engine, ESE）与 Catalyst Studio，通过受治理的智能体运行时与服务账号，赋能一线非技术专家自主构建企业级安全 AI 应用，弥合原型与生产落地之间的鸿沟。 (8:22)
Reading summary
Matt Cashatt, CEO of Polyrific, argues that enterprise AI deployments routinely fail because centralized IT initiatives exclude and alienate the workforce, asserting that successful adoption requires governed platforms that allow frontline workers to build their own targeted automation.
Workforce Resistance and the Enterprise Adoption Paradox (2:10)
Cashatt points to significant labor friction surrounding enterprise AI. In July, Reuters recorded 142 AI protest events across 42 US states, and Pew Research found that 52% of American workers worry about the technology's career impact. Citing workplace survey data from Writer, Cashatt notes that 29% of employees—and 44% of Gen Z respondents—admit to actively sabotaging internal AI rollouts, while 60% of executives plan terminations for employees who refuse to use AI tools.
Despite this overt resistance, workers actively use AI when they retain autonomy. Data cited indicates that 90% of employees use unauthorized personal AI tools for professional tasks, even though only 40% of enterprises officially supply AI software. Furthermore, a substantial majority of surveyed workers report preferring AI guidance over consulting human managers. Cashatt argues this indicates employees do not reject the technology itself, but rather push back against being excluded from deployment decisions, training data usage, and operational design.
Human Cognitive Strengths Versus AI Replacement (5:05)
Cashatt rejects the premise that artificial intelligence should replace human workers, contrasting biological efficiency with current artificial models. Citing a 2024 study by Google and Harvard, he emphasizes that a single cubic millimeter of the human cortex required 1.4 petabytes to map, containing tens of millions of cells and over 150 million synapses, while the entire human brain operates on just 20 watts of power.
Reflecting on his past messaging that warned business owners they would be 'left behind,' Cashatt argues that top-down pressure alienates workers who never requested automation. He asserts that central IT teams cause enterprise pilots to fail when they build systems in isolation. In his view, artificial intelligence provides computational speed, but human judgment remains necessary to identify relevant problems, engage stakeholders, and avoid the generation of low-quality 'AI slop.'
Governed Runtime and the Enterprise Solutions Engine (8:22)
While acknowledging the rise of code-generation tools like Codex, Claude Code, Lovable, and Bolt.new, Cashatt maintains that rapid mockups cannot meet production standards without security reviews, rigorous testing, and integration with subject matter experts. He forecasts the emergence of a new enterprise software category called the Enterprise Solutions Engine (ESE), which he compares to established pillars like ERP and CRM systems.
To operationalize this model, Polyrific developed Catalyst Studio on top of a governed agentic runtime. Cashatt explains that placing governance underneath frontline build environments resolves compliance and impersonation risks through dedicated service accounts. This architecture allows non-technical business employees to build and deploy single-day, customized applications directly, eliminating central IT bottlenecks.
Restoring Human Agency Amid Agent Proliferation (13:43)
During the session's discussion, audience member Kristen from Orin noted a conference observation that enterprise networks are approaching a density of approximately 75 synthetic AI agents per human employee. She argued that equipping frontline workers with governed build tools restores individual agency over their intellectual property, data, and identity, ensuring automated workflows require explicit human authorization rather than running unchecked.
Cashatt concludes that enterprise automation must transform from a top-down threat into a collaborative, frontline-driven capability. However, the model relies on the assumption that providing non-technical staff with governed build studios will successfully overcome deep-seated workplace anxieties and eliminate employee resistance.
精炼阅读摘要
Polyrific 首席执行官 Matt Cashatt 在 Ai4 大会上指出，企业级人工智能部署遇阻的核心在于自上而下强制推行所导致的技术与劳动力脱节；真正的商业价值不在于用算法取代人力，而在于构建受治理的“企业解决方案引擎”，将 AI 应用开发主导权交还给一线业务专家。
劳工抵触与管理层高压下的应用悖论 (2:10)
Cashatt 引用数据指出，全美企业在推行 AI 时面临显著的劳资冲突与信任危机。路透社记录显示，7 月份全美 42 个州共爆发了 142 起针对 AI 的抗议事件；Pew Research 调查表明 52% 的美国劳工担忧该技术冲击个人职业生涯。根据企业软件公司 Writer 的调研，近三分之一（29%）的员工承认曾主动破坏内部 AI 项目的推进，该比例在 Z 世代群体中高达 44%。然而，60% 的管理层高管表示计划解雇拒绝使用 AI 工具的员工，Cashatt 将这种强制策略比作 19 世纪严酷的劳资纠纷。
与这种公开抵制形成鲜明反差的是，职场中存在极大的应用悖论：高达 90% 的员工经常在日常工作中使用未经官方授权的个人 AI 工具，而企业官方正式批准提供 AI 资源的比例仅为 40%；同时绝大多数员工表示更倾向于向 AI 模型寻求专业指导而非向人类主管请教。Cashatt 强调，员工并非抗拒 AI 的技术能力本身，而是反对在未获得充分咨询、未参与日常职责再设计且未知晓个人产出如何被用于训练模型的情况下被动接受指派。
人类大脑复杂性与中心化IT瓶颈 (5:05)
针对主流叙事中“AI 终将取代人类智能”的论调，Cashatt 认为技术人员本身在不经意间推波助澜了这种错误观念。他援引 Google 与哈佛大学在 2024 年开展的脑科学里程碑研究指出，人类大脑仅 1 立方毫米的大脑皮层样本就需要 1.4 PB（拍字节）的数据进行渲染，包含数千万个细胞与超过 1.5 亿个突触连接，而整个人脑运行功耗仅约 20 瓦特（相当于一枚微弱的家用灯泡）。这种极高能效的生物复杂性是当前人工神经网络无法企及的。
Cashatt 反思了过去对企业主宣称“不上车就会被淘汰”的话术，指出许多企业和劳工本就从未主动要求“登车”。在商业实践中，绝大多数企业 AI 试点项目在部署阶段夭折，其根源在于中央 IT 团队孤立构建方案，导致单一开发排队队列成为严重瓶颈，强行将通用工具套在遗留工作流上且毫无投资回报率。Cashatt 指出，AI 的优势在于快速处理与执行，而人类的独特价值在于前瞻思考、辨别真正值得解决的问题并协调各方利益。
企业解决方案引擎与安全代理运行时 (8:22)
Cashatt 分析指出，虽然 Codex、Claude Code、Lovable 以及 Bolt.new 等提示词生成代码（prompt-to-app）工具能够迅速产出原型，但它们无法满足企业级生产环境所需的 UX 深度打磨、严格质检（QA）、安全性审查、领域调研以及用户验收测试（UAT）。为了弥合原型构想与企业合规治理之间的鸿沟，Polyrific 研发了受治理的代理运行时（governed agentic runtime）以及处于实验室阶段的 Catalyst Studio。
Cashatt 预测软件行业将诞生名为“企业解决方案引擎”（Enterprise Solutions Engine, ESE）的新品类，其地位将比肩 CRM 领域的 Salesforce 或 ERP 领域的 SAP 与 Oracle。该架构通过在底层设立具备服务账户身份、可观测性防线及数据访问权限管制的治理运行时，从技术底层防范智能体冒充等合规风险，同时向上让非技术的业务领域专家（SME）通过 Catalyst Studio 在极短时间内（乃至单日）自主构建目标应用。这种一线主导的模式能够有效避免生成低质冗余的“AI 垃圾”（AI slop），使一线员工从技术防范者转变为工具驾驭者。
智能体激增与一线员工数据主权维护 (13:43)
在大会问答环节中，来自波士顿 AI 认证基础设施公司 Orin 的参会者 Kristen 提出，早期互联网旨在连接物理空间中的人类，而当前企业环境正经历合成智能体（synthetic agents）的爆发式增长。据大会现场引用的数据，企业内部环境中的人机比例已达到大约每 1 名人类员工对应 75 个合成 AI 智能体。
Kristen 与 Cashatt 均认为，在合成智能体高密度的企业环境中，必须通过赋予员工构建工具来重构其对自身数据、身份标识、人口统计信息及 API 资产的主权。赋予一线员工对其智力成果与智能体操作权限的最终控制权，能够彻底转变过往单向推动的自动化逻辑，确保任何智能体自动化任务都必须获得人类明确授权，从而化解企业核心知识产权被算法孤立束缚的治理隐患。
Cashatt 总结强调，企业级 AI 的普及不应是单向的技术灌输，而必须是一场包容一线员工、家庭与社区各方利益相关者的集体协作。若技术领袖无法打破以高压裁员威胁和中心化孤立开发为主导的传统路径，AI 普及将持续遭遇抵制；唯有将安全治理与自助构建工具赋能于一线业务人员，才能建立可持续、以人为本的企业级 AI 生态。
Detailed notes — zh
## Background
该内容由 Ai4 频道于 2026 年 8 月 16 日发布在 YouTube 平台，主讲人为 Polyrific 首席执行官 Matt Cashatt。视频简介指出，企业 AI 的下一波浪潮无法单靠工程团队完成，演讲重点介绍了“企业解决方案引擎（Enterprise Solutions Engine）”这一全新范式，旨在为业务团队提供一个兼顾自主构建速度与底层安全合规治理的自服务开发环境。
## TL;DR
Matt Cashatt 剖析了当前企业 AI 落地中技术乐观派与一线员工之间的认知撕裂，指出排斥和暗中破坏的根源在于一线业务专家（SME）从未被赋予话语权。他强调人类智能（HI）具有不可替代的前瞻性，AI 只是执行层面的助推器，传统自上而下的中心化推进方式注定导致试点失败。为此，Polyrific 提出了由治理型运行时与可视化工作台 Catalyst Studio 构成的“企业解决方案引擎（ESE）”，赋能业务人员直接构建贴合实际工作流的生产级 Agent。整场演讲倡导“AI 负责快速思考，人类负责前瞻思考”，主张通过把构建杠杆交还给一线员工来化解抵触，实现企业 AI 的真正价值兑现。
## Atmosphere & style
演讲脱离了技术发布会常见的狂热浮夸，以坦诚的自我检讨和冷峻的社会现实调研开场，直击职场焦虑与人机对立的痛点。Matt 的表达沉稳而极具感染力，通过人脑能耗对比、“时代列车”反思等生动隐喻，让深刻的架构哲学变得引人入胜。后半程与观众关于数据代理权与人机边界的共鸣互动，更将现场推向充满人文关怀与战略启发的高潮。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) 行业中流传的“AI 将取代人类智能”是极其危险的错误叙事，技术从业者在无意中助长了大众对 AI 的恐惧与排斥。
- [02:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=142s) 基层员工和普通公众对 AI 的抵触并非针对技术本身，而是拒绝未经其知情同意、强加变革且威胁生计的推行条款。
- [02:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=142s) 职场存在严重撕裂：近三成员工承认曾暗中破坏公司的 AI 项目，而多数企业管理层却采用强推和解雇威胁等冰冷手段。
- [05:15](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=315s) 人脑在极低能耗下拥有超高复杂度与认知能力，AI 绝非人类大脑的替代品，而是人类智能的“火箭助推器”。
- [07:44](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=464s) 企业级 AI 项目普遍缺乏 ROI 且采用率低下的根源，在于传统 IT 模式边缘化了一线业务专家（SMEs），脱离了实际业务场景。
- [08:33](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=513s) 流行的生成式或氛围编程（Vibe Coding）工具仅能降低原型搭建门槛，缺乏构建企业生产级系统所需的工程治理、安全测试与业务把关。
- [09:47](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=587s) Polyrific 提出“企业解决方案引擎（ESE）”新品类，旨在成为与 CRM、ERP 并列的基础设施，包含治理型运行时与可视化构建环境。
- [09:47](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=587s) 通过服务账户（Service Account）和受控运行时，可有效解决 Agent 身份冒充、权限失控与安全合规风险，打破 IT 排期瓶颈。
- [11:23](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=683s) 人机协作的核心分工是“AI 负责快速思考，人类负责前瞻思考”，人类必须始终掌舵战略目标与价值判断以杜绝“AI 垃圾”。
- [13:43](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=823s) 在未来可能出现“1 名员工对应 75 个 Agent”的时代，必须保障个体的数据主权与代理权，建立授权方可执行的合规机制。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) 重新审视 AI 的叙事：从技术狂热到社会现实
Matt Cashatt 以对前一场专题讨论的回应开场。他直言 Polyrific 开发的平台能够解决前一组嘉宾提到的每一个痛点。但他并没有按部就班地宣讲纯技术架构，而是在演讲前一刻临时调整了幻灯片，决定将焦点转向一个更加紧迫的社会议题——人们当前对 AI 的普遍恐惧与排斥。他指出，当前行业中流传着一个极其危险的错误叙事：“AI 将取代人类智能（AI will replace human intelligence）”。而这一叙事恰恰是由技术从业者和乐观派们在无意中不断助长和放大的。
```mermaid
flowchart TD
accTitle: AI 实施中的核心矛盾与解决路径
subgraph Disconnect["当前企业 AI 推进中的断层"]
A["技术构建者与管理层<br/>（单向推行 / 盲目乐观）"] -->|"强加变革 / 忽视一线"| B["一线员工与业务专家 (SMEs)<br/>（恐惧、抵触甚至暗中破坏）"]
B -->|"缺乏端到端工程与业务整合"| C["AI 试点项目失败<br/>（无 ROI、无实际采用率）"]
end
subgraph Solution["企业解决方案引擎模式"]
D["治理型智能体运行时<br/>(Governed Agentic Runtime)"] --- E["业务赋能工具 Catalyst Studio"]
E -->|"赋予一线设计与部署权"| F["生产级企业 AI 应用<br/>（安全合规、贴合实际工作流）"]
end
```
在场的开发者与管理者大多对 AI 怀抱巨大热情，将其视为职业生涯乃至一生中最具变革性的技术力量。然而 Cashatt 提醒大家，冷酷的现实是：AI 目前并未让所有人受益。如果这项技术不能让广大基层工作者感受到益处，公众对 AI 阴暗面的恐惧与想象就会演变成现实，最终反噬整个行业。要扭转这一局面，必须从构建者自身做起，将构建 AI 的权力下放给那些真正在一线干活的人，让那些感到恐惧的同事、朋友和利益相关者在决策桌前拥有真正的席位。
---
### [02:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=142s) 撕裂的现状：街头抗议、职场抵触与私下依赖的矛盾
Cashatt 坦言，自己依靠构建 AI 驱动的智能体解决方案（Agentic Solutions）为生，公司也开发了专门管理和治理智能体的平台。但在拉斯维加斯的技术盛会之外，现实世界的办公室和街头却在上演完全不同的景象。数据揭示了技术乐观派与普通劳动者之间巨大的鸿沟：
- **街头抗议与广泛担忧**：路透社（Reuters）在某一个周六统计到全美 42 个州爆发了 142 场针对 AI 的抗议游行；皮尤研究中心（Pew Research）的数据显示，52% 的美国劳动者对 AI 对其工作的影响感到焦虑。
- **暗中抵制与管理层高压**：Writer 与 Workplace Intelligence 的联合调研显示，29% 的员工承认曾暗中破坏过自家公司的 AI 推广项目，而在 Z 世代（Gen Z）受访者中，这一比例高达 44%；与此同时，60% 的高管竟表示计划解雇那些拒绝使用 AI 的员工——这一数据充斥着“1892年劳资关系委员会”般的原始与冰冷。
- **管理失灵与私人依赖**：令人讽刺的是，67% 的员工表示他们宁愿向 AI 寻求指导，也不愿向自己的直属经理请教；大约 90% 的员工在日常生活中频繁使用个人 AI 工具，但只有 40% 的企业为员工主动配备了官方 AI 工具。
这种既在街头抗议 AI、回家后又离不开 AI 的矛盾现象，根源在于“没有人征求过他们的意见”。员工们从未被询问过 AI 该如何改变他们的岗位，甚至在不知情的情况下，自己的工作成果被拿去训练模型，反过来威胁自身的生计；社区居民也未曾同意在自家后院建设耗费巨量水电的数据中心与配套发电厂。Cashatt 强调，大众拒绝的不是技术本身，而是推行技术的“霸王条款（The terms）”。
---
### [05:15](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=315s) 人类智能（HI）的不可替代性与“上车论”的反思
为了消除这种恐惧，Cashatt 强调必须重新认识人类智能（HI, Human Intelligence）与人工智能（AI）的本质区别。他坚信人类大脑是宇宙中最壮丽的造物。2024 年哈佛大学与 Google 联合绘制的人类大脑皮层高精度地图显示，仅仅 1 立方毫米的脑组织就包含数千万个细胞、超过 1.5 亿个突触连接，产生的数据量高达 1.4 PB；而整个人脑的运行功率仅仅约为 20 瓦，相当于一颗普通灯泡。
在算力与能耗的巨大对比面前，AI 绝不是人类大脑的替代品，而是人类智能的“火箭助推器”。然而，技术圈过去兜售 AI 的方式却存在严重缺陷。Cashatt 在此做出了自我检讨：他曾在一场面向纽约企业主的演讲中宣称“如果不登上这趟火车，你就会被时代抛下”，结果台下一片死寂，毫无共鸣。
> "他们根本一开始就没要求过要上火车，也不想去任何地方。"
技术人员自以为合理的“时代列车”隐喻，对于只想安心做好本职工作的人来说是一种强加的威胁。一线人员的思维并未对技术关闭，他们每天都在接触 AI，但他们理应获得发言权。最好的赋权方式，就是将 AI 的操控杠杆直接交到实际做业务的人手中。
---
### [07:44](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=464s) 传统 AI 架构的失效：为何缺乏一线参与的试点注定折戟
从商业落地角度来看，将终端用户排除在外的做法正是导致绝大多数企业级 AI 项目流产的罪魁祸首。当前无数企业都在抱怨 AI 项目缺乏 ROI（投资回报率）、实际采用率低下，其根本症结在于传统 AI 实施架构存在严重的系统性缺陷。
| 架构维度 | 传统企业 AI 推进模式（导致失败） | 业务驱动型构建模式（理想状态） |
|:--- |:--- |:--- |
| **主导团队** | 单一中心化 IT/AI 工程团队排队响应需求 | 业务部门与一线主题专家（SME）自主探索与主导 |
| **系统整合** | 将通用 AI 工具生硬嫁接到老旧工作流上 | 针对具体业务问题定制的自适应智能体工作流 |
| **需求输入** | 业务专家（SME）被边缘化，缺乏真实场景反馈 | 业务专家全程参与设计、验证与持续调优 |
| **落地效果** | 员工产生抵触甚至破坏，项目停留于 Demo | 贴合实际作业习惯，实现高采用率与明确商业价值 |
由于设计过程中没有真正理解业务逻辑的业务专家参与，自上而下硬塞给业务部门的通用 AI 方案往往脱离实际业务场景，最终在从原型迈向生产的鸿沟中彻底搁浅。
---
### [08:33](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=513s) 从原型玩具到企业级生产：Catalyst Studio 与解决方案引擎的诞生
Cashatt 随后探讨了当前流行的生成式开发工具（如 Codex、Cloud Code、Lovable、Base44 等“氛围编程/Vibe Coding”工具）。虽然这些工具极大降低了前端界面和原型 Demo 的搭建门槛，但在构建严肃的企业级生产系统时，它们暴露出巨大短板：
- **生产级要素的缺失**：仅靠模型生成的代码并不等于可上线应用，中间缺少了深入的领域研究、严谨的 UX 设计、全面的 QA 自动化测试、底层基础设施编排、代码安全审查以及业务验收测试（UAT）。
- **业务专家的缺席**：最重要的核心盲区在于，缺乏真正懂得该业务逻辑该如何运作的业务专家（SME）把关。
针对这一断层，Polyrific 提出了一套全新的产品范式——**企业解决方案引擎（Enterprise Solutions Engine）**。它包含两大核心支柱：底层是一个兼顾安全合规与调度的**治理型智能体运行时（Governed Agentic Runtime）**；上层则是其研发实验室推出的可视化构建环境 **Catalyst Studio**。该体系旨在打破工程与业务的壁垒，让最懂业务逻辑的一线人员无需从零编写底层工程代码，就能在企业受控与合规的环境中，直接构建安全连接企业数据的生产级 AI 应用。
### [09:47](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=587s) 企业解决方案引擎（ESE）与可治理的 Agent 运行环境
Matt Cashatt 提出，企业级软件市场即将诞生一个全新的品类——企业解决方案引擎（Enterprise Solutions Engine，简称 ESE），它将像当年的 ERP 和 CRM 一样，成为企业不可或缺的基础设施。为了说明品类认知的建立，他与现场观众进行了互动：提到 CRM，大家脱口而出的是 Salesforce；提到 ERP，大家立刻联想到 SAP 和 Oracle。Polyrific 希望未来当人们提及 ESE 时也能产生同样的条件反射。Matt 强调，目前许多团队尝试使用各类云端代码工具构建业务方案，但往往缺少关键的一环——一个能够确保 AI Agent 受到严格管控的可信运行时（Governed Runtime）。针对前一场圆桌讨论中有人担忧“无法判断 Agent 是否在冒充真实用户”的问题，Matt 直言只要 Agent 拥有合法的服务账户（Service Account）并运行在受治理的环境中，这类技术与安全风险早已迎刃而解。
一旦底层治理的基础设施就绪，企业便可以提供给员工低门槛的 Studio（工作台），让业务人员能够放心地在受控的运行时上自主搭建解决方案。这种模式从根本上改变了传统的运作方式：企业不再需要把 AI 当作耗时冗长的 IT 项目来立项，而是能够让那些原本对 AI 感到恐惧和排斥的业务人员直接参与到设计过程中。更重要的是，IT 和管理层可以彻底“让出道路”，让一线员工直接使用 Studio 构建解决眼前问题的应用——哪怕这个应用生命周期只有一天。这正是 Agent 时代的真正价值所在：赋予组织一种在响应速度与覆盖广度上均表现卓越的新型智能。
```mermaid
flowchart TD
accTitle: 企业解决方案引擎 (ESE) 架构模型
subgraph Governance["底座治理层 (Governed Runtime)"]
A["服务账户与身份认证 (Service Account)"]
B["安全合规与权限管控"]
end
subgraph Studio["自主构建层 (ESE Studio)"]
C["业务用户自主编排"]
D["敏捷/即用即弃型 AI 应用"]
end
subgraph Value["业务成效"]
E["摆脱传统 IT 项目排期沉疴"]
F["赋能一线人员，消除抵触情绪"]
end
Governance --> Studio
Studio --> Value
```
- **全新软件品类**：企业解决方案引擎（ESE）将与 CRM（如 Salesforce）、ERP（如 SAP、Oracle）并列为企业核心支柱。
- **运行时治理**：通过服务账户（Service Accounts）与安全运行时，彻底解决 Agent 身份冒充与权限失控风险。
- **去 IT 化瓶颈**：业务人员可在 Studio 中快速生成只使用一天的临时性 Agent 解决方案，无需经历繁琐的 IT 立项与交付流程。
---
### [11:23](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=683s) “AI 负责快速思考，人类负责前瞻思考”：让人类始终掌舵
Matt 分享了 Polyrific 的核心理念：“AI thinks fast and humans think forward”（AI 负责快速思考，人类负责前瞻思考）。他半开玩笑地坦白，这句令他颇为自豪的金句可能还是他在用 Claude 打磨文案时碰撞出来的，但它精准地抓住了人机协作的本质。AI 的核心优势在于速度与广度，能够在数秒内完成一整个团队原本需要耗费一周的繁重工作；然而，AI 永远无法替代人类决定“究竟什么事情才值得去做”、下一个目标该瞄准何方，或者谈判桌对面的哪个人应该被纳入决策流程。只有让业务利益相关者始终坐在“驾驶座”上，才能真正区分出有价值的业务应用与市面上泛滥的“AI 垃圾”（AI slop）——比如那些原本一句话就能说清、却被 AI 扩写得冗长拖沓的无效邮件和粗制滥造的软件。
在 Matt 看来，Agent 时代无疑将属于那些无所畏惧的先行者，但技术向前推进并不意味着人类可以放弃领导权。AI 是否能成为一股造福企业的力量，完全取决于我们选择带谁一起坐到谈判桌前。他坦言，在行业大会的演讲台上，讲者通常都在推销自己的产品，他虽然也介绍了产品，但更希望大家带走的核心认知是：必须把那些对 AI 充满恐惧的员工拉入共创过程。唯有消除不理解与不信任，才能避免项目在内部遭到抵制甚至暗中破坏，让整个行业更加平稳、可持续地兑现 AI 多年以来的美好承诺。
- **核心价值分工**：AI 胜在执行效率与数据覆盖广度（Think fast），人类胜在战略定力、目标选择与人际协同（Think forward）。
- **杜绝“AI 垃圾”（AI Slop）**：缺乏人类深度参与的 AI 应用只会制造信息噪音与粗劣工具，业务人员的把关是实用性的唯一保障。
- **化解内部阻力**：主动吸纳对 AI 抱有疑虑的员工参与共建，是防止企业内部项目被暗中破坏（Sabotage）的关键手段。
---
### [13:43](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=823s) 现场互动与观众共鸣：重构人机连接与数据主权
演讲进入 Q&A 环节后，来自波士顿 AI 证明基础设施公司 Orin 的观众 Kristen 主动接过了麦克风。她对 Matt 的观点表示高度认同，并进一步展开了关于人类与 AI 连接本质的探讨。Kristen 回顾道，互联网（Internet）的词头“Inter”最初代表的是连接坐在椅子上的真实人类；而在当下的新世界中，我们面临着如何连接合成客户（Synthetic Customers）与合成 Agent 的挑战。她分享了本周大会上的一个数据：企业内部每拥有 1 名员工，未来可能就会对应多达 75 个 AI Agent。
面对如此庞大的 Agent 规模，Kristen 指出关键在于 Agent 正在处理谁的数据，以及我们如何确保个体的自主权（Agency）。Matt 所说的“将杠杆交到员工手中”，本质上是将数据的主导权与代理权还给个人——个人的名字、人口统计学特征、专属 API 就是最大的博弈筹码。她呼吁行业必须逆转现有的利益机制，建立“未经我的明确许可，AI 不得擅自行动”的规则体系，打破被锁死的数据与知识产权壁垒。Matt 随后向 Kristen 的发言致谢，并再次总结道，AI 是一场深刻影响每一个人的技术浪潮，唯有赋能身边的所有人，才能构建真正健康的生态。在友好的交流氛围中，Matt 宣布演讲结束，并邀请大家前往展台做进一步深入探讨。
- **Agent 与人类比例**：行业预测显示，企业内每个真实员工未来可能对应多达 75 个 AI Agent。
- **个体代理权回归**：将个人身份、人口特征及 API 接口转化为员工自身的杠杆，建立“默认禁止、授权方可执行”的合规机制。
- **全员赋能共识**：AI 变革关乎所有组织成员的利益，只有让每个人拥有工具的使用权与掌控权，AI 才能真正成为正向生产力。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) AI 将取代人类智能（AI will replace human intelligence） — Matt Cashatt
> [00:04](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=4s) 如果这项技术不能让广大基层工作者感受到益处，公众对 AI 阴暗面的恐惧与想象就会演变成现实，最终反噬整个行业。 — Matt Cashatt
> [02:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=142s) 60% 的高管竟表示计划解雇那些拒绝使用 AI 的员工 — Matt Cashatt
> [02:22](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=142s) 大众拒绝的不是技术本身，而是推行技术的“霸王条款（The terms）”。 — Matt Cashatt
> [05:15](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=315s) 他们根本一开始就没要求过要上火车，也不想去任何地方。 — Matt Cashatt
> [05:15](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=315s) AI 绝不是人类大脑的替代品，而是人类智能的“火箭助推器”。 — Matt Cashatt
> [09:47](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=587s) IT 和管理层可以彻底“让出道路”，让一线员工直接使用 Studio 构建解决眼前问题的应用——哪怕这个应用生命周期只有一天。 — Matt Cashatt
> [11:23](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=683s) AI thinks fast and humans think forward（AI 负责快速思考，人类负责前瞻思考） — Matt Cashatt
> [11:23](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=683s) AI 是否能成为一股造福企业的力量，完全取决于我们选择带谁一起坐到谈判桌前。 — Matt Cashatt
> [13:43](https://www.youtube.com/watch?v=ZEwPu67JWWw&t=823s) 未经我的明确许可，AI 不得擅自行动 — Kristen
## Mentioned
Matt Cashatt, Polyrific, Enterprise Solutions Engine, Catalyst Studio, Governed Agentic Runtime, 路透社 (Reuters), 皮尤研究中心 (Pew Research), Writer, Workplace Intelligence, 哈佛大学, Google, Codex, Cloud Code, Lovable, Base44, Salesforce, SAP, Oracle, Claude, Orin, Kristen
## Video description
Matt Cashatt - CEO, Polyrific
The next wave of enterprise AI will not be built by engineering alone. As agentic AI matures, business teams can increasingly build their own problem-solving applications, but most platforms force a choice between speed and control. This session introduces a new category, the Enterprise Solutions Engine, designed to resolve that tradeoff: a self-service environment where teams design and ship AI applications connected safely to corporate data, on infrastructure the enterprise governs. Matt Cashatt walks through how this reshapes the build-versus-buy calculus, why the traditional AI platform stack falls short, and what it takes to give business users real building power without sacrificing security, observability, or compliance. Executives will leave with a clear lens for evaluating platforms and a playbook for unlocking AI velocity across the organization.
[Source](https://www.youtube.com/watch?v=ZEwPu67JWWw)
Source quotes
[0:04] Matt Cashatt: AI 将取代人类智能（AI will replace human intelligence）
[0:04] Matt Cashatt: 如果这项技术不能让广大基层工作者感受到益处，公众对 AI 阴暗面的恐惧与想象就会演变成现实，最终反噬整个行业。
[2:22] Matt Cashatt: 60% 的高管竟表示计划解雇那些拒绝使用 AI 的员工
[2:22] Matt Cashatt: 大众拒绝的不是技术本身，而是推行技术的“霸王条款（The terms）”。
[5:15] Matt Cashatt: 他们根本一开始就没要求过要上火车，也不想去任何地方。
[5:15] Matt Cashatt: AI 绝不是人类大脑的替代品，而是人类智能的“火箭助推器”。
[9:47] Matt Cashatt: IT 和管理层可以彻底“让出道路”，让一线员工直接使用 Studio 构建解决眼前问题的应用——哪怕这个应用生命周期只有一天。
[11:23] Matt Cashatt: AI thinks fast and humans think forward（AI 负责快速思考，人类负责前瞻思考）
[11:23] Matt Cashatt: AI 是否能成为一股造福企业的力量，完全取决于我们选择带谁一起坐到谈判桌前。
[13:43] Kristen: 未经我的明确许可，AI 不得擅自行动
Source quotes
[2:00] Matt Cashatt: They're marching against AI in the street and then they're leaning on it when they go back home.
[5:05] Matt Cashatt: never asked to be on a train or go anywhere in the first place.
Original transcript
[0:04] All right. Thank you all for being here. Raise
your hand if you were here for the last panel
[0:07] that was just in here. Anybody? There's
a couple of you. Okay. And keep your hand
[0:10] up if that's the the issues they raise.
Were they genuinely a concern for you?
[0:14] Okay. Great. You're welcome to come see us.
I'm with Polyrific. We've built a platform that
[0:18] literally solves every one of the problems
they talked about. And I do mean everyone.
[0:23] Uh we're going to talk about the second part of
that. I did give a speech on that yesterday. We
[0:26] talked about on that one. Today we're going to
talk about the enterprise solutions engine. And I
[0:29] made some lastminute changes to this presentation
because I wanted to do something that's actually
[0:33] interesting for everybody and speaks to a social
problem that we all have with AI right now that
[0:38] we need to be considerate of. It ends on a very
happy note. Don't worry. Um but but I'd like to
[0:43] kind of kind of get into that. We're still going
to talk about when your business builds its own AI
[0:48] because that's what was builded here and that's
part of part of what we're talking here. But
[0:51] we're going to talk specifically about how we
move that closer to the hands that do the work
[0:55] and include the people that need to be included,
the stakeholders who right now are terrified
[0:59] of AI, the people, the colleagues that you work
with, your friends, your family, your neighbors,
[1:03] and give them a place at the table
when we're making these decisions.
[1:08] There's a story going around that AI will replace
human intelligence, and it's the wrong story.
[1:13] And the problem is, it's a story that people
like me and and perhaps some of you all out
[1:17] in the crowd today are nurturing and growing.
Being in this room probably means that like me,
[1:22] you're enthusiastic about AI. I mean, raise your
hand if you're optimistic about the future of AI.
[1:27] I certainly am. I I'm a little bit of a nerd about
it. I I love it. People around me and my personal
[1:32] life probably get tired of hearing me talk about
it. But we recognize AI is something that's going
[1:36] to be a force for good. It's going to be probably
the most influential tech in our careers, maybe
[1:41] our lifetimes. But the question is, is it really
a force for good? And the cold fact that we all
[1:47] need to hear is that it's not helping everybody
right now. AI isn't good for everybody right now.
[1:53] And if it isn't helping everybody, we're all at
risk for the darker side of AI, the darker things
[1:59] that our imaginations can conjure, turning it into
something that's not good for society as a whole.
[2:04] So, making sure that these outcomes are good start
with us. And I'm going to get to how we do that.
[2:10] Let me back Let me back up just a little bit. See
if I can't No, I can't do that. Here we go. Sorry,
[2:14] guys. trying to I like I said I I changed up the
slide deck right at the last minute because it's
[2:17] important to me to to fit this stuff in. I want to
make the point that I make my living creating AI
[2:22] powered agentic solutions. We've built a platform
uh that that manages and governs agents. We're
[2:28] good at it. A lot of you all in this room are
probably good at it as well. Um but it's kind
[2:35] of put us at this crossroads. While we're having
fun here in Vegas enjoying ourselves and and and
[2:39] proud of how smart we are for building these
solutions, we've got a lot of colleagues back
[2:42] at the office that are probably terrified of it.
And I've only come to really realize that lately.
[2:48] One Saturday last month in July, Reuters counted
142 protesters against AI across 42 different
[2:55] states. Pew Research found last year that 52% of
American workers are worried about AI's impact
[3:00] on their jobs. writer and workplace intelligence
this year found 29% of workers admit to sabotaging
[3:07] their own company's AI rollout. 44%
of those are among Gen Z participants.
[3:13] And in the same study, six 60% of executives
plan to lay off employees who won't use AI.
[3:20] That's a research finding that reads
like a labor relations board from 1892.
[3:27] This is what that looks like. Not a think
piece, not a poll. These are people in the
[3:32] street in New York, Atlanta, Berkeley, London. At
home, they could be your neighbors. They could be
[3:37] your friends. They're not going to tell you that
they're afraid of AI, maybe as tridently as as
[3:42] they could because they're going to be friendly
about it, right? And I was deep in a bubble
[3:47] thinking that everybody around me shared my
optimism for what I do for a living. Here's the
[3:53] interesting thing though about it, and it creates
a new kind of tension. 67% of workers say they'd
[3:59] rather seek guidance from an AI rather than their
own manager. And that says something about AI and
[4:04] the optimism we share for it. Many of us share for
it. And it says something worse about management.
[4:09] And roughly 90% of workers already use personal AI
regularly while only 40% of companies provide it.
[4:16] So they're marching against AI in the street and
then they're leaning on it when they go back home.
[4:21] How can both of those things be true? Here's the
fact. Nobody was asked. these people that are
[4:26] pissed off about it, the people that are angry,
the people that are scared, they're terrified,
[4:31] it's because they weren't giving a choice given
a choice about this. They weren't given a choice
[4:34] about their involvement is something that was kind
of foisted on them. People weren't consulted about
[4:38] how AI would change their work or threaten their
jobs, or in some cases, they weren't consulted
[4:43] about how their work fueled the very beast that
they would later be fed to. They weren't asked
[4:49] whether they'd like a large data center and
a power plant to support it in their backyard
[4:53] or the electricity and water bills to match.
They aren't rejecting the technology. They
[4:58] might even share our optimism for it. What they
are rejecting is the terms. The wrong story is
[5:05] about replacement, though. The real story is about
choice and inclusion. It's about giving people
[5:10] who will acutely be affected by the things that
you and I are building a place at the table.
[5:15] Here's where it gets positive. Don't worry. We
got some people leaving. I don't blame them.
[5:19] I want to be clear about what I believe in
this sentiment. The human mind, I think,
[5:27] and I really believe this, is the most magnificent
thing in the universe. I've been fascinated by AI
[5:33] for as long as I can remember, certainly my entire
adult life, and I've jumped into it full throttle
[5:38] in my career. But the more I've learned about
artificial neural networks and how AI works, the
[5:42] more I've learned about the human brain and how
unsurpassed it is. And there's the silver lining.
[5:48] The human mind is a different kind of
intelligence. It's human intelligence.
[5:53] It's hi. It's something different, right?
We're we're not we're not uh interchangeable.
[5:57] In 2024, Google and Harvard published the most
detailed map ever made of the human cortex. One
[6:02] cubic millimeter of it. The map alone took
1.4 pabytes, tens of thousands of cells,
[6:09] more than sorry, tens of millions, more than 150
million synaptic connections. And the whole brain
[6:14] runs on 20 watts. That's a light bulb. That's a
light bulb. I've been fascinated by AI, as I said,
[6:20] forever. And I'm even more fascinated by the human
brain and what we can do together. For those of us
[6:26] that truly understand AI's capabilities, that's
a separate and more frustrating tension to hold
[6:31] because we know the brain's great. We know that
AI isn't going to replace it. It's a rocket ship
[6:35] for it, right? Something really different.
And we want everyone to benefit from that.
[6:42] Here's the confession. And I speak a lot on
this topic and once very recently to a group
[6:46] of business owners in in my home state of New
York. Um, and I've come across as a complete ass,
[6:52] I think, and I only realized it lately because I
sat in a panel a while ago and I talked about how
[6:59] if you didn't get on the train, you're going to
be left behind. What I didn't realize at the time,
[7:03] I coldly realized it when I got no
reception out of that room walking out,
[7:07] was that they never asked to be on a
train or go anywhere in the first place.
[7:10] And I think that's something we would all do
well to kind of remember as we start building
[7:14] this agentic future which will be there which we
will be building for everyone. That train metaphor
[7:20] made sense to me at the time. It sure didn't make
sense to them or it did but they they didn't want
[7:25] to go anywhere in the first place. And that's
the real bomb for all of us as as practitioners
[7:29] to diffuse. How do we open the door to what AI
can do for the people who'd benefit the most
[7:34] if their minds seem closed to the possibility?
Except here's the thing. Their minds aren't
[7:39] actually closed. They use AI every day and they
get it. But since AI will affect them intimately,
[7:44] they deserve a say. And the best way to give them
one is to move the lever of AI closer to the work
[7:49] and let the people doing that work decide
how to deploy it. And in the business case,
[7:54] the business case agrees with most people.
Many enterprise AI pilots don't include
[8:00] the enduser in the design process. I see this over
and over again. They don't include the end user in
[8:05] the design process. And most of those pilots are
failing at exactly that distance. They don't work.
[8:11] You've heard it before. We're not getting ROI.
We're not getting adoption. But that's because
[8:15] you didn't include the people that have to use
it in the decision-m in the design process.
[8:22] That's where the traditional AI stack that
we're starting to build is falling short.
[8:25] One central team, one Q, generic tools bolted on
to old workflows andmemes not being consulted.
[8:33] Quick show of hands. How many of you use something
like codeex or cloud code? Lovable base 44.
[8:39] Several of us in here. That makes sense.
It's very encouraging. Generally, genuinely,
[8:42] that is a good thing. We use it all the time.
[8:45] The thing is they don't give you production
ready applications, do they? There's domain
[8:49] research that's needed. There's UX, there's
QA, there's infrastructure, there's security,
[8:53] there's all kinds of code reviews that we do and
and and tests and user acceptance testing to make
[8:57] sure it works in the real world when we make an
app. and the not the least the subject matter
[9:02] expert who knows how the hell this application
this solution supposed to work in the first place.
[9:07] Great app ideas and they make great mockups but
they're not production and that creates a gap.
[9:13] At Polrific we're closing that gap. We've built
the runtime. I talked about that yesterday and it
[9:17] does solve many of these problems. I think all
the problems the panel before me talked about
[9:22] and then we built something else that's in
the lab currently that I'm really proud of
[9:25] and we think it's a new category. It's it's
called Catalyst Studio and it runs on top of
[9:31] a governentic runtime to allow you to
build solutions to business problems.
[9:36] I'm I'm talking about the category here at AI4.
Make sure I'm on the right slide with you guys
[9:44] because not because we want to claim it
for ourselves, but because I do think
[9:47] it's a category that will come out soon.
It'll sit alongside the ERP and the CRM
[9:51] just as those became very very popular.
Shout out if you if I'm just curious here
[9:55] as a little experiment. Who would
who do you associate with a CRM?
[9:59] Salesforce, right? I heard Salesforce a couple
of times. Who do you associate with an ERP?
[10:07] SAP, Oracle, right? There there's all
these that we reflexively think of.
[10:12] We want to be associated with the
ESSE, the enterprise solutions engine,
[10:15] but I don't think we're the only ones that are
going to do it. I think the missing link for
[10:18] a lot of people in using something like cloud
code to build a solution to a business problem
[10:22] is that runtime that you have to sit on to
have the confidence that your agents are going
[10:26] to be managed in the first place. I sat here and
listened to the panel. I heard them talk about you
[10:30] don't know if an agent's impersonating a user or
not. Well, you do if they have a service account.
[10:34] If they're sitting in a govern runtime, that's
handled. That problem's been solved already.
[10:39] So if we solve those technical problems,
if we put that foundational work first,
[10:43] which at Polyrific we have then you have a studio
that you can build and you can allow even in
[10:49] cloud code or whatever you can allow people to
build solutions on top of that runtime and know
[10:53] that it's governed. What does that do for you? It
allows you to stop treating AI like an IT project.
[11:01] It allows you to now include the people who are
terrified about AI in the design process. In fact,
[11:07] it ultimately allows you to get out of their way.
You're not even between them anymore. Just allow
[11:12] them to have access to a studio where they can
build the solution to their problem. And it might
[11:16] be a solution that they only need for one day.
That's what the agentic future does for all of us.
[11:23] We're adding a new kind of intelligence to our
organizations. It's one one that excels at speed
[11:27] and breadth and everybody everybody should be able
to leverage this. I really believe that nobody
[11:32] should be left behind on this one because the gap
in benefit is too big. The stakes are too high
[11:38] because that's the difference between giving
them a tool that they can leverage and then
[11:42] handing them a tool that's basically wielded
towards them. No one likes the way that feels.
[11:48] Here's the bet we're making at Polyrific.
AI thinks fast and humans think forward.
[11:53] I I'm I'm a little overly proud of that because I
thought about it for a while and and to to confess
[11:58] that probably came from Claude or something
like that as I was workshopping everything,
[12:01] but I liked it because it works and it's right.
What is AI doing for us? It's just faster. It can
[12:08] cover more breath for you more quickly. But that's
important. It's helpful. But we think forward. So
[12:13] AI can do in seconds what it took a whole team
a week. But we can't what it cannot do is decide
[12:18] what's worth doing in the first place, right? It
can't decide where to aim next. or what person
[12:24] across the table it actually needs to to involve
in this process. That part stays human and keeping
[12:29] the human stakeholders in the driver's seat is
what separates actually useful apps from AI slop.
[12:35] Raise your hand if you're tired of seeing AI
slop. I am. I'm tired of reading emails that are
[12:40] way too long, you know, that someone could have
written in a sentence and sloppy applications.
[12:47] Make no mistake, the agentic age is going
to belong to the fearless. we're driving it,
[12:52] right? Anything I said doesn't obiate
the fact that we're still in charge,
[12:55] that this is still a force of nature almost
and we're moving forward with a gentic AI,
[12:59] but whether it becomes a force for good or not
depends on who we bring to the table with us.
[13:05] And I know that this talk, these talks like this
at these conferences, I'm supposed to be hawking
[13:09] my own products and I did that. I think I did
my job there. But what's more important that
[13:13] you take away is that the process of including
people who might be terrified of what we're
[13:18] doing is going to make our technology, our
industry that we work in more sustainable,
[13:25] less bumpy, and we can get right to the the
positive promises of AI that we've all read
[13:30] about for many years now. Without that, we're
going to be operating with a lot of people who
[13:33] don't understand it and don't trust us and
push back and create obstacles. Like I said,
[13:37] that statistic, I forgot the number already of how
many people sabotaging the projects from within.
[13:43] Okay, that's all I had. That was my time. I've
got a lot of time left and I I'm probably going
[13:48] to regret doing this, but does anybody have any
questions or do you have comments? You want the
[13:52] mic? Happy to share it. Hi everybody. I'm Kristen.
I am working in an AI at astation infrastructure
[13:59] company called Orin. We're based out of Boston,
Massachusetts. And I want to continue with what
[14:05] Matt was saying because I really appreciate your
talks. We have a lot of similar overlap here where
[14:09] we're working on the idea of identifying
the information that makes us all human.
[14:16] When we think about the internet, we think about
that prefix of inter. It was connecting humans in
[14:21] chairs. What is the information that is actually
connecting us to synthetic customers, synthetic
[14:29] agents? We learned this week that there's 75
agents for every one person in your organization.
[14:37] What is the data that they're handling?
What is your data that it's handling?
[14:42] And when Matt says we're putting levers into
the worker's hands, what he's trying to say is
[14:48] we're putting agency back into the individual.
Your name, your demographics, your API,
[14:55] that's your leverage. And it's the moment for us
to say, "Hey, let's actually flip the incentives.
[15:03] You cannot do this until I say it's able to." Is
anybody else out there experiencing this issue
[15:10] with locked and shackled IP? Anybody
else having issues with moving data?
[15:15] No. Good. Hey, thanks for coming up. I appreciate
it. I I look, I'm a believer and everybody has
[15:22] something to say about this. This is this is
a AI is a movement that belongs to all of us.
[15:26] we're all impacted by it uh one way or another.
Again, are there any other questions that I can
[15:29] answer or any comments that that people who feel
strongly about what we're doing and how we empower
[15:34] all the people around us with AI?
The comments you'd like to make.
[15:39] No, this time to get the hell out of here, huh?
[15:42] All right. Hey, thank you everybody. I really
appreciate you. Thanks for uh thanks for
[15:46] being here. Thanks for coming to the room.
That's all we got. Come see us at our booth.
