Polyrific CEO Urges 90-Day AI Agent Rollouts Over System Overhauls

Polyrific CEO呼吁以90天AI Agent部署取代系统全面重构 将受边界约束的AI Agent直接部署在遗留基础设施上，可释放员工的决策带宽，并在竞争对手赶超之前实现复合式生产力提升。 企业部署AI Agent无需经历繁琐的系统全面重构，而是可以通过协议直接叠加在现有的ERP、CRM等遗留系统之上，并以只读访问保障安全。依托涵盖推理、自主决策与任务执行的RAT框架，AI Agent已步入早期大众普及阶段，多数企业在部署后12个月内实现了生产力提升、显著的投资回报以及明显的竞争优势。 AI Agent 的核心运作模式可概括为“RAT”框架，即推理分析数据（Reasoning）、在既定护栏内自主决策（Autonomy）以及执行具体任务（Task Execution），全程置于人类监管之下。 (1:43) 企业级 AI Agent 正式跨入早期大众普及期，目前已有 79% 的企业正在积极部署，66% 的企业已获得可衡量的生产力提升，88% 的高管正在追加相关预算。 (3:12) 实际落地已验证出显著 ROI，例如 Verizon 的销售转化率提升 40%，Petrobras 实现了 1.2 亿美元的税收节省，保险核保评估时间也被压缩至 5 分钟以内。 (4:56) AI 架构无需全面替换现有的 ERP、CRM 或数据湖，借助 MCP 与 A2A 协议即可直接挂载在遗留系统之上，初期通过默认只读权限来有效控制安全风险。 (8:14) 建议企业执行 90 天敏捷落地路线图：前 30 天深入一线找出最痛的业务瓶颈，31–60 天坦然面对试错并吸纳批评者以完善安全与权限，61–90 天则向各业务部门横向扩展成功经验。 (16:43)

Matt Cashatt, founder and CEO of Polyrific, argues that enterprise AI agent adoption has transitioned from speculative experimentation into the early majority, offering rapid returns when organizations deploy bounded software over existing IT infrastructure. He contends that commercial success depends less on complex technical overhauls than on eliminating routine employee drudgery and managing organizational change. The RAT Framework and Pragmatic System Integration (1:43) Cashatt defines an AI agent using the "RAT" framework: Reason, Autonomy within guardrails, and Task execution. Rather than representing unconstrained intelligence, enterprise agents function as software integrated with internal systems to make bounded decisions based on contextual data, while keeping human operators in oversight. He emphasizes that companies do not need to overhaul legacy ERP or CRM platforms or wait for pristine data hygiene. Instead, deployments should layer directly over existing infrastructure using lightweight middleware, Agent-to-Agent (A2A) protocols, and the Model Context Protocol (MCP), while enforcing read-only access permissions during initial rollouts to mitigate risk. Market Adoption Benchmarks and Enterprise Case Studies (3:12) According to benchmarks cited by Cashatt, 79% of companies are actively adopting AI agents, 66% report measurable productivity improvements, 88% of executives are expanding AI budgets based on demonstrable return on investment, and 73% of CEOs anticipate gaining competitive advantages within 12 months. These metrics reflect a shift toward standardized implementation patterns rather than isolated experimentation. To substantiate these gains, Cashatt highlights quantifiable corporate deployments. Verizon achieved a 40% increase in sales close rates across business-to-consumer and enterprise channels by modeling agents on top-performing representatives. In Brazil, Petrobras identified $120 million in tax savings in three weeks by using agents to reconcile complex financial datasets, while automated mortgage processors compressed approval windows to two days. Similarly, Polyrific implemented a multi-modal agent for property and casualty insurers that analyzes handwritten notes and inspection photographs to deliver concise recommendation summaries, reducing underwriter review times to under five minutes per application. The AI Value Pyramid and Cognitive Relief (10:20) Cashatt outlines enterprise value across a three-tiered pyramid. The baseline consists of efficiency gains, where automating universally disliked, repetitive tasks establishes early momentum. The intermediate tier relieves workplace decision fatigue by delegating micro-decisions to agents, thereby preserving human cognitive stamina for complex operations. At the peak sits human innovation enablement. Cashatt asserts that artificial superintelligence (ASI) will not arrive in our lifetime, arguing that artificial neural networks merely simulate past patterns rather than generate genuine, unmapped conceptual breakthroughs. Consequently, using agents to remove administrative overhead is designed to unlock inherent human creativity and product development without reducing overall organizational headcount. The 90-Day Implementation Roadmap and Governance (16:43) Cashatt recommends executing rollouts in structured 90-day cycles rather than multi-year transformation projects. The first 30 days require walking operational workflows to identify the most frustrating friction point—the organization's operational bottleneck—and assembling internal advocates. Days 31 to 60 involve iterative piloting, candid internal communication when systems fail, and actively engaging fierce internal critics whose concerns typically pinpoint legitimate security gaps or anxieties about job security. Days 61 to 90 focus on codifying lessons and horizontally replicating proven agents across business units. Because Cashatt views adoption as 80% people and 20% technology, executive sponsorship, transparent town halls, and continuous training are essential. Additionally, he advises organizations—even those based in the United States—to voluntarily adopt strict European Union AI compliance standards to foster workforce trust and preempt future domestic regulatory tightening. Cashatt concludes that early investment yields compounding returns, projecting that basic 90-day proofs of concept can evolve into interconnected multi-agent ecosystems past 180 days. He warns that delayed adoption risks insurmountable competitive disadvantages, though successful execution fundamentally hinges on proactive change management and resolving employee concerns regarding security and job retention.

Polyrific创始人兼首席执行官Matt Cashatt指出，企业AI Agent已跨越早期炒作阶段，正通过解决低效流程和决策疲劳释放核心业务价值；企业无需推倒既有IT架构，而应通过90天敏捷迭代与严密的人员变革管理实现复利增长。 RAT定义与早期多数派采纳现状 (1:43) Cashatt将AI Agent的核心运行机制概括为“RAT”框架，即推理（Reason，结合上下文分析企业数据）、自主性（Autonomy，在人类设定的安全边界内制定决策）以及任务执行（Task execution，连接业务系统付诸行动）。他强调人类依然掌控全局，Agent本质上是在受控权限内执行操作的软件。 行业采纳指标显示，企业级Agent已迈入“早期多数派”阶段：79%的企业正在积极推进Agent部署，66%的企业已录得可量化的生产力提升。此外，88%的高管正在追加AI预算，73%的CEO预计在未来12个月内通过部署Agent获取直接竞争优势。预算的松绑表明CFO们看到了实质性的投资回报，而非单纯追逐概念。 跨行业高价值用例与IT落地架构 (4:56) 多行业实践证实了Agent的工作流价值。例如，Verizon通过提炼金牌销售代表的行为模式并构建Agent进行规模化复制，使B2C与企业渠道的销售成单率提升了40%；巴西国家石油公司（Petrobras）部署税务策略Agent，在3周内对庞大分散的财务数据完成对账，识别出1.2亿美元的节税空间；银行业则利用按揭自动化Agent将贷款审批周期压缩至2天。在财产及意外险（P&C）核保中，Polyrific开发的多模态Agent能够审阅包含手写批注和勘验照片的复杂申请件，生成2至3行的审批建议摘要，将单件核保时间缩短至5分钟以内。 在IT架构集成层面，企业无需替换原有的ERP或CRM系统。现代Agent架构可借助轻量级中间件，结合Agent间通信（A2A）与模型上下文协议（MCP），在传统系统与分散的数据仓库上直接构建连接层。Cashatt强调，即使企业内部数据尚未完全规范化，也可以即刻开展集成，但在部署初期应将Agent的权限默认设置为“只读”，以确保系统安全与运营合规。 价值金字塔与人工超级智能边界 (10:20) 企业AI创造的价值呈三层金字塔结构。底层是作为基础门槛的效率提升，重点是消除员工日常厌恶的重复劳动，如手动录入PDF数据至表格。中层是缓解决策疲劳，现代员工常因日常琐碎的微决策而耗尽精力，将常规运营判断移交Agent可释放高阶认知带宽。顶层则是激发创新，将员工从底层杂务中解放，激活潜藏的业务创意。 Cashatt结合神经科学与人工神经网络分析认为，人类智能在本质上与机器系统截然不同。他断言人工超级智能（ASI）在当代人的有生之年不会到来，因为AI模型仅能基于历史数据进行模拟或重组，缺乏在未知领域构想未来的原生创造力。因此，企业部署Agent的目标并非缩减人员规模，而是通过人机协同强化创新飞轮。 90天落地路线图与人员变革管理 (15:28) 企业应摒弃长达数年的数字化转型周期，转向90天敏捷推进。前30天的目标是深入业务一线寻找最受员工反感的流程瓶颈（类似电影中的“TPS报告”），并动员内部支持者；第31至60天进入试点与修正阶段，团队需坦然面对试错过程并保持透明沟通，同时主动拉拢最坚决的内部反对者，直面其关于数据安全、控制权丧失或失业的担忧，将其转化为方案倡导者；第61至90天则提炼标准化经验，向其他业务条线横向复制推广。 Cashatt指出，AI转型的成败20%取决于技术，80%取决于人员。高管层必须尽早参与以防后期战略脱节，并通过全员大会和系统培训消除恐慌。在合规方面，他建议即便仅在美国本土运营的企业，也应主动向欧盟严苛的AI监管标准看齐，此举不仅能增强员工信任，还能提前防范未来本土监管政策的收紧风险。 早期AI部署具有类似财务储蓄的复利效应：从90天的单一任务验证，到180天覆盖多业务条线，最终演进为互联的多Agent生态系统。未能尽早布局的企业可能会在智能化进程中与领先者拉开不可逆的代际差距。

## Background 本视频由 Ai4 频道于 2025 年 8 月 24 日发布在 YouTube 平台，主讲人为 Polyrific 创始人兼首席执行官 Matt Cashatt。页面简介指出，随着绝大多数企业高管将 AI 视为重塑职场的关键力量，演讲旨在通过真实商业案例与渐进式部署战术，帮助企业跨越 AI 智能体的概念炒作并建立可持续的战略竞争优势。 ## TL;DR 演讲系统阐述了企业落地 AI 智能体的务实战略，首先通过 RAT 模型（推理、合规自主权与任务执行）解析了智能体的本质运作机制。随后，Matt 结合 Verizon 销售提效与保险核保提速等实战案例，指出企业无需推倒重来更换 ERP 或 CRM，通过 MCP 等轻量协议叠加于现有系统之上即可快速见效。他提出了涵盖“提升底层效率、消除决策疲劳、释放人类原生创新”的价值金字塔，并给出了以攻克组织内部痛点流程为起点的 90 天敏捷实施路径。最后，他强调企业应主动向高标准治理看齐并重视人际协同，警示 AI 智能体的快速复利效应将让犹豫不决的企业面临“曲速飞船对阵单马木车”的致命竞争鸿沟。 ## Atmosphere & style 现场氛围幽默风趣且极具务实感染力，演讲者开场便以会后小聚和现场“脱外套”的趣味示范破冰，迅速打破了高阶技术演讲常见的枯燥与神秘感。叙事风格极其贴近一线职场现实，对日常决策疲劳以及人见人烦的“TPS 报告”等痛点的精准剖析频频引发台下共鸣。整场分享的情绪由最初的亲切宽慰自然过渡到后半段对代际竞争差距的严肃警示，既没有悬浮的科技说教，又充满了令人跃跃欲试的实战紧迫感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=4s) 传统与非 AI 原生企业落地智能体无需深陷底层技术泥潭，通过成熟的中间件与抽象层即可快速切入并推进 AI 战略。 - [01:43](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=103s) AI 智能体的核心运作逻辑可解构为 RAT 模型：基于上下文数据的推理（Reason）、在合规护栏下的自主权（Autonomy）以及闭环的任务执行（Task Execute）。 - [03:12](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=192s) 企业 AI 智能体部署已跨越早期采用期步入大众普及期，79% 的企业已引入智能体，CFO 与高管层正因明确的 ROI 大幅追加预算。 - [04:56](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=296s) 从电信销售转化率提升 40%、能源巨头挖掘 1.2 亿美元节税，到财产险核保降至 5 分钟内，AI 智能体在垂直场景展现出高回报价值。 - [08:14](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=494s) 最优秀的 IT 集成往往是非侵入式的，企业无需重构 ERP/CRM 或整理完美的数据湖，借助 MCP 与 A2A 协议并坚持初始“只读权限”即可平滑落地。 - [10:20](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=620s) AI 智能体价值金字塔由底层效率跃升至中层：接管成百上千项低风险微决策，彻底消除员工下班后的“决策疲劳”。 - [12:22](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=742s) AI 擅长总结复述既有数据，但人类拥有独创的前瞻性想象力；AI 充当创意的“压裂液”，旨在赋能现有员工释放创新而非单纯裁员。 - [15:28](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=928s) 推行 90 天敏捷落地路线：前 30 天管理层应深入一线，精准攻克类似《办公空间》“TPS 报告”般全员痛恨的低效瓶颈以建立先发势头。 - [16:43](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1003s) 在试点试错阶段需积极吸纳严苛批评者，将安全合规与失控担忧转化为完善系统的动力，进而横向复制多代理网络。 - [19:06](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1146s) 合规治理是核心护城河，主动对标严格的欧盟标准有助于建立内外部信任，并从容应对未来监管收紧的风险。 - [20:26](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1226s) 数字化转型遵循“20% 取决于技术，80% 取决于人”，项目成功必须依赖高管顶层支持、全员透明赋能与创新型组织文化三大支柱。 - [22:01](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1321s) AI 智能体具备高度压缩的指数级复利效应，先行者与观望者之间的差距将迅速拉大为“曲速飞船与单马木车”的不可逆代差。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=4s) 开场破冰与 Polyrific 的企业级 AI 赋能使命 在拉斯维加斯大会最后一节演讲的现场，Polyrific 创始人兼首席执行官 Matt Cashatt 以轻松幽默的方式开场。面对座无虚席的会议室，他调侃道，虽然这是全天最后一场分享，紧接着就是大家期待已久的欢乐时光（Happy Hour），但他保证这场演讲绝不枯燥，并热情地邀请大家会后一起去喝一杯。为了摸清台下听众的背景，Matt Cashatt 做了一个现场调研，询问在场有多少人所在公司的核心业务并非 AI 或与 AI 强相关。看到绝大多数听众举手后，他肯定地表示：“你们来对房间了，这正是我们想要服务的对象。” Matt Cashatt 指出，对于非 AI 原生的传统企业与大型机构而言，拥抱并落地 AI 智能体（AI Agents）并没有想象中那么困难或令人生畏，也无需深陷底层的技术泥潭。只要选对合作伙伴，底层的技术细节完全可以被高度抽象化。Polyrific 在企业级 AI 和自动化解决方案领域深耕已超过十年，不仅帮助众多企业实现了切实业务价值的落地，目前还在积极布局量子 AI（Quantum AI）解决方案。他提醒企业管理者，在快速掌握 AI 智能体战略的同时，也要为下一波量子计算与 AI 融合的浪潮做好准备。 - **核心受众定位**：非 AI 核心业务的传统企业与大型机构管理者，旨在帮助其建立高阶 AI 战略并快速落地。 - **技术演进前瞻**：Polyrific 拥有十年以上企业自动化与 AI 落地经验，下一步正推进量子 AI（Quantum AI）前沿方案的整合。 --- ### [01:43](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=103s) AI 智能体本质解析：RAT 架构与现场“脱外套”示范 为了消除大家对 AI 智能体的神秘感与过度复杂化理解，Matt Cashatt 在台上当场脱下了西装外套，以此作为最直观的具象化演示。他提出，AI 智能体的核心运作逻辑可以高度概括为 **RAT** 模型：**Reason（推理）**、**Autonomy（在合规范围内的自主权）** 与 **Task Execute（任务执行）**。以脱外套为例，他在台上感知到周围环境温度过高，通过情境数据“推理”出自己感觉偏热；随后在合理的规则边界内行使“自主权”决定脱掉外套（而不会做出全裸等违法失控的行为，这便是所谓的安全护栏 Guardrails）；最终落实并“执行了脱掉外套的任务”。 ```mermaid flowchart LR accTitle: RAT Framework for AI Agents subgraph S1["感知与推理 (Reason)"] A["环境与情境数据"] --> B["分析推断现状"] end subgraph S2["决策与护栏 (Autonomy)"] B --> C{"合规自主决策"} D["企业规则与安全护栏"] -.-> C end subgraph S3["落地行动 (Task Execute)"] C --> E["执行目标任务"] end ``` Matt Cashatt 强调，AI 智能体本质上就是一段能够理解其所处环境中上下文数据的软件程序。人类依然牢牢掌握着控制权，“它现在绝对不是终结者（Terminator）”。企业根据业务场景赋予智能体适度的自主权，让它在设定的边界内完成推理、做出判断并执行具体任务，无需将概念复杂化。 - **RAT 架构三要素**： - **Reason（推理）**：基于业务上下文与环境数据进行逻辑分析。 - **Autonomy（自主权）**：在明确的安全护栏（Guardrails）与业务规则内享有决策空间。 - **Task Execute（任务执行）**：将决策转化为端到端的具体行动与流程闭环。 --- ### [03:12](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=192s) 跨越早期采用期：企业智能体落地的关键数据指标 当前企业级 AI 智能体的部署已经跨越了“早期采用者（Early Adoption）”阶段，全面迈入“早期大众（Early Majority）”时代。Matt Cashatt 提醒在场嘉宾，制定并实施 AI 智能体战略的窗口期就是当下，“等到明天绝对为时已晚”。随着先行企业的探索，行业内已经逐步摸索并沉淀出成熟的设计模式（Design Patterns）和实战打法（Playbooks），这意味着新入局的企业不再是在荒野中摸黑前行，落地的风险与门槛已大幅降低。 资金流向与高管态度同样印证了这一拐点。以往对预算把控极其严苛的首席财务官（CFO）们，如今正主动放开资金支持，因为他们从实际落地中看到了真金白银的回报，而非空洞的技术炒作。如果不立即启动哪怕极小的一步，企业将在接下来的行业竞争中迅速处于劣势。 - **79%** 的企业已经在内部引入并采用 AI 智能体。 - **三分之二（约 67%）** 的企业报告了可衡量的生产力提升，验证了成熟打法的可复制性。 - **88%** 的高管正在增加 AI 相关的专项预算，CFO 看到明确的商业价值回报。 - **73%** 的 CEO 预计在未来 12 个月内通过落地 AI 智能体战略建立显著的竞争优势。 --- ### [04:56](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=296s) 落地实战与商业价值：从电信、能源到财产保险的成功案例 Matt Cashatt 分享了一系列企业通过部署 AI 智能体实现业务突破的典型案例： 1. **电信巨头 Verizon**：通过构建 AI 智能体来深度学习并模拟其顶尖销售冠军的沟通模式与业务策略，并将其能力固化为标准化智能体流程，推广至企业渠道、B2C 渠道、设备及合约销售业务中，在一年内将销售成单率（Close Rates）大幅提升了 **40%**。 2. **巴西能源巨头 Petrobras**：由于涉及数百亿美元的庞大业务体量，人类税务专家受限于精力与沟通协同成本，无法穷尽海量数据中的潜在模式。公司引入税务智能体后，仅在部署上线的 **3 周内就挖掘出 1.2 亿美元的税收节省空间**。 3. **银行业房屋抵押贷款**：各大银行通过抵押贷款审核智能体重塑审批流程，将原本冗长的贷款审批周期压缩至 **2 天以内**。 ```mermaid flowchart TD accTitle: Polyrific PC Insurance Underwriting Workflow subgraph Ingestion["多源非结构化数据输入"] A1["保单申请表"] A2["现场检查报告照片 (如屋顶破洞)"] A3["表单边缘手写批注"] end subgraph AgentProcess["Polyrific AI 智能体审核"] A1 & A2 & A3 --> B["智能体多模态深度预审"] B --> C["提取关键风险与核保依据"] end subgraph Output["核保员快速决策"] C --> D["生成 2-3 行极简封面摘要 (Cover Sheet)"] D --> E["核保人员 5 分钟内完成合格审批"] end ``` 在 Polyrific 自身服务的众多案例中，财产及意外伤害险（P&C Insurance）是极具代表性的高成效场景。该行业普遍具有“高业务量、低利润率”的特点。某保险客户每天需要处理数百份保单，但由于人工核保耗时过长，支付给核保员的时薪成本甚至经常超过保单本身的利润，导致长期亏损。 Polyrific 在该客户核保团队的配合下训练了专用 AI 智能体，执行全方位的预审工作。该智能体不仅能解析文本，还能结合多模态能力分析房屋检验报告中的现场照片（例如自动识别出第 10 页照片中的屋顶破洞），甚至能辨识表单边缘的手写批注。智能体将复杂的审核结果浓缩成一份仅有 2 到 3 行核心结论的“封面摘要（Cover Sheet）”，明确指出建议批准或驳回的具体证据与页码。这使得核保人员对单个保单的审核决策时间骤降至 **5 分钟以内**，彻底扭转了核保成本倒挂的被动局面。类似的逻辑也正在物业管理领域快速推广，用于大幅缩短维护调度、零配件采购与报修响应的周期。 ### [08:14](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=494s) 优秀的 IT 集成往往“很无聊”：不推倒重来，直接叠加于现有系统之上 从 IT 架构的角度来看，最出色的 AI Agent 集成往往显得有些“枯燥乏味”，但这恰恰是企业最大的战略优势。企业在引入智能体时，完全不需要经历伤筋动骨的系统重构——无需更换现有的 ERP，也不必替换现有的 CRM。当前技术生态中涌现出的 Agent-to-Agent（A2A）协议以及模型上下文协议（Model Context Protocol, MCP），就像是翻译各种混乱异构数据的“罗塞塔石碑（Rosetta Stone）”。无论是利用 Polyrific 自主研发的中间件平台还是采用通用开源协议，核心逻辑都在于“就地取材”，直接与企业现有的系统和数据现状无缝对接。 很多管理者常常会因为企业内部的数据状态感到焦虑，觉得数据分散在各个角落、混乱不堪，甚至大量核心信息还躺在一个个零散的 Excel 表格里。对此，Matt Cashatt 宽慰台下的听众：“恭喜你们，你们和所有人一模一样。”即便是一些自豪宣称拥有完整 Data Lake（数据湖）、Data Warehouse（数据仓库）甚至“Data Ribbon（数据带）”的巨头客户，其内部数据资产同样凌乱如麻。企业完全没有必要被数据现状吓退，AI Agent 策略的落地并不复杂，只需构建一层轻量级的集成中间件即可将现有系统串联起来。Matt 还特别提出了一个极其务实的实操铁律：在初期部署智能体时，务必坚持“只给予 Agent 只读权限（Read-only access）”，在确保安全可控的前提下逐步推进。 ```mermaid flowchart TD accTitle: AI Agent 叠加集成架构 subgraph S1["企业既有系统（无需替换）"] A["ERP / CRM 系统"] B["数据仓库 / 数据湖"] C["分散的 Excel 与文档"] end subgraph S2["连接与协议层"] D["MCP 协议 / A2A 协议"] E["轻量级集成中间件"] end subgraph S3["AI 智能体层（初始阶段）"] F["只读权限 AI Agents"] G["业务流程自动化与分发"] end A --> D B --> D C --> E D --> F E --> F F --> G ``` - **非侵入式集成**：AI Agent 运行在既有 IT 基础设施之上，而非替代 ERP 或 CRM。 - **协议标准化**：利用 MCP 和 A2A 协议抹平数据异构差异，充当企业多源数据的“罗塞塔石碑”。 - **安全准则**：智能体上线初期一律配置只读（Read-only）访问权限，避免对底层系统产生误操作风险。 --- ### [10:20](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=620s) 价值金字塔的地基与中层：从效率突破到消除“决策疲劳” 在总结了大量 AI Agent 落地项目的成功与失败教训后，Polyrific 提炼出了一套“价值金字塔”模型。位于金字塔底座的基石是“效率提升（Efficiency Gains）”，这是企业必须拿下的基本盘。企业在启动 AI 项目时，应首先审视内部那些缓慢、低效且令人深恶痛绝的落后流程——例如“在 2025 年居然还需要人工阅读 PDF 再手动录入系统”。从低效流程切入是一种高度明智的组织策略：它能迅速覆盖最多的基层员工，最快展现效果，从而为项目赢得广泛的信任，将团队最初的抵触心理彻底转变为推动变革的顺风助力。 ```mermaid flowchart TD accTitle: AI Agent 价值金字塔 subgraph Pyramid["AI Agent 价值金字塔"] direction TB TOP["顶层：创新赋能 (Innovation Enablement)<br/>释放人类原始创造力与全新业务价值"] MID["中层：消除决策疲劳 (Decision Fatigue)<br/>将成百上千项微决策委托给智能体"] BASE["底座：效率提升 (Efficiency Gains)<br/>解决低效痛点、消除重复录入等苦差事"] end BASE --> MID MID --> TOP ``` 越过底层效率之后，便来到了金字塔的中层：解决职场中无处不在的“决策疲劳（Decision Fatigue）”。Matt 向全场抛出一个极具共鸣的日常场景：很多人明明在办公桌前坐了一整天，下班回家却感到精疲力竭，甚至会因为“今晚到底吃什么”这种琐事和配偶陷入僵局，因为大脑已经“再也不想做任何一个该死的决定了”。他建议大家在桌上贴一张便签，每做一次微决策（无论是给老板发邮件的措辞，还是午餐选什么三明治）就画一笔计数，一天下来便签绝对会被填满。每天成百上千次的微决策正在榨干员工的精力，而 AI Agent 的关键价值正是接管并自动化这些低风险的微决策，让员工彻底从无休止的精力消耗中解脱出来。 - **效率基石作为启动跳板**：优先消灭重复录入、文档搬运等低效工序，以此积累全员对 AI 的信心。 - **缓解决策疲劳**：员工每天需要做出数以千计的琐碎微决策，将常规判断委托给 Agent 能极大减少心理内耗。 --- ### [12:22](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=742s) 金字塔顶端：释放人类独特的想象力与“压裂式”创新 在谈及 AI 与人类大脑的关系时，有着 20 年深度技术沉淀的 Matt 分享了一个有别于硅谷主流叙事的洞见：在他看来，所谓的人工超级智能（ASI）在我们的有生之年绝不可能到来。人类大脑的智能机制与人工神经网络有着本质区别——人类拥有基于未知远景的“前瞻性想象力”，能够在全然未知的“深蓝海洋”中构想未来；而 AI 无论多先进，都只能模拟、重组并复述被喂入系统的既有知识，它无法凭空想象从未存在过的事物。因此，人类独有的创造力才是最宝贵的资产。 一旦通过 AI Agent 释放了员工的时间并消除了决策疲劳，企业就能直达金字塔的最顶端——“创新赋能（Innovation Enablement）”。Matt 形象地回忆道，每个人在孩提时代早餐前都能冒出 100 个奇思妙想，但随着成长，外界的环境、老师和说教都在不断打磨甚至压制这些创意火花。企业员工其实人人都有好点子，只是被低效的苦力工作和繁重的微决策死死掩埋。AI Agent 就像是“创意的压裂液（Fracking Fluid）”，能够冲破繁杂事务的阻碍，将潜藏在员工脑海深处的创新能量彻底释放出来。当员工有精力去构思更好的产品、为客户创造全新价值时，企业就会进入“持续创新—不断获胜—士气高涨”的正向飞轮。这也正是为什么企业不应借助 AI 盲目裁员，而应利用 AI 赋能并激活现有人才。 - **人机智能的本质分野**：AI 擅长总结与模拟既有信息，但面向未来的前瞻性想象力是人类独有的核心壁垒。 - **AI 如同“压裂液”**：AI Agent 负责扫除琐碎杂务，帮助企业压裂并释放员工被埋没的原生创造力。 - **构建创新正反馈**：员工从繁杂事务抽身后投入业务创新，推动企业形成持续打胜仗的良性循环。 --- ### [15:28](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=928s) 90 天落地战术：前 30 天找到全员痛恨的“TPS 报告” 对于企业推进 AI Agent 的落地节奏，Matt 直言自己是个“极度缺乏耐心的人”，比起长达三年的宏大叙事，企业必须在 90 天内看到真金白银的成果。为此，他给出了清晰的时间表规划：在战略启动的前 30 天内，管理者的首要任务是“实地走访业务一线”。无论是在单一业务纵深内由浅入深摸排，还是跨越多个垂直业务线进行横向考察，核心目的只有一个——与一线员工深入交流，揪出那个全员痛恨、效率极其低下的共同痛点。 Matt 借用经典职场讽刺电影《办公空间》（*Office Space*）中的桥段，将这种人见人烦的低效流程比作电影中人人白眼的“TPS 报告”。管理者在第一阶段要做的，就是精准定位自己组织里的那份“TPS 报告”。找到这个公认的痛点后，应立刻召集各方利益相关者共同参与设计，借助成熟的 Agent 构建平台快速将智能体开发上线。这种从最痛点切入的战术，能以最快速度在组织内部形成示范效应并确立战略先发优势。 - **90 天敏捷见效**：拒绝长周期的空中楼阁，以 90 天为一个完整见效周期推进智能体部署。 - **前 30 天核心目标**：深入业务一线，寻找组织内部如同《办公空间》“TPS 报告”般被全员反感的低效流程并优先攻克。 ### [16:43](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1003s) 90天实施路径（中后期）：试点纠错与化敌为友 在前30天内，企业需要动员并赋能跨部门的数百名员工，让他们成为 AI Agent 项目的支持者，以此扫清推进中的各种组织壁垒。进入第31天至60天阶段，核心任务远不止表面上的“试点与学习（Pilot and Learn）”，这更是主动试错、经历曲折探索的时期。在实际落地中，产品刚上线时大概率无法尽善尽美，团队必须撤回、重新配置、推倒重来并再次发布。贯穿这一阶段的生命线是“沟通、沟通、再沟通”，必须坦然面对失败，主动向利益相关者说明故障原因、已知现状以及修复方案。 ```mermaid flowchart LR accTitle: AI Agent 90-Day Implementation Lifecycle subgraph S1["第 1-30 天：破局与动员"] A["动员跨部门员工"] --> B["消除组织壁垒"] end subgraph S2["第 31-60 天：试错与吸纳"] C["小步快跑/承认失败"] --> D["拉拢最严苛批评者"] D --> E["解决安全与失控疑虑"] end subgraph S3["第 61-90 天：横向扩展"] F["沉淀学习经验"] --> G["跨业务线复制 Agent"] G --> H["构建多代理协同网络"] end S1 --> S2 --> S3 ``` 这一时期最关键的策略是“找到对项目最持批评态度的人，并把他们拉到身边”。正如老话所说：“亲近你的朋友，更要亲近你的敌人（Keep your friends close and your enemies closer）。”在实际落地中，最激烈的批评者往往能被转化为最坚定的支持者，因为他们的反对意见通常切中要害： - **安全性担忧**：批评者能帮助发现团队忽视的盲区，从而构建更健壮、安全的 Agent。 - **失去控制权的焦虑**：这是合理的诉求，团队必须正面应对并制定缓解措施，不能逃避或无视。 - **职业替代恐慌**：必须承认员工对失业的担忧。尽管人机结合的“协同模式”理论上无需裁员，但现实中仍需正面安抚员工顾虑、共同化解风险。 进入第61天至90天的冲刺阶段，团队将把沉淀下来的经验在组织内部进行规模化扩散，并将成熟的 Agent 架构复制到不同业务线与应用场景中，推动 AI 能力在整个企业内横向生长。 --- ### [19:06](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1146s) 合规与治理：以欧盟高标准筑牢信任护城河 AI 治理与监管虽然听起来枯燥，却是战略落地的重中之重。当前美国在联邦监管层面整体偏向“自由放任（laissez-faire）”，这种宽松环境让企业能够迅速推进并建立先行优势。然而，欧盟（EU）的监管环境则截然不同且严格得多。 针对跨国开展业务或追求长期稳健的企业，演讲者建议在不严重拖慢业务节奏的前提下，主动向欧盟的合规标准看齐： - **建立内部信任**：遵循高标准合规要求能够显著提升员工和客户对 AI 系统的信任度。 - **应对政策收紧风险**：美国未来的监管法规极有可能逐步向欧洲靠拢并趋于严苛。如果从一开始就满足更严苛的标准，未来政策收紧时企业就能从容应对，立于不败之地。 - **低成本合规学习**：对于大多数企业而言，合规入门并不复杂，花一天时间通过 ChatGPT 等工具系统了解相关法案与合规框架，即可掌握绝大部分实操要领。 --- ### [20:26](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1226s) 组织变革与文化：20% 技术与 80% 人的协同 AI Agent 的落地与过往所有数字化转型项目一样，核心不在于技术本身，而是遵循“20% 取决于技术，80% 取决于人”的规律，甚至 AI Agent 变革对“人”的依赖程度超越了以往任何一次技术浪潮。如果仅有中层推动而忽视高层沟通，极易导致项目在汇报时因与高层战略脱节而“彻底告吹”。 ```mermaid flowchart TD accTitle: Organizational Transformation Pillars subgraph Pillars["组织变革三大支柱"] E["高管支持与赋能 (Executive Champions)"] T["全员培训与透明沟通 (Team Enablement)"] C["创新型组织文化 (Innovation Culture)"] end E --> C T --> C C --> O1["产品迭代更快 / 击败竞争对手"] C --> O2["摆脱内耗 / 准时下班"] ``` 要构建健康的变革生态，企业需要打通自上而下的三个关键支柱： - **高管层深度参与（Executive Buy-in）**：避免在项目最后关头给高层“突袭式”惊喜，而应尽早拉入高管作为项目冠军（Champions），确立顶层方向。 - **全员赋能与沟通（Team Enablement）**：AI 关乎员工的职业生涯与饭碗，领导层必须通过全员大会（Town Halls）、系统性培训与答疑，给予员工明确的安全感与信心。 - **文化进化（Cultural Evolution）**：当高管与基层达成协同，企业将演进为“创新赋能型文化”，如同配合默契的顶级乐队，员工不仅能高效打造更优产品、战胜竞争对手，还能免除繁重内耗、“开开心心准时下班”。 --- ### [22:01](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1321s) 复利效应与战略迫切性：曲速飞船与马车的代际竞赛 AI Agent 的部署具有极强的“复利效应（Compounding Returns）”，其发展逻辑与年轻时的养老金储蓄极为相似：初期投入看似微不足道、见效缓慢，但随着时间推移，收益将呈指数级爆发。不同之处在于，AI 的复利周期被极大压缩，演进速度远超传统认知。 ```mermaid flowchart LR accTitle: Compounding Speed of AI Agent Strategy A["0-90 天：首个 Agent 概念验证"] --> B["90-180 天：各业务线铺开 Agent"] B --> C["180 天+：多 Agent 协同生态"] C --> D["摆脱低效流程与决策疲劳"] D --> E["释放全员创新潜能"] E --> C ``` 如果在 90 天内打造出首个验证成功的 Agent，接下来的 90 天（即第 180 天）就能在所有纵向业务线中铺开，并最终构建出相互协作的多 Agent 生态网络。这一过程将员工从低效流程与“决策疲劳（Decision Fatigue）”中彻底解放，形成正向反馈闭环，推动企业飞速发展。 反之，若竞争对手率先完成了 Agent 生态布局而自身止步不前，双方的差距将迅速拉大为“曲速飞船（warp speed spaceship）”与“单马木车（horse and buggy）”的技术代差： - **不可逆的竞争鸿沟**：一旦落后，企业将彻底失去追赶的机会，被对手远远甩在身后。 - **企业存亡危机**：在这场巨大的经济与技术转型浪潮中，大批墨守成规的企业将面临倒闭破产。 演讲最后总结强调，入局 AI 并不需要一开始就进行庞大投资，关键在于“立即行动（Get started now）”。演讲者展示了联系方式（hello@polyrific.com）与现场展位信息，呼吁在场嘉宾把握战略窗口期，尽快启动自身的 AI Agent 布局。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=4s) 你们来对房间了，这正是我们想要服务的对象。 — Matt Cashatt > [01:43](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=103s) 它现在绝对不是终结者（Terminator）。 — Matt Cashatt > [03:12](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=192s) 等到明天绝对为时已晚。 — Matt Cashatt > [08:14](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=494s) 恭喜你们，你们和所有人一模一样。 — Matt Cashatt > [10:20](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=620s) 在 2025 年居然还需要人工阅读 PDF 再手动录入系统。 — Matt Cashatt > [10:20](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=620s) 再也不想做任何一个该死的决定了。 — Matt Cashatt > [16:43](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1003s) 亲近你的朋友，更要亲近你的敌人（Keep your friends close and your enemies closer）。 — Matt Cashatt > [20:26](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1226s) 20% 取决于技术，80% 取决于人 — Matt Cashatt > [20:26](https://www.youtube.com/watch?v=EufiQ6YjanQ&t=1226s) 开开心心准时下班 — Matt Cashatt ## Mentioned Matt Cashatt, Polyrific, Ai4, RAT 模型 (Reason, Autonomy, Task Execute), Verizon, Petrobras, Agent-to-Agent (A2A), Model Context Protocol (MCP), 《办公空间》(Office Space), TPS 报告, 欧盟 (EU), ChatGPT ## Video description Matt Cashatt, CEO, Polyrific Abstracts Ready to harness the AI disruption that 75% of executives believe will reshape the workplace even more profoundly than the Internet did? AI agents have reached a critical tipping point, moving from theoretical concept to practical reality, with 79% of companies already adopting them and reporting measurable productivity gains. With 88% of executives boosting AI budgets and 73% seeing competitive advantage within 12 months, the window for strategic advantage is now. You'll discover real-world case studies showing dramatic results, learn how to identify high-impact opportunities in your organization, and understand proven adoption strategies that start small but scale effectively while ensuring responsible AI deployment and real ROI. Attend to transform your uncertainty about AI agents into a strategic competitive advantage before your competitors do. [Source](https://www.youtube.com/watch?v=EufiQ6YjanQ)

[0:04] Matt Cashatt: 你们来对房间了，这正是我们想要服务的对象。 [1:43] Matt Cashatt: 它现在绝对不是终结者（Terminator）。 [3:12] Matt Cashatt: 等到明天绝对为时已晚。 [8:14] Matt Cashatt: 恭喜你们，你们和所有人一模一样。 [10:20] Matt Cashatt: 在 2025 年居然还需要人工阅读 PDF 再手动录入系统。 [10:20] Matt Cashatt: 再也不想做任何一个该死的决定了。 [16:43] Matt Cashatt: 亲近你的朋友，更要亲近你的敌人（Keep your friends close and your enemies closer）。 [20:26] Matt Cashatt: 20% 取决于技术，80% 取决于人 [20:26] Matt Cashatt: 开开心心准时下班

[0:04] This is what I like to see. This room is [0:06] packed. You people showed up. It's the [0:08] last session of the day. Some of them [0:10] are not so interesting and some of them [0:12] are pretty interesting, but happy hour [0:14] is next. So, that is going to be an [0:17] awesome act to follow me, right? Who's [0:18] excited for happy hour? That's all [0:19] right. Go ahead and clap. I'm excited [0:21] for it. I can't clap right now. Who's [0:22] excited to be in Vegas? This is pretty [0:24] great, right? Okay. All right. Good. So, [0:26] we have a few partyters. I'm with you [0:28] guys. After this, we we'll go to happy [0:29] hour together. All right. So, we're here [0:31] today to talk about AI agent strategy [0:34] and try not to make it too boring. This [0:35] is going to be higher level than some of [0:37] the talks that you all have been to. I [0:38] want to get to know my audience just a [0:39] little bit before I dive deep. How many [0:41] of you raise your hand if you don't mind [0:43] if the company you work for, your core [0:45] business is not AI or extremely AI [0:48] adjacent? All right, good. That's a lot [0:50] of you. That's then you're in the right [0:51] room then. That's what we want. Okay. [0:53] So, we're here to talk to you today. [0:55] Hey, we're here to help you understand [0:56] at a high level what you need to be [0:58] doing to um embrace AI and AI agency and [1:01] get them rolled out at your your uh [1:03] places of business, your your [1:04] enterprises. It's actually not as hard [1:06] as you think. It doesn't have to be as [1:07] painful or as technical. All of those [1:09] technical details if you find the right [1:11] vendor, we're one of them at Polyrific [1:13] has all of that abstracted for you. We [1:14] can explain all of it, but we can also [1:16] just get right to the point and help you [1:17] roll these things out and start winning. [1:19] And that's what we're here to do. So [1:20] again, I got a great introduction there. [1:22] Thank you for that. But again, my name [1:23] is Matt Cashad. I'm founder and CEO of [1:25] Polyriic. We've been around for a little [1:26] over a decade rolling out AI and [1:29] automation solutions for enterprises and [1:31] helping them win at that, create great [1:32] value. And pretty soon, we're going to [1:34] be rolling out quantum AI solutions as [1:36] well. So, that's the next big thing. You [1:37] guys think you're just barely getting a [1:38] handle on an AI agency? Quantum's the [1:40] next one for you to have to buckle up [1:42] for. So, get ready for that next. All [1:43] right. So, let's talk about real quick a [1:46] level set. I feel like you all have seen [1:47] a variation of this slide probably a [1:50] hundred times today. In fact, just give [1:51] me one second here. I took my jacket [1:53] off. Right. It's a reasonable thing to [1:54] do because I have agency. That's easy [1:56] for me. AI agents, we follow the acronym [1:59] RAT. They can reason. They have autonomy [2:02] within reason. And they can task [2:04] execute. I just got a little hot up [2:06] here, right? So, I took my jacket off. I [2:08] reason from the environmental data [2:10] around me that I'm a little more warm [2:12] than what makes me comfortable. So, I [2:15] had the autonomy to say within [2:17] guardrails, within guidelines. I had the [2:19] autonomy to say, I want to take this [2:21] jacket off so I'm not so hot. I couldn't [2:23] completely strip down, right? Because [2:24] that would be ridiculous and illegal. [2:26] Those are your guardrails. And then I [2:29] had the the ability to execute a task. I [2:31] took off the jacket. It seemed a little [2:33] silly and dramatic. Sure, but I was [2:34] actually pretty hot. So, it's hotter in [2:36] here than I thought it would be. Um, now [2:38] I got to find my clicker. Here it is. [2:40] Okay. So, if we just level set here, [2:42] that's what an agent is. It's not more [2:43] than that. It's not less than that. It's [2:45] really not complicated. It's just a [2:47] little piece of software that can make [2:49] sense out of the contextualized data in [2:51] its world. And it has enough autonomy [2:54] that you've granted it. You're in [2:55] control. We're in the driver's seat [2:57] still. I promise it's not the Terminator [2:58] yet. It has a little bit of autonomy to [3:00] make a decision. Once it makes that [3:01] decision, it can execute on that [3:03] decision. It can execute a task. So, [3:04] RAT, it's kind of a cheesy acronym, but [3:06] it'll help you remember. Reason, [3:08] autonomy, task, execute. Let's not make [3:09] it harder than that. [3:12] All right, let's talk about some [3:13] numbers. 79% of companies adopting AI [3:15] agents already. We are past early [3:18] adoption. By the way, for everybody [3:19] raising your cameras, feel free to do [3:20] that, but I'm happy to send you a copy [3:22] of this. There's going to be a QR code [3:23] at the end, and I think they put it up [3:24] with the uh the conference, too. I [3:27] always wish somebody would have told me [3:28] that when I was in the audience. Um, all [3:30] right. 79. Again, we're past early [3:32] adoption. We're in early major majority [3:34] days. So, the fact that you're in here, [3:35] if you're thinking about or have already [3:37] started your journey of implementing an [3:38] AI agent strategy, it's really, really [3:40] appropo right now to be doing that. this [3:42] is the time. Waiting till tomorrow is [3:44] way too late. I guarantee you. And we're [3:45] going to get into that in a little bit. [3:46] Twothirds of companies reporting [3:48] measurable productivity gains. That [3:50] means that the learning curve has [3:52] already been climbed. We're starting to [3:54] establish playbooks. Design patterns are [3:56] starting to get established for how we [3:58] do these things. And that's crucial for [4:00] you guys that are just getting started. [4:01] If that's where you are in the cycle, um [4:03] it means that you're not going it alone. [4:05] This isn't on the frontier. It's less [4:06] scary now, right? But there is a little [4:09] bit of criticality of time here as well. [4:11] You need to get in there and start [4:12] developing your own playbooks as well if [4:13] you want that competitive edge. 88% of [4:16] executives boosting AI budgets. That's [4:19] real value that CFOs are seeing. CFOs [4:21] aren't good at opening up the purse [4:22] strings, are they? They like to hold [4:24] those suckers tight. Believe me, I know. [4:26] Um CFOs are opening it up because [4:28] they're seeing real value, not hype, [4:30] coming out of AI implementations now, [4:32] especially AI agency. And then we have [4:34] 73% of CEOs seeing a competitive [4:36] advantage in the next 12 months by [4:39] rolling out an AI agent strategy. What [4:41] that means is if you aren't doing this [4:42] already, your competitors probably are. [4:44] And again, we're going to get into why [4:46] it's so critical that you get engaged, [4:49] that you get in there now and start [4:50] rolling out, even if it feels like a [4:52] small step that you're starting with, [4:53] which is fine. [4:56] All right, let's talk about a few [4:57] success stories and then we're going to [4:58] get into more of the strategy, but [5:00] Verizon seeing 40% in close rates for [5:02] their sales, enterprise channels, uh, [5:05] BTOC channels, devices, plans, all based [5:09] on rolling out AI agents that first [5:11] modeled their best salespeople. That's [5:13] that's kind of a little strategic tip [5:14] right there, but uh, more of a social [5:17] proof for this point in the slide deck. [5:18] That is that is what they did. They [5:20] modeled their best salespeople and [5:21] within within a year boosted close rates [5:24] to 40%. That's a pretty phenomenal [5:26] number because now they're automating [5:28] the best uh bestness goodness, whatever [5:31] it is that those best salespeople and [5:33] they are distilling it into an agent [5:35] that can repeat it. So that's pretty [5:37] pretty awesome. [5:39] Uh big uh Brazilian energy company Petro [5:42] found $120 million in tax savings within [5:46] three weeks of deploying their AI [5:48] agents. Now they're a big company to be [5:49] fair. They're dealing with billions and [5:51] billions of dollars, but their human [5:54] actors that were involved in devising [5:55] tax strategy can't possibly look at all [5:58] of that data and make sense out of all [5:59] those patterns all the time. They just [6:01] don't have enough people. And if they [6:02] did have that many people, they can't [6:04] all communicate in synchrony the way [6:05] that we would like for them to be able [6:07] to. That's where agents can come in to [6:09] great effect and and uh help augment [6:12] your team and lift up those tax [6:14] strategists. $120 million in three [6:16] weeks. Man, I'd like to make that kind [6:17] of money. Uh banks are using mortgage [6:20] processing agents now to cut down uh [6:22] loan approval times. I think I have a [6:24] little typo there. It's not uh loan [6:25] approval times by two plus days. It's [6:27] down to two days. That's that's how [6:30] quick it's getting. We're going to get [6:31] into a little bit of that in a minute, [6:32] too. Here's one of my favorites before [6:34] we kind of leave the social proofs here. [6:36] Things we've done at Polyriic. We work a [6:38] lot with insurance companies. We work [6:39] with a lot of industries all the way [6:40] from major motion picture studios to big [6:42] oil and gas companies and to insurance [6:45] companies, healthcare. We've done a lot. [6:46] But one of my favorite use cases is [6:48] insurance. If you don't know a lot about [6:50] the insurance industry, a lot of their [6:52] business model is high volume, low [6:54] margin. So, one of our clients, for [6:56] example, does property and casualty [6:58] insurance policies. They process [7:00] hundreds a day and they were spending [7:02] often spending more on their [7:04] underwriters time to review an [7:06] application than they were making on the [7:08] policy. They were losing a lot of money. [7:09] This is a big problem obviously, right? [7:11] So, kind of like Verizon did by by [7:14] mimicking and learning from what their [7:15] best salespeople can do, we trained uh [7:18] with the with the the buyin and the help [7:20] of their underwriters. We trained an AI [7:23] agent to basically pre-review these [7:26] applications for their underwriters and [7:28] put a cover sheet on it that says you [7:29] need to approve this application and [7:31] here's why you should deny it. Here's [7:33] why. Go to page 10, you're going to see [7:34] a big hole in the roof. Yeah, we're [7:35] analyzing photographs, too, from [7:37] inspection reports. We're analyzing [7:38] handwriting in the margins. We're [7:40] analyzing and catching things that [7:41] sometimes the underwriters miss. And [7:43] we're putting all that on a cover sh uh [7:45] page and about two or three lines that [7:48] they can go through and less than five [7:50] minutes to make a qualified decision [7:52] about that package about that [7:54] application. So that's huge for their [7:56] bottom line. So if you have an [7:58] imagination, and I'm sure you do, you [7:59] can extrapolate from that what we could [8:01] probably do for your industry as well or [8:03] what agents can do for your industry. Um [8:05] property management, same thing. cutting [8:07] down response times for for scheduling [8:09] in maintenance um scheduling parts, [8:11] scheduling repairs and other issues that [8:14] uh that some of our property management [8:15] clients have had across their [8:16] properties. Um again, AI agents to the [8:20] rescue in a big way. Um this slide's a [8:23] little boring, but it's uh but it's [8:26] telling. I guess the thing that's [8:27] important to take away from it is the [8:29] best AI integrations, the best AI agent [8:32] integrations from an IT perspective [8:34] actually are kind of boring. And that's [8:35] to your advantage. You know why? We're [8:37] not changing your ERP. We're not [8:39] changing your CRM. We don't need to do [8:41] any of those things. If you've been in [8:42] this room or some of the other rooms [8:43] around here listening to these talks, [8:45] you've probably heard a lot today about [8:46] agent to agent A2A or MCP, model [8:49] context, protocol. These are the bits of [8:52] technology that have they're kind of [8:54] like a Rosetta Stone for translating all [8:56] the crazy data you have. We've built our [8:59] own at Polyriic, but we also use others [9:00] as well. We leverage anything that's [9:02] going to get you a win. That's what we [9:03] do with our platform. Um, but we [9:06] basically leverage what you have in [9:08] place. This is the great thing about AI. [9:10] It's not just unique to Polyrific, but [9:12] everybody's kind of doing it this way or [9:13] should be doing it this way. Now, we're [9:15] coming to meet you where you are. [9:17] Because if you're sitting there thinking [9:18] right now, geez, our our data is kind of [9:20] scattered. A lot of it's parked in Excel [9:22] spreadsheets here and there. None of [9:23] it's clean. Congratulations. You're just [9:26] like everybody else. You're just like [9:27] everybody else. Some of our biggest [9:29] clients who are really proud of their [9:32] data lake, their data warehouse, they're [9:33] now their data ribbon actually have some [9:35] of the messiest closets that I've seen. [9:37] So like it's everywhere. Don't be [9:39] intimidated by the state of your data. [9:41] Just get to work. We we got you. We got [9:43] you. Got the tooling for this. Sometimes [9:45] to to accomplish that, we build that [9:47] integration middleware. It's a usually a [9:50] light little bit of software that we [9:51] need to build to tie together some of [9:52] your systems. It's not always necessary. [9:54] Sometimes it is, but that's how we help [9:56] you get the most out of it. But the [9:57] point here is that it's not that [10:00] difficult to deploy AI agent strategies. [10:03] They go on top of your existing system. [10:05] They don't replace it. [10:08] I think I had something in my notes too [10:09] which I found out late. We can't have [10:11] speaker notes up here uh you know on a [10:13] screen that I had something about give [10:15] your agents only read it only access. [10:17] Makes sense, right? Common sense. When [10:18] you get started, start with readonly [10:20] access. Okay. So we've deployed a lot of [10:24] AI agent solutions and it's led to a lot [10:26] of successes and sometimes some [10:27] failures. We learn from all of it, [10:28] right? That's that's that's the case [10:30] with everybody. What we have discovered [10:31] is that efficiency gains, the bottom of [10:33] that value pyramid right there needs to [10:35] become your table stakes. What you're [10:37] going to do is when you go back to your [10:38] places of business, you're going to look [10:40] around at the processes that are slow [10:43] and inefficient and just suck. I mean, [10:46] sorry for the language, but like there's [10:48] we all have them. You're going to look [10:50] around. You're going to find those [10:51] first. I don't want to have people [10:54] spending time reading a PDF and [10:57] re-entering the data. Why would we do [10:59] that? It's 2025 for crying out loud. [11:01] That's where you start first, efficiency [11:03] gains. But there's a strategy here. You [11:05] start with the efficiency gains because [11:07] that's going to affect the most people [11:09] in your organization and it's going to [11:12] buy you the most confidence and that's [11:14] going to start turning resistance into a [11:16] tailwind for you, which is what you need [11:18] to do moving forward. Then you're going [11:20] to start getting to this next part, [11:22] which I I love. How many of you, by show [11:25] of hands, go home at the end of the day [11:27] or you get done with your workday and [11:29] you're just mentally drained? You're [11:30] just kind of tired. You don't know why. [11:32] You're like, "Man, I've been sitting on [11:32] my butt all day. Why am I so tired?" [11:35] Nobody feels like that. I do. Okay. I [11:38] feel like that all the time. My wife and [11:40] I have a standoff over what are we going [11:41] to make for dinner? I don't know. You [11:42] decide. I don't know. You decide. I [11:43] don't want to make another freaking [11:44] decision. So, that's decision fatigue. [11:46] What I'll challenge you all to do the [11:48] next time you go into your office, the [11:49] next time you're sitting down, put a [11:50] sticky note on your desk, and every time [11:52] you have to make some kind of decision, [11:54] I don't care it's whether to send that [11:55] email to your boss or to to pick a [11:57] certain kind of sandwich for lunch, make [11:59] a little tick mark on that note. I [12:00] promise you, you're going to run out of [12:01] space by the end of the day. We make [12:03] hundreds of decisions and micro [12:05] decisions every day, sometimes [12:06] thousands, and it is draining. It's [12:09] draining. But the key is in the age of [12:12] AI agents, we don't have to do that [12:14] anymore. you can defer certain decisions [12:17] certain decisions to the AI agent so [12:20] that your people aren't so drained. And [12:22] this is going to get to my favorite part [12:23] and it's going to maybe get a little [12:25] weird. I don't know. I'm really [12:27] passionate about what we have between [12:29] our ears. I've been fascinated by AI for [12:32] as long as I can remember and I've [12:33] become an expert in it at least in the [12:35] past decade. I've been developing those [12:37] expertise for 20 years. I'm the type of [12:39] person who dives really really deep and [12:41] gets really really interested in certain [12:43] things and AI was probably the biggest [12:45] in my lifetime. The deeper I got in [12:48] learning about what AI is and how it [12:49] works and how artificial neural networks [12:52] work and how they they simulate to the [12:54] best of our knowledge what our human [12:55] brains do, the more I had to go learn [12:57] what the human brain did. And I started [12:59] to realize this beautiful thing in my [13:01] opinion and a lot of Silicon Valley [13:02] would disagree with me. Super [13:04] intelligence or what they call [13:05] artificial super intelligence ain't [13:07] coming. Not in our lifetime. It's a [13:09] different fundamentally different kind [13:11] of intelligence. Ours is human [13:12] intelligence which is something [13:13] different. We can imagine what [13:15] forward-looking looks like. We can [13:17] imagine a future. AI can't do that. It [13:21] can simulate or regurgitate what an [13:23] imagined future was that was fed into [13:25] its system, but it can't really picture [13:27] something that isn't there in a deep [13:29] blue ocean right now. That's something [13:30] that's unique to us. It's something that [13:32] makes us special. So if you get through [13:34] those first two things, you free up [13:36] people's time, you get them less fatigue [13:38] from from uh uh you know, enhanced [13:40] decision-m, less decision fatigue, you [13:43] are going to enable the tip of that that [13:45] pyramid, the tip of the spear, [13:46] innovation enablement. It's kind of a [13:49] businessy term, but let's put it in in [13:51] in kind of terms that probably more of [13:52] us can relate to. Do any of you remember [13:54] in a kid being a kid and you had a [13:56] hundred happy, good, excitable ideas [13:59] before breakfast and then the world [14:02] slowly ground all that out of you? Oh, [14:05] you got to prioritize. You can't you're [14:07] you're off in the clouds. You got, you [14:08] know, blah blah blah blah blah. Teachers [14:10] do it. Everybody does it. I've done it [14:11] to my kids all the time. But the fact of [14:14] the matter is those idea that raw [14:17] creative power that we all have is what [14:19] makes us awesome as a species. And AI is [14:23] actually going to lift that up and and [14:25] uncover that again from us. It can it [14:27] can do that. And this matters in your [14:29] place of business. I'm not talking about [14:31] freeing up your your your team members [14:33] to draw a better picture unless that [14:35] picture is part of your core business. [14:37] I'm talking about when they are free to [14:39] be creative and it's this is a cultural [14:40] movement too. It takes a little time, [14:42] right, to to get their confidence. When [14:44] you have freed them up from having to [14:46] make to to to deal with inefficient [14:48] drudgery processes and to have to make [14:51] decisions that are relatively [14:53] inconsequential all day long, they're [14:55] going to start coming up with better [14:56] ideas, better products, ways to create [14:59] more value for your customers. And that [15:01] creates a feedback loop where you start [15:03] winning and winning and winning. And the [15:04] more you're winning, the more ideas are [15:06] rolling in, the more encouraged your [15:08] team's getting. And that's why you also [15:09] don't get rid of your people. I'm sure [15:11] that some of you have members of your [15:12] team that you could do with or without, [15:14] right? But the fact is all of us have [15:16] good ideas. Everybody does. They're in [15:18] there. They're just buried and they're [15:19] blocked for certain reasons. AI is like [15:22] a fracking fluid to get those ideas out. [15:24] So if you deploy it correctly, that's [15:26] what you can expect. [15:28] All right. Um I'm an impatient guy. I've [15:31] worked my whole life to try to become [15:32] more patient. It doesn't take. Um so I [15:35] like to do things in short spurts. I [15:36] mean, and and again, just quick raise of [15:38] hands. How many of you would like to see [15:39] results in 90 days instead of three [15:41] years? Thank you. I mean, I knew I [15:44] wasn't alone. My mom always told me I [15:45] was too impatient and she's wrong. She's [15:47] still wrong. Okay. I love her. Um, [15:51] first 30 days, what you need to do is [15:53] you need to when you get back to your [15:54] offices, go walk your operations either [15:56] figuratively or literally if you can. If [15:59] you're in a single vertical, just do go [16:01] up and down that vertical. If you're [16:02] managing across verticals, walk across [16:05] all those verticals. Get to know the [16:07] people. Find the one common pain point [16:10] that nobody can stand. The one process [16:12] that everybody loathes and that's the [16:14] one you're going to fix first. How many [16:16] of you guys remember the movie Office [16:17] Space? Okay, you remember the TPS [16:19] report? You know, the manager's walking [16:21] around telling you TPS report. [16:23] Everybody's rolling their eyes. Find [16:24] that one. Find the TPS report. Once you [16:26] find the TPS report, that's the one you [16:29] bring as many people, as many voices [16:31] into to start solving that problem to [16:32] designing your agent. Use a good [16:34] platform like our inferring platform. [16:36] and we can crank those agents right out [16:37] for you. But you find that problem. The [16:40] reason why you're going to do that, the [16:41] strategic advantage of doing that is [16:43] because now you have galvanized and [16:46] deputized perhaps hundreds of people [16:49] across verticals in your company to be [16:51] fans of what you're doing. That's going [16:53] to start eliminating all kinds of [16:55] barriers for you to move forward with [16:56] your agent strategy. So it's really [16:58] important that you do that in the first [16:59] 30 days. Now, the next 30 days, the 31 [17:02] to 60 there, pilot and learn, [17:05] we can do better than that. It's more [17:06] than piloting and learning. This is [17:08] where you're going to make your [17:09] mistakes. You're going to try to roll [17:10] things out and fits and starts. I mean, [17:12] let's be honest about it. You're going [17:13] to launch it. It's not going to work [17:14] that well. You're going to have to pull [17:15] it back, reconfigure, go back to the [17:17] drawing board, launch it again. But [17:19] throughout this process, what you have [17:21] to do is communicate, communicate, [17:22] communicate. Own your failures. you [17:24] know, let the stakeholders know that, [17:26] hey, this fell down and here's why and [17:28] and we're aware and here's how we're [17:29] fixing it. That's really really crucial. [17:31] That's going to lead to the next point [17:33] in this this middle tranch of 30 days is [17:36] find the people who are most critical of [17:38] what you're doing and bring them the [17:39] closest. You remember the old saying, [17:41] keep your friends close and your enemies [17:42] closer. We've turned some of our most [17:45] fierest critics into our most fierest [17:48] advocates through AI implementations [17:50] that we've involved them. And the reason [17:52] why we do that is because their opinions [17:55] are usually correct. They're they have [17:58] they're holding out for a lot of [18:00] different reasons that actually matter [18:01] whether you think they do or not. Some [18:03] of them are basic. They're concerned [18:05] about security. You got to answer those [18:07] questions. What if there's a blind spot [18:08] you missed? They just helped you uncover [18:10] it so you can build a stronger agent. [18:12] What if uh they're concerned about [18:14] losing control? That's a valid concern. [18:16] How do we how do we mitigate that? These [18:18] are things that you have to face and [18:20] think about. It's not something you can [18:22] dodge and ignore. And then, of course, [18:23] they're going to be worried about losing [18:24] their job, too. It's okay for us to [18:26] acknowledge that's a risk in what we're [18:28] doing. I have a personal theory that [18:30] we're better off if we if we combine [18:32] forces with AI and that there's no [18:34] reason to eliminate a job, but that's [18:35] not always going to be the case. So, you [18:37] have to answer that. You have to put [18:38] their mind at ease, you know, and and [18:39] and mitigate that together. When you do [18:41] that, you're going to galvanize them to [18:43] be some of your best advocates. Now, you [18:45] have even more of a tailwind. That last [18:48] 30 days is when you are going to take [18:50] what you've done, everything you've [18:51] learned. It's the learning that you're [18:52] going to scale out. But you can also [18:54] replicate these agents across different [18:57] verticals and in different problems that [18:59] you have um to make something that that [19:02] is is truly amazing and starts growing [19:04] horizontally across your organization. [19:06] I'm just going to touch on this because [19:07] it's it's boring, but it's important. Um [19:11] it is important. I mean, it's really [19:13] important. It is. No, it is. It is. So [19:17] we we have we don't have that much [19:19] governance or or much federal regulation [19:21] here in the US yet we as far as it goes [19:24] it's pretty lazy fair right um here in [19:26] the US and we're lucky for that it [19:27] enables us to get ahead quickly there's [19:29] strategy behind that as well in the EU [19:31] it's a big different story right now so [19:33] if you do business in the EU one of the [19:35] challenges that I will will put in front [19:37] of you and even if you don't perhaps [19:39] within reason without slowing yourselves [19:41] completely down try to live up to their [19:43] standards because it's going to build [19:45] more trust among your people if you can [19:47] do that and then by the time you launch [19:49] something that's making you super [19:51] successful your AI agents that are [19:53] rolling out and just doing really really [19:54] well regulation could change here in the [19:57] US and become more restrictive not less [19:59] restrictive right so they could catch up [20:00] to what Europe's already doing and if [20:02] you're already in play with the European [20:05] the the more severe regulations then [20:08] you're fine so if you can stomach it try [20:10] to learn a little bit spend a day go and [20:12] chat GPT learn what you can about those [20:14] because for most of that's actually all [20:16] you need to do, you know, for what [20:17] you're doing at work. So, learn about [20:19] those and learn how to be in compliance [20:21] with those and then Bob's your uncle. [20:23] You're okay. Um, that's important. So, [20:26] um, this is I'm not going to spend a lot [20:28] of time here because it's really like [20:30] something you've heard with every other [20:31] transformation project. Adoption isn't [20:34] just about the technology. It's 20% [20:35] technology, right? And then 80% people. [20:38] And that is more the case with AI agents [20:40] than any kind of transformation thrust I [20:42] can remember in my career. Um and and so [20:46] yeah, you got to get the executive buy [20:47] in. I've we've had implementations where [20:49] we've had kind of like middle management [20:51] that wants to hold it back from [20:52] executive management as we're going [20:54] through it. We get to the whim, they [20:55] take it to executive management, it [20:56] falls flat. It happens. And it's because [20:59] there was, you know, policy making [21:00] happen from the top that they weren't [21:01] aware of. I understand the the need to [21:04] get some of your ducks in the row before [21:05] you go to the executives, but you know, [21:07] resist the urge to keep them completely [21:08] blindsided and then then show them at [21:10] the end. Um, get your team involved [21:14] early on. Offer a lot of training. Do [21:16] town halls. Answer question. Answer [21:17] concerns. Make sure leadership is [21:19] involved in this because if they're not [21:21] giving confidence to your people about [21:23] where they're taking you with AI, this [21:24] is their careers you're messing with, [21:25] right? So, you have to do that. But if [21:27] you enable the team, if you get the [21:29] executives bought in as champions, [21:32] that's going to create that third [21:33] pillar, the culture, cultural evolution. [21:36] Now, you've become this innovation [21:37] enabled. remember the tip of that [21:39] pyramid culture that is outstanding at [21:42] working together. You're jelling. You're [21:44] you're just like an awesome band. [21:45] Everybody's doing their part. Beautiful [21:47] music is starting to happen. Great [21:49] things are starting to happen. You're [21:50] beating your competition. You're coming [21:52] up with better products. You're coming [21:54] up with better ways to do things. And [21:55] people are generally happier and they're [21:56] going home on time. That's huge. That's [21:59] that's the culture you want to build. [22:01] This is my second to last slide, I [22:03] promise. Almost happy hour time. Um, you [22:06] know, like this slide didn't really do a [22:08] great job of communicating what I want [22:10] to communicate here. So, yeah, here's a [22:11] cool timeline, but what I want to [22:13] communicate is [22:15] when we were all younger, I know that we [22:17] all got told in school or whatever it [22:19] was, hey, you need to invest early. Put [22:21] put money in in your savings account. [22:23] You're going to get interest and it's [22:24] not going to do much and it's going to [22:25] sit there and then it's going to [22:26] compound at the end. And I never had the [22:28] patience for that until I got a little [22:30] bit older and I started looking at the [22:32] business end of retirement at some point [22:34] in time because at some point we all [22:36] have to retire, right? Like, oh man, [22:38] like I really hope I saved enough when I [22:40] was younger and and I did, you know, it [22:42] wasn't a lot, but it it was enough and [22:44] it's starting to compound now. And [22:46] that's cool. AI agent strategies and [22:48] rolling out an AI program is couldn't be [22:50] more alike. The difference is it's going [22:52] to move a lot faster. So, right now [22:55] we're talking about building your first [22:56] agents probably for a lot of you. If [22:58] your competitors are already doing that, [22:59] here's what happens. You build the one [23:01] agent proof today, right? Say in 90 days [23:04] if you can roll it out and you can we [23:05] can help you do that all day long. Then [23:08] you're going to start building agents [23:09] for all your verticals tomorrow, right? [23:12] In your next 90 days when you get to [23:14] 180. Now, you're going to start building [23:16] ecosystem of agents that work together [23:18] and start accomplishing things that you [23:20] can't even imagine right now. Now that [23:22] starts moving you faster and faster and [23:24] faster. Remember that pyramid that we [23:25] talked about? You're freeing people from [23:27] inefficient processes. You're freeing [23:28] their time up and their focus. You're [23:31] freeing them from decision fatigue. [23:33] You're enabling that innovation talent [23:35] that we all have. It becomes a feedback [23:37] loop where you move faster and faster [23:39] and faster toward success. So if your [23:43] competition's doing that and you aren't, [23:45] what happens is they're in a warp speed [23:48] spaceship and you're in a horse and [23:49] buggy. they will kick your they they'll [23:53] kick your butt, right? Like like they're [23:54] going to go past you, way past you. [23:55] You're not ever going to be able to [23:56] catch up. Businesses will fold in this [23:58] huge economic transformation with AI [24:00] that we're going through right now. Some [24:02] businesses will go under. A lot of them [24:03] will go under. You don't want to be in [24:05] those ranks. So, you get started now. [24:07] That's how you start investing. Doesn't [24:08] take a lot. Just like with your [24:10] retirement, doesn't take a lot, just a [24:11] little. And then it's going to have [24:12] compounding returns that are going to [24:14] come really fast. All right, that's it [24:15] for me. That QR code right there will [24:17] get you to our website. I think you can [24:19] reach out to us at helloprific.com. I [24:22] will be around after this talk and we [24:23] are happy to help you. We also have a [24:25] booth so come by and see us. Thank you [24:26] very much everybody.