HRTech Leaders Deploy AI Agents to Automate Recruitment Workflows

HRTech领袖部署AI智能体自动化招聘流程 企业采用HR科技的关键在于干净的运营数据与针对性的流程设计，而非宽泛的自动化功能。 与会嘉宾探讨了AI在HR科技中的实际应用与整合，强调自动化与预测模型如何在不牺牲人工监督的前提下提升候选人评估和内部运营。他们讨论了自建定制AI基础设施与采购现成软件之间的战略抉择，以及维护招聘诚信与数据规范的重要性。讨论还重点指出，应赋能非技术员工构建定制AI工具，同时确保稳健的流程设计，并专注于员工留存等高影响力的应用场景。 自动化与预测模型可在保持人工监督的同时，提升候选人评估与内部运营效率。 企业在自建定制AI与购买现成软件之间进行战略权衡，并强调招聘诚信与数据规范。 赋能非技术员工构建定制AI工具，同时需注重稳健的流程设计并聚焦员工留存等关键场景。

HR and technology leaders argue that artificial intelligence should serve as an adaptable delivery mechanism that supports human-centric work rather than replaces it, emphasizing that successful adoption depends on rigorous process design, targeted use cases, and strict human oversight. Human-Centric Delivery Across Varying Adoption Stages (2:05) BambooHR's Kristie Rowley observes that while HR professionals face mounting 'AI fatigue' following industry-wide hype, organizations sit across a wide spectrum of maturity—from introductory experimentation to operating fleets of autonomous agents. She argues that software providers must meet organizations at their current stage rather than bolting on superficial chatbots or forcing full-scale automation. In Rowley's view, AI alters the mechanism through which software is delivered, but it does not change the core human-centered purpose of HR functions. Safeguarding Hiring Integrity Through Human Oversight (3:31) HireVue CTO Alberto Silveira warns against sacrificing hiring integrity and candidate career outcomes for deployment speed. Because securing employment represents a critical personal milestone, Silveira maintains that recruitment decisions should not be abdicated to automated systems without human intervention. Grounded in HireVue's operating principle—'We don't guess, we validate'—he contends that AI should assist in objectively verifying concrete candidate skills while keeping human recruiters in ultimate control, rather than chasing unvetted, novelty features. Enabling Domain Experts to Build Custom Tools (6:16) Bryan Hong outlines Astranis's shift toward custom-building internal AI infrastructure, moving from early ChatGPT tests to deploying agentic swarms across General and Administrative functions within 18 months. Hong explains that modern AI tools enable talent operations specialists—who already understand core business processes—to build functional solutions directly rather than waiting on IT queues. Astranis filters these automation requests by evaluating frequency and judgment risk, targeting high-volume daily tasks while keeping complex, irreversible decisions under human discretion. However, opening AI development to non-engineers introduces operational hurdles. When non-technical personnel build custom applications, they often encounter the difficulties of software reliability, system architecture, and ongoing production support. To manage this, specialized engineering teams act as advisors, guiding approved employee projects through deployment and maintenance. Process Hygiene and Targeted High-Impact Deployment (42:27) The panel emphasizes that deploying AI cannot compensate for broken operations or disorganised data. Because AI acts as an amplifier of existing workflows, applying autonomous agents to poorly structured tasks only compounds inefficiencies. Organizations must prioritize workflow redesign and data hygiene as essential prerequisites before introducing automation. To capture tangible value, the speakers advise leaders to avoid broad, superficial rollouts that turn AI into an unproductive playground. Instead, organizations should concentrate deeply on one or two specific, high-friction operational challenges, such as performance review cycles or employee retention strategies, tailoring specific models to solve concrete business problems. The discussion concludes that AI cannot autonomously correct underlying operational flaws or weak data quality. The sustained challenge for HR leaders lies in balancing internal experimentation with production-grade engineering discipline while ensuring critical personnel decisions remain anchored in human judgment.

来自 HireVue、BambooHR 与 Astranis 的人力资源技术领导者指出，人工智能在人力资源领域的价值不在于盲目叠加代理或盲目追求自动化速度，而在于以人为本的技能验证、严密的底层流程设计与针对高频业务痛点的纵深落地。 技术定位：以人为本避免盲目跟风 (2:05) 在人工智能技术快速演进并引发业界“AI 疲劳”的背景下，BambooHR 的 Kristie Rowley 指出，不同客户的采纳阶段跨度极大，涵盖从基础模型实验到运作自主代理集群的各类组织。Rowley 认为，软件必须契合客户所处的实际阶段，而非强行提高系统复杂度；企业不应为了追求“AI 原生”标签而机械地拼接通用聊天机器人或自主代理。她强调，AI 仅是软件交付的新形式，人力资源系统满足个体需求、自适应服务员工的核心目标并未改变。 招聘底线：技能严谨验证与人类主导 (3:31) 针对人才招聘与筛选场景，HireVue 首席技术官 Alberto Silveira 强调，找工作是求职者个人生涯中的重大节点，企业绝不能为了追赶技术上线速度而牺牲招聘流程的公信力或求职者的职业前景。HireVue 主张“不靠猜测，依靠验证”（We don't guess, we validate），明确将 AI 定位为辅助与技能核验工具，坚决反对在缺乏人工介入的情况下过早将决策权完全移交给自动化程序。行业管理者应保持谨慎，保留人机协作（Human-in-the-loop）监督机制，而非被新奇的市场卖点牵着走。 自研策略：业务专家主导构建与运维协同 (6:16) 在“自研还是外购”的路径选择上，Astranis 的 Bryan Hong 分享了其在 18 个月内从 ChatGPT 基础测试拓展到整合学习管理系统（LMS）、定性分析与代理集群的经验。Hong 指出，精通业务流程的人资与组织运营专家如今能够直接主导定制工具的开发，缩减了对传统 IT 排期的依赖。然而实践也表明，非工程人员在构建工具后往往会直面系统维护、宕机预防与生产可用性等工程现实，因此组织需要组建专门的工程支持团队，在管理层立项后提供架构设计与生产部署指导。 落地前提：流程治理与聚焦核心场景 (42:27) 多位专家一致强调，在复杂业务中引入 AI 前，严密的流程设计与数据质量治理是不可或缺的基础。AI 仅会放大现有流程的效率或缺陷，如果基础数据混乱或工作流缺乏结构，直接引入智能代理只会放大错误结果。与其全面铺开进行表面化的实验并陷入无明确回报的“技术游乐场”，管理者应挑选一到两个具备高业务影响力的核心场景（例如员工留存策略或绩效评估机制）进行深度挖掘，精准匹配模型与系统架构以产生可衡量的商业价值。 讨论总结表明，HRTech 的未来竞争重点正在从单纯的工具堆砌转向严密的流程工程与高标准的工程运维支持；但在非技术人员广泛参与工具构建的趋势下，企业如何在保障数据卫生、规避自动化偏差与维持合规工程治理之间达成长期平衡，仍是各方需要持续探索的实践课题。

## Background 本视频由 Ai4 于 2026 年 8 月 22 日发布在 YouTube 平台上。根据页面简介，本次研讨会邀请了多位来自 HireVue、BambooHR 等企业的 HRTech 与 AI 领域专家，共同探讨人工智能如何嵌入到招聘、入职和留存等环节，以及如何通过预测性分析与生成式 AI 提升员工体验并应对合规挑战。 ## TL;DR 研讨会深入探讨了 AI 在现代 HR 系统中的多场景应用与挑战。HireVue 介绍了 AI 面试官如何应对海量大规模招聘，并解决求职者使用 AI 工具应试引发的“作弊与真实性”困境；Astranis 分享了通过自动化工具将通用与行政部门生产力提升 30% 的经验，并剖析了传统 HR 系统循环依赖的底层数据架构瓶颈；与会专家一致认为，数据与具体业务上下文（Context）的结合是释放 AI 创新价值的关键，且 AI 工具正从大型 enterprise 快速普及至 SMB 及企业内部全员。 ## Atmosphere & style 整场研讨会气氛专业而务实，嘉宾们摒弃了空洞的营销噱头，以业内前沿实践者的视角展开深度对话。现场既有对“计算机与计算机对讲”等行业奇观的幽默坦诚，也有对底层技术架构与合规痛点的冷静剖析。嘉宾结合自身的真实职业转型与项目经验，传达出一种既积极拥抱技术变革、又保持理性与务实的严谨态度。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=4s) HRTech 正经历从传统手动工作流向具备预测与自适应能力的智能化平台演进，其核心在于以更高的个性化服务于“以人为本”的行业本质。 - [20:31](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1231s) 在大规模招聘场景中，AI 面试官能够在两周内对超过 10 万名求职者进行扩展与相对公平的评估筛选，极大地提升了匹配效率。 - [20:31](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1231s) 求职者广泛使用 AI 工具引发了新型“作弊”现象，导致“计算机对讲”的技术伪装，如何识别候选人的“真实自我”成为当下的技术攻坚重点。 - [20:31](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1231s) AI 评估不仅能加速求职者与原申请岗位的对接，还能通过跨岗位分析发现求职者的隐性技能，实现“异岗匹配”。 - [26:06](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1566s) 企业在招聘自动化方面表现审慎，主要是出于严格的合规与资格审查要求，避免盲目进行“完全自动化”。 - [26:06](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1566s) 市场上对“人才工程师”与“前沿部署工程师”的需求迅速增长，这类既懂领域业务又具备 AI 工具开发能力的复合型人才正推动部门生产力基线显著提升。 - [26:06](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1566s) 传统 HR 系统（如 Legacy ATS）在底层数据架构设计上缺乏扩展性，存在复杂的循环依赖问题，严重限制了高阶 AI 与分析 Dashboard 的集成。 - [26:06](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1566s) AI 能够解锁离职面谈、同行评估等海量非结构化定性数据，帮助 HR 跨越多年维度深度挖掘复杂的深层管理与组织问题。 - [31:35](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1895s) 在 HR 领域应用 AI 的两大核心是数据与业务上下文；缺乏上下文的 AI 仅是个花式模式识别器，只有注入丰富上下文才能产生创新价值。 - [31:35](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1895s) AI 不仅弥补了人类考官连续面试数十人后的认知与记忆极限，更作为高效的“思考伙伴”大大降低了个人技术学习与转型的门槛。 - [31:35](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1895s) 虽然个人可以通过 AI 在短时间内构建自动化流程，但网络上宣称“几小时做出的 AI 能运行整家企业”是夸大不实的谎言，企业级落地需克服诸多现实复杂性。 - [39:48](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=2388s) AI 的适用范围已打破传统门槛，从大型 Enterprise 全面覆盖至 SMB 中小企业，全员赋能非技术岗位使用 AI 工具将激发出巨大的自下而上创新力。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=4s) 研讨会嘉宾致辞与 HRTech 的行业新前沿 研讨会由多位专注于人力资源与人工智能领域的行业专家拉开序幕。首先发言的是来自 HireVue 的首席技术官（CTO）Alberto Silveira，他指出 HireVue 在当前这波 AI 浪潮爆发之前就已经在该领域耕耘了二十余年，公司核心业务在于提供技能评估与验证的平台。紧接着发言的是 BambooHR 的 AI/ML 工程高级经理 Kristie Rowley，她负责 BambooHR 内部所有 AI 项目的落地与推进，致力于打造现代化的 HR 软件。最后发言的是来自于旧金山卫星制造初创公司 Astranis 的 People Ops 与 AI 服务总监 Bryan Hong，他不仅负责整体 G&A（通用与行政）职能部门的运营，还主导着公司内部 AI 方案的跨部门部署。 随后，Kristie Rowley 分享了她近期参加 SHRM（美国人力资源管理协会）大会的切身感受。她提到，对于 HR 专业人士而言，现在正是一个令人无比兴奋的时代，尽管很多人可能已经对纷至沓来的“AI”概念感到些许审美疲劳。然而，HR 与 AI 的融合正在创造出全新的前沿领域——HR 本身是一个高度“以人为本”而非“以计算机为中心”的行业，而 AI 的真正使命恰恰在于实现更高的个性化与适应性，去更好地服务于它所面对的人类。因此，如何让 AI 与 HR 完美契合，并让技术 ### [20:31](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1231s) AI 面试官与大规模招聘：公平性、真实性与岗位匹配 Alberto Silveira 详细介绍了 HireVue 如何将 AI 技术嵌入到大规模招聘与筛选流程中。为了应对高难度的招聘挑战，HireVue 推出了名为“AI Interviewer”（AI 面试官）的功能。在许多大型企业客户的招聘活动中，经常出现需要在短短两周内筛选并面试超过 10 万名求职者的极端场景。Alberto 强调，若单纯依赖人力，在如此紧迫的时间内保障筛选的公平性并挑选出最佳人选“几乎是不可能的，或者说就是不可能的”。 然而，AI 的普及也引发了全新的作弊与虚假表现问题。随着候选人广泛使用各种 AI 助手的工具与 Bot，很多求职者“一夜之间都变成了专家”。这导致实际面试中甚至出现了“电脑与电脑对话、人类在中间假装”的荒谬现象。Alberto 坦言，如何透过这些技术伪装去发现候选人真正的“真实自我（true self）”，成为了 HireVue 当前技术攻坚的核心重点。 针对招聘覆盖面的问题，Alberto 指出 AI 模型和算法能够以更公平、更具扩展性的方式评估每一位投递者。针对合作时间较长的客户，HireVue 还会根据其独特需求定制专属模型，这种深度个性化正是其竞争优势所在。AI 技术的引入极大地加快了用人单位与匹配人才之间的对接速度。此外，AI 还能实现意想不到的“异岗匹配”——当求职者申请岗位 A 时，系统分析其评估数据后可能发现其更适合岗位 B，从而帮助求职者开启此前未曾关注的职业机会。 对于“AI 是否在剥夺就业岗位”的担忧，Alberto 表示，上一场研讨会也提及了这一话题，但他认为 AI 的本质并非“消灭”岗位，而是“转型”岗位。虽然部分传统职位可能随时间推移被淘汰，但更多全新的工作岗位正在被创造出来，核心在于适应与转变。 * **大规模招聘规模**：某些客户需要在 **2 周** 内对超过 **100,000 名** 候选人完成面试筛选。 * **面试技术组合**：结合了动态按需面试（On-demand interviewing）、AI 自动化面试以及人类考官的实时面试。 * **AI 面试困境**：候选人使用 AI 工具与 Bot 辅助应试，导致面试沦为“计算机与计算机对讲”的奇特现象。 * **跨岗位智能推荐**：系统能根据评估数据，自动将不适合岗位 A 的候选人精准推荐至其未曾发现的岗位 B。 --- ### [26:06](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1566s) 自动化合规考量、前沿工程角色与底层数据架构的技术瓶颈 Bryan Hong 分享了 Astranis 在 Recruiting 与 People Ops（人力资源运营）流程中应用 AI 的实际情况。他指出，作为一家特定领域的初创公司，Astranis 在自动化招聘方面表现得较为审慎，避免进行“完全自动化”，主要原因是为了满足严格的资格审查与合规要求（Compliance）。 在探讨 HRTech 的岗位变迁时，Bryan 提到市场上对“Talent Engineer”（人才工程师）、“People Engineer”（人力工程师）以及“Forward Deployed Engineer”（前沿部署工程师，该概念最早由 Palantir 提出）的需求正爆发式增长。这类人才既是业务领域的专家，又具备直接为特定领域构建 AI 工具的能力。在初创公司内部，People Ops 经常需要将数据跨 5 个甚至更多不同的系统进行传输与整合，而传统软件工程团队往往专注于单一系统的代码开发，最初很难理解 HR 操作的痛点。但当 People Ops 自行构建自动化流程并成功将整个 G&A（通用与行政）部门的生产力基线提升了 **30%** 时，这种技术赋能的巨大价值便不言而喻。 ``` 传统 ATS 架构循环依赖问题 (Circular Dependencies): ┌──────────────┐ 引用 ┌──────────────┐ │ Candidate │ ───────────> │ Applications │ └──────────────┘ └──────────────┘ ▲ │ │ 引用 (Loop) │ 引用 └─────────────────────────────┘ ``` Bryan 深刻剖析了当今主流 Legacy（传统）HR 系统的底层架构缺陷。许多遗留系统的“Plumbing”（管道/数据架构）在设计之初并未考虑未来的数据爆发与 AI 集成需求。以应用追踪系统（ATS）为例，由于历史上不断叠加数据库表，导致表与表之间出现了严重的循环依赖（Circular Dependencies），甚至无法在单一模型中构建完整的分析 Dashboard。这种架构极大地限制了 AI 的发挥，也促使大量原生的、高度可定制的下一代 HRTech 工具不断涌现。 此外，Bryan 揭示了 AI 如何将过去极其繁琐的定性数据（Qualitative insights）转化为高价值的决策依据。以往对面试反馈、同行评审（Peer feedback）、上下级评价以及离职面谈（Exit interviews）等海量文本数据的分析完全依赖人工阅读，既耗时又主观。现在利用 AI，HR 团队可以直接提出极其复杂的管理质询，例如“某位主管是否是一名合格的管理者？”系统可以跨越多年时间维度，综合分析其团队内 **10 名** 员工离职的真正原因，从而揭示出以往单纯依靠数字统计或经验主义无法发现的深层管理问题。 * **生产力提升幅度**：在 G&A（通用与行政）职能部门中，通过自动化手段将生产力基线（Productivity floor）提升了 **30%**。 * **前沿岗位类型**：重点提及了 Forward Deployed Engineer（前沿部署工程师），即深入业务一线构建定制 AI 工具的领域专家。 * **底层架构困境**：传统 ATS 等系统因数据库设计缺乏扩展性，存在严重的循环依赖问题，无法直接支持复杂的 AI 数据分析。 * **定性数据解锁**：AI 使得对离职面谈、同行评估等海量非结构化文本的纵向深度分析成为可能。 --- ### [31:35](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=1895s) 为什么必须是“现在”：数据、上下文与“使用 AI 来学会使用 AI” 当主持人提出“为什么企业应该现在就引入 AI，而不是再等一年”时，Kristie Rowley 首先给出了坚定回应：“AI 不会消失，它将长久留存。”她强调，未来构建的所有 HR 产品都将具备 AI 组件。对 HR 领域而言，AI 成功的两大决定性要素是“数据”和“上下文（Context）”。 Kristie 精辟地指出，几乎所有 HR 部门都拥有大量数据，但“缺乏上下文的 AI 充其量只是一个非常非常优秀的花式模式识别器（Pattern recognizer）”，无法产生真正的创新。唯有将丰富、细腻、具象的 HR 上下文注入 AI，才能激发出具有变革性的业务价值。 ``` AI 在 HR 中的核心价值驱动模型: ┌─────────────────┐ ┌─────────────────┐ │ HR 数据资产 │ + │ 业务上下文 │ │ (Data Assets) │ │ (Context) │ └─────────────────┘ └─────────────────┘ │ │ ▼ ▼ ┌─────────────────────────────────┐ │ 高阶 AI 模式识别与深度洞察引擎 │ └─────────────────────────────────┘ │ ▼ ┌─────────────────────────────────┐ │ 组织创新与高绩效团队构建 (Value) │ └─────────────────────────────────┘ ``` Alberto Silveira 补充道，AI 能够为人类考官提供实时洞察，在进行多场面试时动态对比候选人之间的表现。对于人类而言，连续面试 **15 到 20 名** 候选人后，要保持客观公平的记忆与评估几乎是不可能的，而 AI 恰好弥补了人类在处理复杂上下文时的认知极限。Alberto 强调：“机会就在眼前的此时此刻，甚至可以说是昨天，因为技术演进太快了。” Bryan Hong 提出了一个别开生面的观点：现在的核心理由是“这太有趣了（It's fun）”。他分享了自己的切身经历：自己曾经长期担任 Talent 负责人，常羡慕软件工程师的高薪，甚至在 **38 岁** 时在 AI 的辅助下成功转型为一名真正的工程师。他认为 AI 最大的价值在于赋予每个人强大的学习能力与“思考伙伴（Thought partner）”。 Bryan 分享了一个具体的自动化实践案例：上周日他在听了几期播客后想将其制作成培训课程，于是他坐在电脑前使用 Claude Code 提出需求。系统在短短 **30 分钟** 内便搭建起了一套自动化 Agent 工作流——只需输入任何 RSS Feed，系统就能在 **10 分钟** 内自动抓取音频、生成逐字稿并提炼出完整的培训模块。这种效率在两年前是不可想象的。 Bryan 同时也打破了一些网络上常见的“AI 神话”。他指出，那些宣称“在周日花几小时做出的 AI 系统就能自动化管理整个公司”的言论完全是谎言（"That's a lie"），个人探索与企业级落地有着巨大的鸿沟。但他强烈建议大家不要被技术吓倒，应当积极尝试与实验，因为在当下，掌握 AI 已经不再是一门可选课，而是必不可少的职业要求（"Job requirement"）。 * **个人转型范例**：Bryan Hong 在 **38 岁** 时借助 AI 的辅助，成功从 HR 领导者转型为软件工程师。 * **播客转培训 Agent 案例**：利用 Claude Code 在 **30 分钟** 内开发出工作流，可自动将 RSS Feed 播客转换为 **10 分钟** 的标准化培训课程。 * **认知澄清**：明确指出宣称“周日用 AI 构建的项目就能运行整家企业”是夸大不实的说法，企业落地需克服诸多现实复杂性。 * **AI 核心驱动力**：数据（Data）提供基础，上下文（Context）激活创新，二者结合才能摆脱单纯的“模式识别”。 --- ### [39:48](https://www.youtube.com/watch?v=MSK2WFVXIF4&t=2388s) 技术的全民普及：从 enterprise 走向 SMB 与全员赋能 针对“AI 是否只是大企业（Enterprise）的专属工具，中小企业（SMB）表现如何”这一提问，Kristie Rowley 斩钉截铁地回答：“是所有人，现在的 AI 适用于每一个人。” Kristie 分享了 BambooHR 内部的技术普惠文化。作为一个工程经理，在传统的科技公司中，往往只有工程师团队才能优先享用最新的软件与开发工具。但在 BambooHR，公司将 AI 技术与工具无差别地开放给了全公司所有部门的每一位员工。 这种全员赋能带来了惊人的创造力迸发，员工们纷纷利用周末或下午的闲暇时间构建出各种令人惊艳的自动化工具与应用。虽然 Kristie 也幽默地开玩笑说“我们可能并不希望市场部的人跑去开发 HR 软件”，但这种全员拥抱 AI、探索技术边界的文化，正代表着 HRTech 与未来工作方式变革的方向。 * **适用客户群体**：覆盖从 SMB（中小企业）到 Enterprise（大型企业）的所有规模公司。 * **内部全员赋能**：BambooHR 将 AI 工具的权限打破部门壁垒，全面开放给非技术岗位的全员使用。 * **自下而上的创新**：员工自发在周末或课余时间利用 AI 构建解决自身业务痛点的小工具。 Sorry, something went wrong. Please try your request again. ## Video description Alberto Silveira - Chief Technology Officer (CTO), HireVue Kristie Rowley - Sr. Manager of AI/ML Engineering, BambooHR Marshall Bush - Founder, Baleen Research Bryan Hong - Director of People Ops & AI Services, Astranis HR software is moving past manual workflows to intelligent platforms that anticipate both employer and employee needs. This panel will highlight how HRTech providers are embedding AI into recruiting, onboarding, and retention, as well as how companies are using predictive analytics to understand employee engagement and flight risk. Panelists will address fairness and compliance challenges, while exploring how personalization and generative AI can elevate employee experiences. Walk away with insights into how AI is changing HR systems—and what it means for the future of work. [Source](https://www.youtube.com/watch?v=MSK2WFVXIF4)

[2:05] Kristie Rowley: AI is a new delivery mechanism for how we serve up software, but I don't think it changes the core role of what we want to accomplish with that. [3:31] Alberto Silveira: We don't guess, we validate.

[0:04] Uh hello everyone, Alberto Sivera, CTO at  H Higherview. Um we are a company that have [0:10] been way prior to this world of AI. We help uh we  are platform that help assessing and validating [0:17] skills. Um happy to be here and Christy Rowley.  I am the AI ops leader at Bamboo HR. Uh we build [0:27] HR software. Um, and then of course I'm running  internal AI projects all the time for Bamboo. Uh, [0:34] yep. My name is Brian Hong. I'm the director  of people ops and AI services for a startup [0:40] called Astronis. Uh, we build satellites in San  Francisco. Um, and basically kind of oversee the [0:46] operational capacity throughout all of our people  teams, talent, and then also deploy AI solutions [0:52] across the entire GNA function. Awesome. Christie,  you mentioned that you were at Chararm recently. [0:58] Can you tell us sort of what the the broader state  of HR is and how Bamboo fits into that? Yes, sure. [1:05] Sherm was fantastic. Super fun. For any of you HR  professionals out there, you know exactly what I'm [1:10] talking about. Um, it's exciting. It's an exciting  time to be in HR. We have a lot of emphasis on AI. [1:20] Many of you out there are probably sick and tired  of hearing about AI. Um, and this is this great [1:26] moment where AI and HR, something that you tend  to think of more human centric and less computer [1:32] centric, are merging together and coming together  to create really interesting frontiers in what I [1:37] think is the purpose of AI to be more individual,  to be more adaptable and to be more amendable [1:44] to the humans that it's supposed to interact  with. So, uh, we have a lot of really exciting, [1:50] um, challenges ahead of us to make AI and HR  really compatible and make AI, um, human and [1:58] as human as possible and supporting the humans  that we work with. Awesome. And what are your most [2:05] evolved customers look like right now? And how is  Bamboo um, changing their lives? That's a great [2:12] question. Um, I think that all of our customers  are on an AI journey and it's okay that you're [2:21] on a journey whether you're just barely starting  out or whether you're very very advanced in the [2:25] journey where you're you've got fleets and fleets  of agents that are automating a lot of workloads [2:29] for you. Um, every step along that journey is  really really critical and important starting [2:35] with the foundational let's play around with it  for a minute all the way to let's automate a lot [2:41] of our workflows with it. So, so whether they're  super advanced or whether they're not, we're [2:45] meeting people wherever they are. That's what good  software always should do. And AI is no different. [2:51] AI is a new delivery mechanism for how we serve  up software, but I don't think it changes the core [2:56] role of what we want to accomplish with um with  that. So, what we're trying to do as an HR company [3:03] is meet our customers wherever they are. And that  means not just bolting on another agent or another [3:11] bot or another whatever to call us an AI native  platform. [laughter] It's actually architecting [3:17] those experiences to meet our customers where  they are to make those individualized and to make [3:22] them personalized. Um so we can meet them wherever  they're at. [snorts] And Alberto, can you talk to [3:31] us a little bit about how you're helping with um  hiring and assessment? Yes. Um it's um it's kind [3:40] of a paradox because in the world of AI, we try to  humanizing things. So that's how we're trying to [3:48] uh and that's what we tell our customers. Um,  for those that don't know about Higher View, [3:53] I mean, we, as I mentioned, we are like, we've  been here for a while, over two decades. So, [3:58] we are way prior to AI, uh, at least the way that  we know AI uh, today. And um I think um as every [4:09] day probably during this talk already that there's  a new company creating a new software delivering [4:16] like a AI solution for helping hiring and uh  so the way that we are trying to help is like [4:25] we cannot compromise integrity uh over speed  and um every talk uh every new feature we try to [4:37] build we we try to let's step back a little bit.  Let's not rush too much. Um at the end of the day [4:46] um in everyone's life is one of the most important  milestones is to get a job and to make sure that [4:52] those individuals have a successful career. So  we cannot just you know uh prematurely deliver [5:00] that to uh technology without human intervention  and I I heard this in many talks that I attend [5:07] today about the human uh and the humanizing  aspect. So um yes we we we leverage uh AI for [5:18] helping assessing to validate skills. uh we uh we  like to say we we don't guess we we validate and [5:28] um and and that's how we are trying to help our  customers like introducing the technology but in a [5:35] very responsible way uh not compromising integrity  and making sure that uh the whole hiring process [5:45] um has the AI to assist but have the human is  still in control. Um, and I like to emphasize [5:52] that a lot because uh, one of the recommendations,  don't keep seeking the newest shine feature over [6:02] and over like we we we have to kind of pause and  slow down at times and that's what we as leader [6:09] in the industry we try to convey to customers and  to uh, the audience here of course. Cool. Brian, [6:16] you've um sort of come to a a different decision  on a lot of the build versus buy um decisions. Why [6:24] don't you walk us through your journey and and  tell the audience what what you've done and and [6:28] how you decide whether you're going to build  something? Uh yeah. So about two years ago, [6:34] um everybody was talking about chat GPT. Um and I  was uh I guess kind of skeptical, I guess, in the [6:42] beginning. Um but obviously very curious. And so  started to really just dabble into understanding [6:48] how to use it. And then that kind of led to a  higher level of curiosity where I started to, [6:53] you know, do things that, you know, I guess in the  past I would never have thought I would be able [6:58] to do such as like, you know, integrating like  LMS to our systems, doing qualitative analysis, [7:06] starting to build like agentic swarms. Um, and  that was maybe 18 months ago. Fast forward um [7:13] I basically own a entire kind of Google project  uh where me and a very small team essentially own [7:21] the infrastructure end to end built the DevOps  methodology and essentially deploy agents to [7:26] all the different GNA teams and kind of partner  with them to understand what their workflows are, [7:31] what their problems are, what they'd like to  optimize and we took a long time to get there. [7:37] I mean there are a lot of things around governance  uh security um you know even just choosing what to [7:44] automate right we consider things that are you  know how much volume is this right like how many [7:49] times a day is someone doing this if it's once  a month like should we even build this like is [7:53] it worth it um does it require a lot of human  judgment you know is this something complex [7:58] will it have very challenging consequences  that you know you can't rewind uh then you [8:04] know maybe it's something that we want to avoid  uh we try our best to as we experiment and build [8:09] we definitely try to tackle the things that are  you know I would say simple to more complex and [8:14] continue to kind of extend those solutions um but  I do think it is a very you know it's not even a [8:21] shift I think it's definitely a complete uh kind  of like hopping of lanes because now people who [8:28] are these HR experts or these talent operations  individuals who know their process most deeply [8:36] have the option to build something or buy and not  build like as in you ask your engineering team or [8:43] you have to it's like they can just sit there and  build it and of course there is a discipline and [8:49] obviously a lot of I think learning that you have  to do uh to be able to make something that is both [8:55] consistent reliable um but it really does allow  you to say okay like do we want to just build [9:01] it or buy it and in the past it was really I  think for the most part especially in people [9:05] tech it was do we buy this or that. It was never  a question of do we build like I I hardly know any [9:11] companies outside of you know your very very large  corporations that will say hey let's build our own [9:16] application tracking system let's build our own  HR um and so it it's definitely a very different [9:22] dynamic and I think a very different environment.  What are some examples of some things you've built [9:27] just so the audience has a finer point on this?  Yeah. So I guess to start out astronated company [9:34] we provide a lot of satellites and services in  contingency with the United States and so um you [9:41] know I feel very envious of those who are at work  and they're like I have cloud co-work uh we have [9:48] a lot of restrictions on what we can use which  models we can use where they need to be located [9:54] and so there is a kind of embedded uh requirement  for us to say we should build certain things [10:00] uh just due to sensitivity. Um but I would say  probably like the first thing that I built was for [10:05] my talent team. Uh, I built kind of this master  uh agent that would basically delegate things [10:11] like writing job descriptions, uh, doing like  an educational brief to ramp up recruiters, uh, [10:18] mapping out talent uh, from a data perspective,  building sourcing strategies, outreach campaigns, [10:24] interview plans, and it could basically do it,  you know, basically like the seven to eight in uh, [10:29] executions uh, via one kind of chat interface  for any recruiter. And you know a lot of that [10:35] research in the past would take you know one a lot  of expertise and you know a lot of hours sometimes [10:41] weeks to to put together and you know now we were  able to kind of get that in minutes and that was [10:45] a very big time save and and benefit and kind of a  consolidation of knowledge for the team. And would [10:52] you argue that you're obviously what you're doing  right now is rare. Would you argue that you're [10:57] sort of the future or or you're just unusual based  on, you know, how tech forward your company is, [11:04] etc.? Um, I I I I would [clears throat] say I  know uh communities of builders. I definitely [11:12] think that there's a very strong interest. I don't  know if people are there yet. Um I think people [11:19] are definitely starting to put together the like  understanding that the you know idea person and [11:26] the builder are now kind of the same person.  Um but I do think there's very you know there [11:31] are very strong cases for you know best-in-class  solutions purchases like you know a lot of what [11:37] we try to build is you know do we think you know  our application tracking system which you know [11:43] I won't name but is perfect probably not um  but we build a lot around it to make it more [11:51] uh valuable to give us the data that we want in  the way that we want it to serve our purposes to [11:56] meet our strategies and business objectives. And  so I think there will be uh some that are on the [12:02] very fringe I guess like similar to me and my  teams where you know we are outright building [12:07] you know agents and for different use cases and  you know building our own databases and you know [12:12] maybe choosing not to buy and then I think there  are also teams that will kind of be in the middle [12:17] of like building around it and building with  it and extending the solutions. Um, I think it [12:23] really just kind of depends on, you know, company  maturation level, your budgets, um, you know, [12:28] what resources you have, what you want. Um, so  yeah, same kind of like not one sizefits-all, but [12:33] I I do think there's a stronger interest in build,  people wanting to build, I think, and Christie, [12:41] we could argue that with SAS, software became  less customizable. um you know if if a if someone [12:49] here wants to build something but they don't  necessarily want to build their own data model and [12:54] they want to leverage something like bamboo how  to what extent are you selling complete modules [13:00] and to what extent is can they develop their own  workflows on your systems it's a good question so [13:06] we've been around for about 18 years I think so  this is not new technology we're built on as a [13:14] traditional SAS platform form. We got in front of  the AI curve about six years ago and took a look [13:21] at where AI was headed, where machine learning was  headed and thought we need to kind of get in front [13:26] of this. So, we started integrating AI pieces  then. So, this is a six-year-old endeavor for us, [13:32] but we still have the legacy SAS platform where  you you build for the 80%. you go out and you [13:38] interview your customers and you figure out what  do the what do most customers need and you build [13:42] for those 80% of customers and you sell. Um what  we're finding now um is that with AI you can [13:52] build core foundations but the experience can be  highly customizable to the individual and highly [14:00] customizable to the person. I'm not seeing tons  and tons of people in the HR space that want to go [14:04] build their own agents. There there are some there  are some out there. Uh there are not a lot that [14:10] necessarily want to go build their own agents.  Um in fact I I think when I went to Sherm last, [14:15] one of the things that people were were talking  about in the HIS space was, oh, we have 10 agents, [14:20] we have 25 agents, we have 500 agents. Um I would  argue that no one cares how many agents you have. [14:27] And HR professionals probably certainly don't  care how many agents you have. They do care about [14:32] the experience they get when they navigate the  software. Uh, so talking about number of agents [14:37] is kind of like saying we add another cello to our  symphony. Like I don't care unless it's YoYo Ma. I [14:43] do not care. Don't care about your extra cello.  Um, but if if it's YoYo Ma and they can give me [14:49] a really amazing, interesting experience, then  of course they care about that. So that's what [14:54] we're trying to focus on. We're trying to focus on  opportunities to meet our customers uh, where they [15:00] are and also to give them the opportunities  to invest in the experiences that they want. [15:05] uh easily easily if they want to build agents,  great. You can build agents on top of our [15:09] platform. All the all the software and all the  hardware is there. You can definitely do it. [15:13] Um if you want to invest in professional services  so somebody else can build your agents, we've got [15:19] that. We can do that for you, too. Uh if you just  want it out of the box, more or less plugandplay, [15:27] um that's also what we're working on. And that is  the core focus as of just two weeks ago when we [15:33] rolled out our big AI platform. That is the core  focus of what we're what we're the experience that [15:38] we're driving towards. And when you look at that  new platform, can you talk to um you know do you [15:45] have any alpha customers yet where you can say  this this is the impact? Yeah. Yes. We we've [15:50] been beta testing and alpha testing for quite  some time now. Um, and many of those are rolled [15:55] out more comprehensively in the last two weeks to  most of our customer platform. Um, just with some [16:01] of the early stuff that we've rolled out, we're  hearing that this is saving HR professionals up [16:05] to like 17 hours a week and just not having to  answer redundant questions that people have. Um, [16:12] and it's exciting for them and it's an engaging  experience and it's uh it's useful when the the [16:20] end users not necessarily the HR professionals but  the the employees are using that on their own and [16:25] finding value in that and can use AI to drive some  value in familiar formats and familiar context [16:32] uh in ways that save everybody a lot of time.  Uh so so that's one of the things we're driving [16:36] towards at least that's our early win out of  the gate. You mentioned about the the experience [16:43] and like the number of agents which is not a  meaningful thing, right? Quantity over quality, [16:49] whatever uh that that that is. But um um uh do  you remember those times where you would apply [16:57] for a job and um you wouldn't hear back and  or or you would get like a automated message [17:05] and the experience was just awful. Right?  So these are like examples where I think [17:12] uh technology can help improve the experience  from the moment that you apply for a job to all [17:21] the interactions where it's a more personalized  for your application and one extreme would be okay [17:30] uh now I'm only talking with computer and  I don't have the human the other would be [17:38] I I'll only have the humans but my application is  never analyzed in uh with a fair shot compared to [17:47] all the applicants. So that's uh answering a  little bit of your earlier question how we are [17:53] trying to do is like it's a mix of yes trying  to apply AI to um in the context of the user [18:02] experience like the from the applicant from the  recruiter from the evaluator from the different [18:07] uh angles um where every candidate have a  fair shot uh uh uh throughout the process And [18:18] uh that doesn't mean that is the number of agents  will make the difference but how can we as as you [18:26] know in the HR HR tech industry literally  create that uh good experience and I think [18:32] that's the question that we are always asking  ourselves um and it's not an easy easy uh easy [18:38] job to do but I couldn't agree more all of that  to say I couldn't agree more with you that it's [18:46] um the experience That's what matters and that  it's between hire but the retention as well like [18:56] uh how how how many of you is actually thinking  about the data and an uh analyzing after the [19:04] person gets the job is is was that the right hire  and how you get that and reapply to the models to [19:12] improve the experience from that candidate and I  think that is uh all comes back to the experience. [19:18] But yeah, no, I love that because when we narrow  down on that one use case, the hiring experience, [19:25] um, it highlights how that there's not a  shortage of information anymore. There's [19:30] so much information we're drowning in it and  it's hard. It's hard to compile. It's hard to [19:35] think about. It's hard for humans to compile it  and think about. Um, and one of the things we're [19:40] focusing a lot on, and I I'm guessing all of you  are because it's [laughter] really important is, [19:46] um, we want to get the right information to the  right people at the right time when they need it [19:53] and when it's really valuable. And AI can do that  in some instances way better than humans. Not all [19:58] of the instances. In some instances, humans are  far more valuable at that. Um, especially when you [20:04] talk about final decision- making and hiring and  and things like that. But I think you're you're [20:08] highlighting these these ideas for we would love  to have everyone to have this lovely wonderful [20:12] human interactive experience every time they apply  for a job. It's not going to happen. Um but we can [20:18] use AI to give them a far better experience than  what most people are experiencing today. So I I [20:24] love that example. That's amazing. So to to  better contextualize this for the audience, [20:31] if everyone here applied for you know the same  job and hired in a company had retained hire to [20:39] help before AI how many would you interview on  a video interview and now with AI how many might [20:45] you interview? Yeah, it's a great question. Um  hire definitely helps organization with largecale [20:54] hiring. So we recently res um u released what we  call like the AI interviewer is literally to help [21:03] u you know like from screening resumes uh and  giving like the fair shot that was mentioned [21:08] earlier but uh depending on the customer um and  the timelines that customers have to hire like [21:17] there's like some customers that they do hiring  events they have like two weeks and they have over [21:23] a 100,000 candidates to be interview in two weeks.  How do you do that with just humans? It's nearly [21:29] too impossible. I mean, it's impossible. I guess  at least with like the the level of fairness and [21:36] uh with the intent to hire the best person for  the job. I think all of our customers, that's what [21:41] they want to do is to hire the best person for the  job. So um it's um leveraging you know a mix of on [21:51] demand interviewing AI interviewing like live  interviewing with humans it's a combination of [21:58] uh technologies um and now we have most recently  one of the things that we realized uh and the [22:04] data is showing this with AI it becomes it became  even harder to interview people because everyone [22:11] became an expert overnight you know you you're  you're you're interviewing someone and you see [22:16] the eyes of someone actually and there's like bots  and tools and and at the end of the day is like [22:24] uh it's computers talking with computers and  and people trying to pretend in between. So [22:29] I think this is the challenge that we're uh  building technology to help candidates and [22:35] uh uh employers to find the true self of the best  person for the job. So um answer your question is [22:44] a combination of tools and and practices and  and ways of assess if people can do what they [22:51] say they can do. Uh and I think this is where  where we try to help the our customers. But [22:58] just in terms of how broad you're casting the  net, are there situations now where you might [23:02] have you know historically screened interviews  and interviewed a hundred candidates on video [23:07] and now you'll interview everyone who applies? We  we can definitely evaluate everyone that applies [23:16] uh to a certain degree with uh um with  more fairness. We we can definitely [23:25] um scale more uh the quality of interviews uh  with AI with the models and the trainings that [23:33] we do. And then we do customize by by customers  but uh the customers that we have for longer we [23:40] have specific models for those customers and what  works for those customers and that personalization [23:47] is what actually really makes uh us more unique.  Interesting. I mean, I I keep pulling this thread [23:54] because I guess what's interesting to me is do  do you ever find that people who you previously [23:58] would have screened out and wouldn't have  been interviewed are now getting jobs? [24:05] Um I I I I would say what the data shows is that  uh customers are able to hire the people that [24:21] they want faster and we have some several actually  use cases where uh the match it's happening more [24:30] accurately and faster. uh and that is the exciting  part. If you use the technology in the right way, [24:35] you can achieve those better results. Um and  I think that's what if I could, you know, [24:42] interview one person and get the right person for  the job, beautiful. We can actually move on. But [24:47] um the reality is you have like thousands of  applicants for like the same job. And uh and [24:54] then another thing that is happening now with the  technology is allowing us to do is you know many [24:59] times people applying for job A they do the  assessment and we realize that actually that [25:06] person would be better qualified for job B and  that person didn't even know about that position. [25:11] So doing that kind of u it helps expedite um and  getting people the right people hired faster. [25:19] Cool. because there's there's so much concern  about AI killing jobs. It would be interesting [25:23] if we have real examples of AI actually helping  more people find jobs and find the right jobs. [25:29] I don't think is um I think like in the previous  panel there was like this conversation or like [25:36] there an aspect of it. I don't think it's  killing it's transforming jobs. I we yeah I I see [25:43] positions uh being transformed. Uh so yes certain  jobs will probably be uh you know eliminated over [25:52] time but many others are being created. So, it's  it's about the transformation uh and adapt I [25:58] guess. Super interesting. Um Brian, are you guys  using any of the things you built in the actual [26:06] um applicant identification and hiring process or  is it more once they're inside your four walls? [26:13] Um, typically for evaluating candidates, I  would say we have certain tools. Definitely, [26:18] I think ones [clears throat] that are embedded  within the softwares that we use. Um, as well [26:22] as our own. Um, I think maybe my the startup that  I work at is a little bit more peculiar in that [26:31] because of some of the eligibility requirements,  we typically veer away from like full automation [26:37] primarily just to make sure we're not uh we  are remaining within compliance. um and maybe [26:45] uh a difference of opinion uh but I do think there  is a definitely growing demand in in expectations [26:52] and and in capability uh you know I think if  you've paid attention to the job market there's [27:00] like like a lot of listings for talent engineers  people engineers um there's something called a [27:07] Ford deployed engineer which you know originally  was coined by and I think Um but you know these [27:13] are essentially individuals who are domain experts  that go in and engineer AI or build for their [27:20] specific domain but you know are not we're not  maybe necessarily former builders but you know [27:25] that is what we've found at our company that can  make it very difficult which is you know we'll [27:30] talk with our engineering team and we'll say hey  we need to you know hook up all these systems and [27:34] we want to automate all these things and you  know they're just kind of look at this like a [27:38] like a kind of a huge like hack just waiting to  happen. Um you know but for us you know like you [27:45] know passing data between like five systems it's  like a very normal thing in like people ops right [27:50] um engineers are typically you know working on  a single application within a single scope of [27:55] code and so they don't typically see one engineer  that works across like five different platforms. [28:01] You know there's usually a team per the section  or module or whatever. Um so sometimes it could [28:06] be hard. It was very difficult in the beginning  to kind of communicate like what we're trying [28:09] to accomplish. Um but I think it became very  apparent that um you know if we are able to [28:17] build this and if we build it fast uh and we are  able to increase the productivity floor by 30% [28:24] across the entire GNA function that is like at  least for a startup like very very impactful. [28:30] um for larger companies of course like also very  very impactful but I think it's the way that maybe [28:36] you think about it in terms of what I've found  with a lot of the you know best-in-class kind of [28:42] legacy systems is they can build AI on top of  it and it's definitely helpful uh I think the [28:48] fundamental issue is the plumbing so when you look  at the databases the way it was structured um you [28:56] know a good example is you know our application  tracking system you try to build dash dashboards, [29:01] you have to build three sets because you can't  build it in an entire model uh because there are [29:07] circular dependencies because and that basically  just means if you try to connect one database with [29:11] another data or one table with another table  and another table, it basically just loops and [29:15] it keeps referencing itself and it can't figure  out what's going on. And that was because in the [29:19] past they just didn't know that they were going  to grow this much and they didn't know that they [29:22] had this many fields. So they just keep adding  tables but it's not really optimized for AI. [29:28] Um and so I think I mean this is my personal  opinion but I think either the solutions [29:34] that exist today will need to really adapt and  and really think about like the entire system [29:42] um or uh I think new ones which I'm already seeing  quite a bit um will pop up because they are much [29:50] more optimized and and built to be uh you know  I guess out of the box customizable. versus, you [29:59] know, here's like one solution that most people  will like and you can kind of tinker things here [30:05] and there, but for the most part like you can't  change everything like you can't just do whatever [30:09] you want. Um but I think you know talking to a  lot of founders and a lot of other startups and [30:15] you know they're building like you know tools in  cloud code where they can have live feedback chats [30:21] while they're interviewing someone and a dashboard  that they can use uh that they just then automate [30:26] notes into their you know ATS or um to uh's point  uh you know we are thinking about building a data [30:34] lake that really measures these very specific  metrics. So things like uh interview feedback, [30:41] interview scores, um performance feedback from  their peers and you know people uh that are their [30:48] superiors and people that report to them and then  also looking at like exit interview and now you [30:53] know you can ask some really tough questions  because in the past if you wanted to analyze a [30:58] lot of this data it was just mass amounts of text  and so of course someone's going to sit there and [31:02] have to read it. It's very subjective. Uh but now  you can ask questions like is this person a good [31:07] manager? And you can look at that over time. Maybe  they I don't know 10 people quit on their team in [31:12] a year and they're saying oh these people were all  unqualified. Um and you can look at the data and [31:18] you might be able to get some really good insights  but I don't think you could have done that in the [31:23] past and everything was always numerically driven  or just purely you know empirical. Um but I think [31:30] AI unlocks a lot of qualitative insights as  well. If we sort of pull on that last sentence, [31:35] I don't think you could have done that in the  past. I want to throw this out to all of you. Um, [31:39] if there are people in the audience who aren't  using AI and HR right now, why is now the right [31:45] time to do it? Why not wait another year?  What what can they do today that is unique [31:50] and powerful that wasn't recently available?  And why why should people go home and and act? [31:58] I'll go first. Um, it's not going away. It's  here to stay. Uh everything that we build from [32:05] here on forward is going to have some kind of  AI component to it. Uh I think the better that [32:10] we can interface with it and interact with it  and get to know it and figure out how it works, [32:14] uh the better off we are. Um there are two things  that are missionritical to AI and one is data. [32:21] We've we've mentioned that already. Uh for all  of the HR folks, this is not new information. [32:27] Everybody has data. Everybody has data challenges  and now it's mission critical to get those really [32:33] organized. But the second one is context. And  this is the most powerful piece to me about where [32:40] AI and HR kind of collide in really interesting  ways is because um without context uh AI is just a [32:48] really really really good pattern recognizer. And  so you can't do anything particularly innovative [32:54] with that um because it's just going to recognize  all kinds of patterns and patterns that you can [33:00] prove out mathematically. Um, with context,  you can give it rich, insightful, nuanced, [33:08] truly innovative information and work with it and  navigate it and play with it in ways that can be [33:16] fun and and exciting and will drive our businesses  to much much much better places than we could [33:21] drive them before. And so that combination of data  and context that brings about that innovation is [33:28] where we need to get really comfortable with AI.  And the sooner we do that and the faster we get [33:34] up to speed on that, I think the easier this next  transformation is going to be for all of us as we [33:40] enter that that next um advancement space where we  we know it's coming. It's right around the corner. [33:47] Um I I will go back to the the context which I  agree very much and I will give like some good [33:53] examples but um where context can be very helpful  but what's the opposite right if it's not now then [34:00] when because it's not going anywhere so I agree  on that but um and I think we're um we're like in [34:11] the the beginning of like this still you know I  think everyone is trying to figure out in a way [34:16] in another but um uh I I think the opportunity  and when we connect with context so the way [34:25] like the way that you know um imagine like when  you're like interviewing someone and and and you [34:35] as as you're interviewing you can actually have  insights and comparing with other candidates that [34:42] you already talked um and then how like those  people are actually answering and because for [34:49] humans is really you interview one two three yeah  you remember now interview 15 20 like it becomes [34:56] like human impossible so I think what AI brings to  the table is like this context this intelligence [35:05] that for us humans becomes really challenging  to to do that in a fair way at least so I think [35:14] building these systems are leveraging these  systems uh giving the AI the context to really [35:22] help building these high performance teams that  we all want and retaining them because one thing [35:28] is hiring the other thing is retaining um I think  that's where the opportunity really uh lands and [35:36] uh and that's why I think it's now and honestly  yesterday because things are moving fast and [35:44] uh and and I think the opportunity is uh to do  things that before in naked eyes, let's puts that [35:51] way, it wasn't possible. So, um Brian, what what  are your thoughts? Uh yes, maybe um my answer is a [36:01] little bit less um uh like I I don't know. I don't  know what you call it. I would say first reason [36:08] for me at least is it's fun. Um it's fun. It is  amazing that uh you know historically I I was like [36:17] ahead of talent and I had been doing that um very  fortunate to have a lot of people teach me and you [36:24] know there are a lot of times where I'd like hire  engineers and see how much money they're making [36:28] and you know like maybe you know 5 years ago every  like 3 four years I'm like man should I maybe I [36:34] should have been a CS major. Um and I literally  converted to be an engineer at like the age of 38. [36:41] with the help of AI and I think that is maybe the  biggest value is the ability to learn. It is the [36:49] ability to be a thought partner. Um I can uh I  I I don't want to disagree but I'll half agree [36:55] on the innovation. I think the models are good  enough at this point where um a lot of people view [37:03] themselves as a limiter and my motto has always  been that like you should use AI to use AI. So you [37:09] don't know everything it can do. you don't know  everything it knows and so you ask it you should [37:13] ask it like how do I do this and how do I make you  do these things um you know the use cases that I [37:20] have like I don't know last Sunday I listened to  a couple podcasts and I was like man this is some [37:26] really good content I wish I could make a training  and I just sat down at my computer on Sunday and [37:31] went to my claude code and I was like hey I need  something that will record these podcasts from a [37:37] RSS feed and then just build the transcript script  then I want you to build it into a training. I [37:42] already had a training module like with agents  and I don't know it did it in like 30 minutes or [37:45] 30 yeah 30 minutes and I can just literally plug  in any RSS feed it'll build me a training in 10 [37:52] minutes. Um this was impossible for me like two  years ago. Um and so I really encourage people one [38:01] I think half of using AI is just literally sitting  down and trying it. I think a lot of people are [38:06] very like daunted and they're like they see these  crazy YouTubes and and don't let them fool you. [38:11] Anybody who says like, "Oh, I just like built my,  you know, I I built this like on a Sunday and it [38:16] runs my whole life." That's a lie. Um it you might  be able to do it personally, but you'll never [38:21] do it at a company. Um but I think you should  definitely experiment and you know it can help you [38:27] think about things in a different way and bring  you solutions that maybe you yourself or your team [38:32] would never think of. It's a great sounding board.  Um and uh to Alberto and and as well as her point [38:38] uh it is it's I think at this point probably a job  requirement. It's not a it's not like a choice. [38:44] It's it's you know I think as usual like people  tech you know sales sales tech is a little bit [38:52] more forward than people people is probably like  the latest tech like it's just like the last ones [38:57] to get like the best technology in my opinion and  I feel like a lot of people are now realizing like [39:02] is this going to kind of part a part to be  required in my job and so yeah if anything I [39:08] definitely think it's fun I think it's an amazing  tool um and the great part is is you know it [39:14] extends ends value. So you can build an agent and  I did that for a lot of my teams and you come back [39:19] a month later and they have built something else  that's done like different things and extended the [39:25] value of what you originally accomplished. And I  think not a lot of softwares in the past did that, [39:30] right? You know, you buy something, you know,  like it tracks something and then, you know, [39:35] it emails someone and that's it. Like that's  all it could do. You can never do more than [39:39] that just if you just give it to somebody uh  for like two weeks, right? And so um yeah, [39:44] I guess that's that's my take. Cool. And and so  if the answer to when we should do it is now, [39:48] I guess the other thing I forgot to ask is is  you know who should do it? What do your ideal [39:53] customers look like? What sizes? Is this an  enterpriseonly technology? What are you seeing [39:57] out of the SMB space? Everybody. Everybody right  now. Um I I'm very fortunate to work at Bamboo [40:07] and I'm an engineering manager. So I I run teams  of engineers and uh it was it's pretty typical in [40:16] my space in tech tech companies where engineers  get all of the software. We're very fortunate at [40:22] Bamboo. They have just given all of the technology  to everybody. It's amazing to see to Brian's point [40:29] what everybody can build on the weekend or on  an afternoon or whatever. And maybe, you know, [40:36] we don't want the marketers building HR [clears  throat] software. Maybe that's not what we want [40:41] what we want to happen. Um, but I feel like it's  we've been very fortunate to turn all of these [40:47] things on for everybody and let them play with  them and figure out what they're what they can [40:51] do and figure out how they can multiply themselves  in their roles. Uh, it's been absolutely amazing. [40:59] One of the other thing that's really great for  for me a side benefit is that uh as other people [41:05] build maybe not your typical software engineers  um they realize that there are some things that [41:13] we in software engineering have had to deal with  and had to navigate for a long time that are hard. [41:18] Like oh cool I built this cool thing but now it  broke and I don't want to fix it. [laughter] I [41:23] don't I don't want to have to be the one that  everybody calls when it goes down or whatever. [41:28] like this is software engineering engineering.  So you learned how to build something, [41:31] now you have to learn how to build it in  production and maintain it. And so and [41:35] they're doing that and when the when the roles  apply um and when the when the value is there, [41:40] we're absolutely helping people and uh navigating  uh that world. And it's been great. I have a team [41:47] expressly dedicated to just helping people  who aren't necessarily software engineers [41:52] build AI applications that they think of that  they create. and uh their their adviserss and [41:59] supervisors have said yes we want that for our  business and so we'll help them build it so that [42:04] they have engineering background there in their  corner while they go and and build the thing that [42:10] they want to build and it's it's amazing it's  absolutely amazing so if you haven't done it [42:15] Brian's right so fun just do it try it very  few people I think are going to regret that [42:27] um yeah I mean I think everybody should use it  like it's like I said I don't think there's I [42:32] think it would be a very difficult case to make  to say these people probably don't need AI or [42:39] like they wouldn't benefit from it. Um but  I do think um I do think one thing that is [42:46] like probably really important to note is that  it doesn't unobligate teams to think about how [42:53] they actually want to do things. Um you know  if you have a really bad process like you will [42:59] get a very bad automation like if you have a  really crappy you know un like data hygiene is [43:05] like super bad and like all this the AI is not  going to help you. You got to help yourself. [43:10] uh you know if you try to you know basically  throw like agents at a really messy problem [43:15] that you haven't thought through or you know  tried to clean up before you try to apply AI [43:20] it will not turn out well for you. So you  still need to think about process design. [43:26] I think that's like very critical but um  but yeah I think everybody should use it. [43:33] not different. Um but I think um you know to  be very pragmatical I would say um get like [43:42] um [clears throat] a use case uh or a pain point  or like uh don't try to do all um all at once or [43:54] everything because I I seen people just say  oh yeah I'm going to be using for everything. [44:00] um but then it's just s very superficial and and  that that's not not where the value actually is. [44:08] So I think getting you know if we're talking about  retention and u you know in your organization and [44:14] you're leading HR or like or even if you're like  in your teams how do you do performance reviews [44:20] and so get that use case um do like retention  and you get performance reviews and then you [44:25] double click triple click yeah and and and that's  like the use case that you're going to try to get [44:30] uh and to improve. So instead of like going  shallow, I would say get one or two use cases [44:37] and then then you can actually go deeper. Then you  can find all technology and tools and models that [44:42] can actually help you solve a really business  problem. Otherwise it becomes too much play [44:46] playground but why we still have like a business  values to deliver. So awesome. Thank you so much. [44:53] We are out of time, but I appreciate all of your  um insights and um onto the next presentation.