ThinkingAI Chief Says Agentic OS Needs Metric Unified Knowledge Base

Agentic Growth: From Analytics That Report to Agents That Act With ThinkingAI — Ai4

Ai4

Enterprise AI adoption fails at execution because raw large language models lack unified metric definitions and secure sandboxed runtimes.

AI-assisted summary of Agentic Growth: From Analytics That Report to Agents That Act With ThinkingAI — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published by the channel Ai4 on YouTube on August 20, 2026, this presentation features Chris Han, Co-Founder and President of ThinkingAI. According to the page description, the session explores the concept of Agentic Growth, detailing the shift from static analytics dashboards to autonomous AI agents that analyze behavior and execute operational decisions in real time while maintaining data sovereignty.

## TL;DR Chris Han outlines the historical evolution of business growth, tracing data tools from ancient Mesopotamian record-keeping to modern Big Data and agentic AI. He highlights three major challenges organizations face in adopting AI agents: fragmented tool infrastructure, outdated business processes, and management frameworks that only account for human employees. To resolve these issues, Han introduces ThinkingAI's architecture—comprising personal agents, an Agent OS control panel, and a focus on 'atomic opportunities' over useless dashboards. Ultimately, he envisions a future where humans manage specialized agent teams to achieve true agentic growth while retaining final accountability.

## Atmosphere & style The presentation carries a reflective, pragmatic, and visionary tone, opening with a warm personal anecdote about the speaker's son encouraging him before the conference's final session. Han blends historical storytelling with sharp industry critiques, speaking with engaging authority on the pitfalls of traditional enterprise data while keeping the audience grounded. The energy feels collaborative and forward-looking, transforming a technical discussion on software architecture into an inspiring exploration of how daily work and leadership are being reshaped.

## Detailed walkthrough Here is a detailed, faithful walkthrough based on the transcript provided:

### [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) Personal Intro, Rebranding to ThinkingAI, and the Mesopotamia Story Chris Han opens by acknowledging that he is presenting during the very final session of a three-day conference, remarking that he is glad to see so many people still in attendance. Before taking the stage, his 10-year-old son tried to reassure him by saying, "Dad, you don't have to be nervous. You don't have to worry about it because this is your final session and nobody will be there." Han joked back that at least he would have his son there to listen. Han introduces himself as the co-founder and president of ThinkingAI, noting that the company underwent a rebranding in April. Previously known as ThinkingData, the company has operated for 11 years, primarily serving clients in the gaming industry—a sector known for moving faster than almost any other, where live campaigns and product updates happen weekly. The goal of the talk, titled "Agentic Growth: From Analytics That Report to Agents That Act," is to untap how agentic technology can drive business growth.

Han observes that everyone in the room carries a responsibility for driving business growth, whether in HR, product management, or revenue, asking themselves every morning: "How can I grow my business?" He draws a historical parallel to 5,000 years ago, noting that when humans first invented writing, it was not for poetry, prayers, or laws. Instead, writing originated as business records in ancient Mesopotamia to track grains, livestock, and commercial transactions. For most of human history since then, business growth relied on human gut feeling and experience—much like the early days of Las Vegas, where brave pioneers took wild risks to build casinos and achieve success through sheer luck or courage. However, as technology evolves, the fundamental question remains "how to grow the business," but the tools and methods used to answer it have completely transformed.

- **Company History:** ThinkingAI operated as ThinkingData for 11 years before rebranding in April. - **Industry Background:** The company's core experience stems from the gaming industry, characterized by fast-paced weekly product updates and live operations. - **Historical Fact:** The earliest human writing in Mesopotamia (5,000 years ago) was created to track business records (grain, livestock) rather than poetry or law.

---

### [04:14](https://www.youtube.com/watch?v=rhR3DBcPT70&t=254s) The Evolution of Data & Insights and the Shift to Agentic AI Han outlines the technological evolution of business growth over the last two decades. Around the year 2000, websites proliferated, and everyone claimed to be an "internet company." Next came the era of session tracking, where companies could monitor every scroll and interaction to see *what* was happening. Then came the Big Data era—powered by open-source frameworks like Hadoop and Apache Spark—which introduced real-time communications, dashboard visualizations, and machine learning libraries. This shift allowed companies to understand *why* events were happening and derive key actionable insights. In the gaming industry, where metrics like Day 7 retention are critical (if Day 7 retention is below 30%, the project is often abandoned), rapid application of insights directly into monetization and product strategy became essential.

Moving into the current era of AI, Han points out that while anyone can build a simple agent in seconds or talk about agents daily, a massive gap remains between raw Large Language Models (LLMs) and enterprise-grade agent deployment. In every technological era, powerful solutions bring even bigger underlying challenges: 1. **The Data Era:** Companies drowned in raw data without knowing which metrics actually mattered. 2. **The Insights Era:** Teams struggled with how to turn raw insights into concrete business execution. 3. **The Agent Era:** Companies now have the most powerful tools in human history, but struggle to untap them effectively.

Han highlights that many industries—such as real estate, insurance, and education—are still stuck in the early data-driven era. Only about 24% of companies consider themselves truly data-driven, while around 80% of enterprise data remains trapped in "very dark and deep servers" without being utilized. Data remains the essential foundation upon which the AI era must be built.

- **Key Retention Metric:** In gaming, if Day 7 retention falls below 30%, teams often scrap the entire project. - **Enterprise Data Adoption:** Only 24% of companies report being truly data-driven, and roughly 80% of data sits unused in enterprise servers. - **Big Data Stack Mentioned:** Hadoop and Spark are highlighted as foundational open-source frameworks for real-time analytics.

---

### [08:44](https://www.youtube.com/watch?v=rhR3DBcPT70&t=524s) Three Core Challenges: Tools, Process, and Management/Culture Han outlines the three main challenges organizations face when transitioning to an agentic inflection point:

1. **Tools & Infrastructure:** Modern enterprises use fragmented, siloed toolsets— attribution software, user analytics, monetization platforms, and middleware data tools—making cross-team alignment difficult. Simply adding an "agent layer" on top does not solve the underlying issue. There is a huge gap between an LLM and enterprise execution, requiring an "Agent OS runtime" that manages security, shared context, unified knowledge bases, and specific skills. 2. **Process:** Organizations must decide whether to integrate agents into existing business workflows or completely reshape and rethink their processes from the ground up. 3. **Management & Organizational Culture:** For over a century—from Taylor's Scientific Management to OKRs and Agile/Scrum—management theories have focused exclusively on directing *people*. With autonomous agents (and eventually physical robotics) joining teams, companies lack a clear framework to ensure agents align with core corporate values like customer priority.

- **Management Theories Mentioned:** Frederick Winslow Taylor's Scientific Management, OKRs (Objectives and Key Results), and Agile Scrum frameworks. - **Tooling Silos:** Attribution software, user analytics, and monetization tools often operate independently across different teams.

---

### [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) ThinkingAI Architecture: Personal Agents, Agent OS, and "Atomic Opportunities" Han explains how ThinkingAI tackles these challenges through a specialized technical architecture. Moving past standard agent demo UIs (which feature simple prompts alongside skill or context panels), ThinkingAI focuses on the underlying infrastructure behind the interface: - **Application Layer:** Contains both business agents and personal agents. Han shares a personal routine change: rather than spending 1.5 hours in routine status-update meetings, each team member keeps a work log. Before the meeting, a personal agent reviews everyone's updates and generates targeted questions. During the meeting, team members skip routine status updates and instead challenge each other to extract real value. - **Agent OS / Control Panel Layer:** Provides the secure runtime environment. It includes sandboxing, security governance, shared context, unified metrics definition, and domain-specific skill sets developed over 10+ years in analytics and live operations. Han notes that when clients complain their agents calculate metrics incorrectly, it is usually not a model issue, but rather a lack of a unified enterprise knowledge base that explicitly defines those metrics. - **Process & "Atomic Opportunities":** Han criticizes traditional data visualization, asserting that "90% of dashboards are useless" and waste hardware resources recalculating unused metrics every day. Instead of staring at dashboards, companies should focus on "atomic opportunities"—small, critical operational friction points. He shares an anecdote about checking into a hotel where he had to wait in line for two hours; had the hotel manager possessed an "atomic opportunity mindset," they would have reassigned staff immediately to fix the bottleneck, saving the customer relationship.

- **Architecture Layers:** Application Layer (Personal & Business Agents) and Agent OS Control Panel (Security Sandbox, Shared Context, Unified Knowledge Base, Skill Sets). - **Meeting Optimization Example:** Using personal agents to pre-read work logs and generate challenging questions, shifting 1.5-hour meetings from passive updates to active discussion. - **Dashboard Critique:** Asserts that 90% of traditional enterprise dashboards are unnecessary resource wastes.

---

### [17:21](https://www.youtube.com/watch?v=rhR3DBcPT70&t=1041s) The AI-Native Organization and ThinkingAI's Vision Addressing the concept of the "AI-native organization," Han admits that no one yet possesses a fully mature definition or blueprint for how it operates. However, everyday work routines are already shifting. Han shares his own morning routine: upon waking up, he uses voice dictation to assign tasks directly to his agent team, checking back an hour later on their progress. In this new paradigm, "everyone is a leader right now," directing agents that have clear roles and operational boundaries.

However, Han stresses that the ultimate accountability remains squarely on humans today. As platforms improve skill sets and unified knowledge bases, agents will earn higher levels of operational trust. There is still a long path ahead to achieving fully AI-native organizations. Han concludes by sharing ThinkingAI's ultimate vision: to empower every individual and business with a highly capable agent team to achieve true agentic growth.

- **Morning Routine Example:** Han uses voice dictation every morning to delegate tasks to his agent team, acting as a manager to autonomous agents. - **Human Accountability:** Emphasizes that accountability stays with humans while agents operate within defined boundaries and trust levels.

## Notable quotes > [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) Dad, you don't have to be nervous. You don't have to worry about it because this is your final session and nobody will be there. — Chris Han's son

> [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) Agentic Growth: From Analytics That Report to Agents That Act — Chris Han

> [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) How can I grow my business? — Chris Han

> [04:14](https://www.youtube.com/watch?v=rhR3DBcPT70&t=254s) very dark and deep servers — Chris Han

> [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) 90% of dashboards are useless — Chris Han

> [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) atomic opportunities — Chris Han

> [17:21](https://www.youtube.com/watch?v=rhR3DBcPT70&t=1041s) everyone is a leader right now — Chris Han

## Mentioned Chris Han, ThinkingAI, ThinkingData, Mesopotamia, Las Vegas, Hadoop, Apache Spark, Frederick Winslow Taylor, Agile, Scrum

## Video description Chris Han - Co-Founder & President, ThinkingAI

Every enterprise has spent a decade buying analytics that tell them what already happened. The winners in the next decade will be the ones that close the gap between insight and action, automatically. Join ThinkingAl Co-Founder Chris Han for a clear-eyed look at Agentic Growth, the move from dashboards that report on your business to autonomous agents that run it, analyzing behavior, diagnosing root causes, and executinggrowth and operations decisions in real time. Chris will show where autonomy delivers outsized returns, where humans stay in the loop, and how to keep your data sovereign and compliant when Al is making the calls. If your team is drowning in reports but starving for outcomes, this is the session to catch. You'll take away： • The shift from reporting to operating: a practical framework for agentic growth • Where autonomy pays off across growth and operations, and where humans stay in the loop • Sovereign, compliant Al: acting on your data without giving up control

[Source](https://www.youtube.com/watch?v=rhR3DBcPT70)

中文简报

ThinkingAI负责人称智能体操作系统需要统一指标知识库 企业落地AI往往在执行阶段受阻，原因是原始大语言模型缺乏统一的指标定义和安全的沙箱运行时环境。 Chris Han梳理了商业增长的历史演变，回顾了从古美索不达米亚记账手段到现代大数据及智能体AI的数据工具发展历程。他重点指出了企业在引入AI智能体时面临的三大核心挑战：碎片化的工具基础设施、过时的业务流程，以及仅考虑人类员工的管理框架。为应对这些难题，Han介绍了ThinkingAI的架构方案，该架构涵盖个人智能体、Agent OS控制面板，并主张聚焦“原子级机会”而非流于形式的仪表板。他最终展望了一个未来图景：人类通过管理专业化的智能体团队来实现真正的智能体驱动型增长，同时仍保留最终的权责归属。 Chris Han追溯了数据工具演进史，并指出企业应用AI智能体面临工具碎片化、流程过时和人力管理框架局限三大挑战。 ThinkingAI推出包含个人智能体与Agent OS控制面板的架构，主张聚焦“原子机会”而非无用的数据仪表板。 未来人类将通过统筹管理专业化智能体团队实现增长，同时保持人类自身的最终责任归属。

Reading summary

ThinkingAI co-founder Chris Han argues that enterprise growth requires shifting from passive analytics dashboards to autonomous agent execution, a transition currently stalled by fragmented infrastructure and management models designed strictly for human labor. The Evolution from Big Data to Agentic AI (4:14) Tracking business activity has evolved from ancient Mesopotamian commercial ledgers to modern digital telemetry. Han traces modern digital analytics across three phases: early session tracking that documented user behavior, the Big Data era driven by Hadoop and Apache Spark that explained user actions, and the present AI era focused on direct agent execution. In fast-paced sectors like gaming, rapid operational cycles are standard, and failing to achieve benchmarks such as a 30% Day 7 retention rate often results in project cancellation. Despite the availability of advanced Large Language Models, enterprise adoption remains constrained by foundational data issues. Han points out that roughly 80% of enterprise data sits idle in server storage, while only about 24% of companies consider themselves genuinely data-driven. Moving from insights to autonomous agents requires first resolving these underlying data accessibility and operational gaps. Infrastructure and Management Bottlenecks (8:44) Han identifies three primary obstacles preventing organizations from reaching an agentic inflection point: fragmented tooling, unresolved workflow integration, and an absence of agent-oriented management frameworks. Enterprise software stacks frequently isolate user analytics, attribution platforms, and monetization tools within separate operational silos, creating barriers that a standard LLM interface cannot bridge on its own. Simultaneously, management frameworks spanning the past century—including Frederick Winslow Taylor's Scientific Management, OKRs, and Agile Scrum—were designed exclusively to direct human workers. Han observes that enterprises currently lack operational standards to govern hybrid teams composed of humans, autonomous agents, and eventual physical robotics, particularly when ensuring adherence to core organizational priorities. Agent OS Runtime and Shared Metric Context (11:49) To address enterprise integration challenges, ThinkingAI relies on a layered architecture that decouples user-facing agents from an underlying Agent OS runtime. The runtime layer governs security sandboxing, shared context, and unified metric definitions. Han notes that when agents fail to calculate business metrics accurately, the error typically stems from missing or ambiguous enterprise knowledge definitions rather than model inference failures. The application layer incorporates both business agents and personal agents to alter operational workflows. In internal operations, Han notes that team members maintain work logs that personal agents review ahead of time to generate analytical questions. This practice replaces standard 1.5-hour status-update meetings with focused discussions centered on operational challenges. Targeting Atomic Opportunities Over Static Dashboards (17:21) Han criticizes traditional business intelligence practices, asserting that 90% of dashboards are useless and consume computing resources recalculating unused metrics. In place of static visualization, he advocates organizing workflows around 'atomic opportunities'—identifying and correcting immediate friction points as they occur, analogous to a service manager rapidly reassigning staff to clear customer check-in bottlenecks. In this operational model, managers direct specialized agent teams via daily task delegation. While this allows individual contributors to direct multiple agents, Han maintains that operational and strategic accountability remains entirely with human operators as trust frameworks and specialized agent capabilities continue to mature. Han acknowledges that a definitive model for the AI-native organization does not yet exist. Full implementation remains limited by the need for unified enterprise knowledge bases and security governance, leaving final accountability with human managers while agent platforms evolve toward operational maturity.

精炼阅读摘要

ThinkingAI 联合创始人兼总裁 Chris Han 指出，企业商业决策正在经历从被动报表分析向主动智能体行动（Agentic Growth）的转变，但要跨增大模型与企业级交付之间的断层，组织必须解决底层数据利用率低、工具孤岛以及缺乏人机协同管理体系等根本挑战。 数据演进与现实断层：从直觉到数据壁垒 (4:14) Han 回顾了商业增长工具的技术演进：人类从 5000 年前美索不达米亚记录商业交易的文字开始，经历了依靠直觉与运气的商业冒险，逐步发展到 2000 年左右的互联网网站会话追踪，再到依托 Hadoop 和 Apache Spark 等大数据框架实现对业务原因的深度洞察。在产品更新频繁的游戏行业，数据洞察对变现与留存至关重要，例如第 7 日留存率若低于 30%，项目通常会被直接放弃。然而，当前企业迈入 AI 时代面临着严峻的数据地基断层：目前仅有约 24% 的企业认为自身真正做到了数据驱动，约 80% 的企业数据仍封存在深层服务器中未被唤醒，房产、保险、教育等传统行业仍滞留在早期的数据探索阶段。 三大核心瓶颈：工具割裂、流程与管理滞后 (8:44) Han 认为，在大语言模型与企业落地执行之间存在巨大差距，组织转型面临三大维度的瓶颈。在工具与基础设施层面，企业内部归因分析、用户行为洞察与变现平台相互割裂，单纯在大模型之上叠加一层智能体无法解决根本问题，必须依赖包含安全沙箱、共享上下文与统一技能的运行环境（Agent OS）。在业务流程层面，企业面临是将智能体强行嵌入旧流程还是全面重构工作流的抉择。在管理与组织文化层面，从泰勒科学管理理论、OKR 到敏捷开发，上百年的管理学体系均针对“人”而设计，当自主智能体甚至实体机器人进入团队协作时，企业尚未建立能确保智能体与“客户第一”等核心价值观对齐的治理框架。 系统架构解法：Agent OS 与原子机会 (11:49) 针对落地痛点，Han 介绍了 ThinkingAI（深耕游戏数据分析 11 年的 ThinkingData 升级而来）的技术架构方案。该架构在应用层部署个人智能体与业务智能体，底层则依托具备安全沙箱、治理机制、共享上下文与领域技能集的 Agent OS 控制面板。Han 指出，智能体若出现指标计算偏差，根源往往是企业缺乏明确且统一的指标知识库，而非模型缺陷。在流程优化方面，Han 直言传统可视化报表中“90% 的仪表盘毫无用处”，仅是在浪费硬件资源重复运算。他主张将精力转向捕捉“原子机会”（即业务中的微观阻碍点），例如团队成员利用个人智能体提前阅读日志并自动生成提问，将 1.5 小时的例行汇报重塑为实质性的深度讨论，或者通过动态响应瓶颈改善客户体验。 AI 原生组织雏形与人类责任边界 (17:21) 在探讨“AI 原生组织”时，Han 坦言行业目前尚无清晰且成熟的操作蓝图，但个体工作模式已经在发生重构。他分享了自己的工作习惯：早晨通过语音指令向个人智能体团队派发具体任务并在后续跟进进度，形成“人人皆管理者”的新范式，员工负责指挥具备清晰职责边界的智能体。不过，Han 明确强调，当下的最终业务责任与问责机制依然完全由人类承担。随着底层系统不断沉淀统一知识库与专业技能，智能体才会在安全边界内逐步建立更高水平的运营信任，企业实现完全的 AI 原生化仍有一段漫长的路径需要探索。 Han 总结认为，向具备行动能力的智能体演进是商业增长的必然趋势，但企业当前面临模型能力与企业落地脱节、知识库标准缺乏及新型管理框架空白等挑战；在推动智能体团队赋能业务的同时，明确操作边界并维持人类的最终问责权，是推动智能体增长的关键前提。

Detailed notes — zh

## Background 该视频由 Ai4 频道于 2026 年 8 月 20 日发布在 YouTube 平台。内容为 ThinkingAI 联合创始人兼总裁 Chris Han 的主题演讲，探讨企业如何打破传统数据看板仅能事后汇报的局限，迈向由自主智能体进行行为诊断与实时决策执行的“智能体增长（Agentic Growth）”新阶段。 ## TL;DR Chris Han 回顾了商业增长从直觉驱动到数据分析、再到行动落地的演进历程，指出当前 Agent 真正嵌入企业业务面临着工具孤岛、流程重塑和组织管理对齐三大瓶颈。为此，ThinkingAI 打造了集成了安全沙箱、统一上下文和标准化指标技能库的企业级 Agent OS 架构。他直言当前 90% 的数据看板毫无用处，企业必须将重心转向捕捉全链路中的“原子机会”，通过人机协同构建人人皆可调度 Agent 的 AI 原生组织。 ## Atmosphere & style 演讲以一段温馨幽默的家庭父子互动轻松开场，随后迅速切入务实、干练的企业级技术与管理探讨。Chris 的表达坦率而敏锐，既有对行业盲目堆砌看板与泛滥工具的犀利剖析，又有立足于 11 年高频运营实战的底层架构推演。整场分享节奏明快、生动写实且极具前瞻感，让听众在接地气的案例中清晰感受到人机协同组织转型的紧迫与张力。 ## Key takeaways - [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) ThinkingAI 前身为深耕游戏数据分析 11 年的数数科技（ThinkingData），依托游戏行业极高频率的策略调整与运营迭代，积累了实时数据驱动与即时行动的实战方法论。 - [01:37](https://www.youtube.com/watch?v=rhR3DBcPT70&t=97s) 商业数据分析经历了记录“发生了什么”与诊断“为什么发生”的演进，当前 Agent 时代的核心在于填补洞察与行动之间的落差，但从快速搭建 Agent 到企业级无缝集成仍存在巨大鸿沟。 - [07:41](https://www.youtube.com/watch?v=rhR3DBcPT70&t=461s) 传统企业的数据驱动转型仍未完全走通，目前仅约 24% 的企业实现真正的数据驱动，近 80% 的企业数据仍处于沉睡未被利用的状态。 - [07:41](https://www.youtube.com/watch?v=rhR3DBcPT70&t=461s) 企业在多工具并存的环境下面临上下文与技能严重割裂的困境，单纯在现有工具上叠加 Agent 无法见效，必须依赖统一的企业级 Agent OS 底座支撑。 - [07:41](https://www.youtube.com/watch?v=rhR3DBcPT70&t=461s) Agentic 增长要求企业不能停留在对原有流程修修补补，而必须从头重新思考并对业务增长流程进行端到端的彻底重构。 - [07:41](https://www.youtube.com/watch?v=rhR3DBcPT70&t=461s) 现代管理理论（如泰勒制、OKR、Scrum）均基于人类员工设计，在人类与 Agent 协同的新型组织中，问责机制、信任建立与核心价值观对齐均面临新的挑战。 - [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) 大模型在业务指标上经常计算不准，其根本原因不在于 LLM 模型本身能力不足，而在于企业底层缺乏统一定义的标准知识库与指标体系。 - [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) 个人 Agent 可以颠覆低效的团队例会模式，通过会前自动研读成员更新并生成针锋相对的挑战性问题，将例会重心从单向汇报转向高价值思辨。 - [15:27](https://www.youtube.com/watch?v=rhR3DBcPT70&t=927s) 企业中高达 90% 的数据看板（Dashboard）属于低价值消耗，企业应摆脱看板依赖，聚焦捕捉并迅速化解业务流中的“原子机会”（微小业务摩擦点）。 - [15:27](https://www.youtube.com/watch?v=rhR3DBcPT70&t=927s) 实现数据洞察向增长行动的闭环，必须彻底打破数据洞察（BI）、运营与产品团队之间的组织孤岛与职能隔阂。 - [15:27](https://www.youtube.com/watch?v=rhR3DBcPT70&t=927s) 在 AI 原生组织中，每一位员工都将转变为管理者与业务调度者，通过向专属 Agent 团队分配工作来驱动业务闭环增长。 ## Detailed walkthrough ### [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) 开场轶事与 ThinkingAI 的演进之路 演讲开始时，ThinkingAI 的联合创始人兼总裁 Chris Han 调侃道，这是为期三天的峰会的最后一场分论坛，但现场依然座无虚席。他分享了一个温馨的家庭小插曲：登台前，他 10 岁的儿子坐在台下安慰他说：“爸爸，你不用紧张，反正这是最后一场了，台下肯定一个人都没有。”这引来了现场的一阵笑声。Chris 幽默地回应儿子：“至少我还有你来听。” Chris 接着介绍了公司的背景。ThinkingAI 刚刚在今年 4 月完成了品牌重塑（Rebranding），其前身为深耕数据分析领域长达 11 年的 ThinkingData（数数科技）。公司绝大部分早期客户来自游戏行业。游戏行业堪称全球产品迭代和运营活动变动最快的领域，几乎每周甚至每天都要更新活动和策略。正是这种高强度的业务场景，让团队在业务增长、数据分析和即时行动方面积累了大量实战经验。 - **品牌历史**：公司深耕行业 11 年，2024 年 4 月从 ThinkingData 升级重塑为 ThinkingAI。 - **行业根基**：在游戏等高频迭代场景中验证了实时数据驱动与精细化运营方法论。 --- ### [01:37](https://www.youtube.com/watch?v=rhR3DBcPT70&t=97s) 商业增长的本质与数据分析的历史演进 谈及商业增长的本质，Chris 提出了一个有趣的洞察：人类在美索不达米亚发明的最早文字，既不是为了写诗、祈祷，也不是为了制定法律，而是纯粹为了记录粮食、牲畜和交易等商业数据。人类摸索商业增长已有 5000 年历史，但在绝大部分时间里，企业增长主要依赖人类的直觉（Gut feeling）和个人胆量——就像当年在拉斯维加斯创立赌场的开拓者一样，只要足够大胆，碰上机遇就能成功。 然而，随着互联网与大数据技术的普及，增长的底层逻辑被彻底重构。回顾过去 20 年的技术发展脉络： 1. **2000 年前后（Web / 流量时代）**：互联网爆发，企业开始监测每次会话（Session）、滑动和页面交互，能够直观看到**发生了什么（What happened）**。 2. **大数据与可视化时代**：Hadoop、Spark、实时通信架构和机器学习库兴起，不仅能看到发生了什么，更能深度分析**为什么发生（Why it happened）**。 3. **行动落地时代（从洞察到行动）**：在游戏行业中，次日留存与 7 日留存（Day 7 Retention）至关重要。如果一款游戏的 7 日留存率低于 30%，整个项目往往会被直接叫停。企业必须快速将数据洞察转化为产品调整和 LiveOps 策略。 进入当前的 Agent 时代，虽然 LLM 技术爆发，每个人都能在几秒钟内搭建一个基础 Agent，但要将其无缝嵌入企业实际业务，仍存在巨大的鸿沟。 --- ### [07:41](https://www.youtube.com/watch?v=rhR3DBcPT70&t=461s) 智能体时代的转型瓶颈：工具、流程与组织管理 在迈向智能体增长（Agentic Growth）的过程中，企业面临着多重现实挑战。即使在普及多年的“数据驱动”阶段，很多传统行业（如保险、房地产、教育等）仍未完全走通：目前仅约 24% 的企业能真正称得上是“数据驱动型公司”，而企业内部有近 80% 的数据仍沉睡在黑暗、冰冷的底层服务器中未被唤醒。 Chris 剖析了 Agent 真正落地企业面临的“三大核心挑战”： ```mermaid flowchart TD accTitle: 三大核心挑战 subgraph S1["挑战一：工具孤岛与底层架构"] A["多工具并存 (归因/行为/变现)"] --> B["Agent 层上下文与技能割裂"] B --> C["缺乏统一 Agent OS 运行时"] end subgraph S2["挑战二：业务流程重塑"] D["单纯叠加 Agent 到现有流程"] E["端到端重构业务增长链路"] D -.vs.-> E end subgraph S3["挑战三：组织管理与价值观对齐"] F["百年管理学基础 (泰勒/OKR/Scrum)"] --> G["人类与 Agent 协同机制未知"] G --> H["问责制与信任体系重塑"] end ``` 1. **工具碎片化与上下文割裂（Tools & Context）**：企业内部往往同时运行着归因工具、用户行为分析工具、变现分析系统和各种数据中间件。跨团队沟通成本极高，直接在上方叠加一层 Agent 无法根本解决问题，必须提供统一的安全沙箱、上下文协同和知识库支撑。 2. **业务流程重塑（Process Reshaping）**：Agent 的引入不仅是在现有流程中打补丁，而是需要重新思考并彻底重构工作流。 3. **管理理论与文化对齐（Management & Alignment）**：从泰勒科学管理原理到 OKR、敏捷 Scrum，过去百年的管理哲学都是针对“人类员工”设计的。当 Agent 甚至未来的具身智能（Robotics）加入团队时，如何确保它们始终对齐“客户价值第一”的文化，目前业内尚无成熟标准。 --- ### [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) ThinkingAI 的 Agent 架构：从应用层到 Agent OS Chris 指出，市面上许多 Agent 产品展示千篇一律：左侧配置技能、知识库和上下文，右侧提供对话框，但这种界面表象并非关键，真正的核心在于底层的架构支撑。ThinkingAI 构建了一套完整的 Agent 企业级操作系统（Agent OS）： ```mermaid flowchart TD accTitle: ThinkingAI Agent 企业级操作系统架构 subgraph ApplicationLayer["应用层 (Application Layer)"] PA["个人 Agent (Personal Agents)"] BA["业务/运营 Agent (Business Agents)"] end subgraph AgentOSLayer["Agent OS 控制与运行时底座 (Agent OS Control Panel)"] SEC["安全与隔离沙箱 (Security & Sandbox)"] CTX["统一上下文引擎 (Unified Context)"] KB["标准化知识库与指标体系 (Unified Knowledge Base)"] SKILLS["预置行业技能集 (Analytics & LiveOps Skills)"] end ApplicationLayer --> AgentOSLayer ``` 在日常协作方面，Chris 分享了个人 Agent 改变例会流程的实践： 传统的远程例会往往长达 1.5 小时，所有人轮流做枯燥的工作汇报，效率极低。现在的全新流程是：每个人在工作日志中记录动态，在开会前，**个人 Agent 会预先阅读所有团队成员的更新，自动针对性地生成挑战性问题**。会议期间，团队成员不再复述琐事，而是直接相互推敲工作中的核心价值，极大提升了会议的产出。 针对很多客户抱怨“为什么大模型 Agent 算不对业务指标”，Chris 直言这并非大语言模型本身的能力缺陷，而是企业缺乏统一定义的标准化知识库（Unified Knowledge Base）。ThinkingAI 将过去 10 余年积累的深度数据分析和 LiveOps 技能固化为标准能力库，让 Agent 开箱即可投入实战。 --- ### [15:27](https://www.youtube.com/watch?v=rhR3DBcPT70&t=927s) 捕捉“原子机会”与构建 AI 原生组织 Chris 直言批评了当前的报表泛滥现象：“目前 90% 的企业数据看板（Dashboard）都是无用的。”企业每天消耗大量硬件计算资源去刷新成百上千个图表，却极少能从中提炼出有价值的商业洞察。企业应当摆脱看板依赖，转向关注业务全流程中的**“原子机会（Atomic Opportunities）”**。 Chris 举了一个亲身经历的生动例子：他在某家酒店办理入住时，由于前台处理缓慢，整整排了两个小时的队。如果酒店管理者具备敏锐的“原子机会意识”，就应在发现拥堵的第一时间调度人手化解痛点。这种糟糕体验会导致顾客发誓“这辈子再也不住这家酒店”，而快速抓住并修复这些微小的业务摩擦点，就能彻底扭转客户满意度。 商业增长链路必须是一个端到端的闭环： - **打破数据孤岛**：不能把业务割裂为独立的“洞察团队/BI 团队”、“运营团队”和“产品团队”。 - **人人成为管理者**：在 AI 原生组织中，每个人每天清晨通过语音听写向专属 Agent 分配工作任务，随后跟进产出，每个人都转变为业务调度者。 - **权责与信任边界**：虽然现阶段的最终责任与问责制（Accountability）仍落在人类身上，但随着技能库与知识库的成熟，人类与 Agent 之间的信任阶梯正在逐步建立。 演讲最后，Chris 总结了 ThinkingAI 的愿景：赋能每家企业都拥有极其强大的专属 Agent 团队，打破报表与行动之间的隔阂，实现真正的智能体驱动增长（Agentic Growth）。 ## Notable quotes > [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) 爸爸，你不用紧张，反正这是最后一场了，台下肯定一个人都没有。 — Chris Han的儿子 > [00:08](https://www.youtube.com/watch?v=rhR3DBcPT70&t=8s) 至少我还有你来听。 — Chris Han > [01:37](https://www.youtube.com/watch?v=rhR3DBcPT70&t=97s) 人类在美索不达米亚发明的最早文字，既不是为了写诗、祈祷，也不是为了制定法律，而是纯粹为了记录粮食、牲畜和交易等商业数据。 — Chris Han > [01:37](https://www.youtube.com/watch?v=rhR3DBcPT70&t=97s) 如果一款游戏的 7 日留存率低于 30%，整个项目往往会被直接叫停。 — Chris Han > [07:41](https://www.youtube.com/watch?v=rhR3DBcPT70&t=461s) 目前仅约 24% 的企业能真正称得上是“数据驱动型公司”，而企业内部有近 80% 的数据仍沉睡在黑暗、冰冷的底层服务器中未被唤醒。 — Chris Han > [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) 市面上许多 Agent 产品展示千篇一律：左侧配置技能、知识库和上下文，右侧提供对话框，但这种界面表象并非关键，真正的核心在于底层的架构支撑。 — Chris Han > [11:49](https://www.youtube.com/watch?v=rhR3DBcPT70&t=709s) 为什么大模型 Agent 算不对业务指标 — 客户 > [15:27](https://www.youtube.com/watch?v=rhR3DBcPT70&t=927s) 目前 90% 的企业数据看板（Dashboard）都是无用的。 — Chris Han > [15:27](https://www.youtube.com/watch?v=rhR3DBcPT70&t=927s) 这辈子再也不住这家酒店 — 顾客 ## Mentioned Chris Han, ThinkingAI, ThinkingData (数数科技), Hadoop, Spark, LiveOps, Agent OS, OKR, Scrum, 泰勒科学管理原理 ## Video description Chris Han - Co-Founder & President, ThinkingAI Every enterprise has spent a decade buying analytics that tell them what already happened. The winners in the next decade will be the ones that close the gap between insight and action, automatically. Join ThinkingAl Co-Founder Chris Han for a clear-eyed look at Agentic Growth, the move from dashboards that report on your business to autonomous agents that run it, analyzing behavior, diagnosing root causes, and executinggrowth and operations decisions in real time. Chris will show where autonomy delivers outsized returns, where humans stay in the loop, and how to keep your data sovereign and compliant when Al is making the calls. If your team is drowning in reports but starving for outcomes, this is the session to catch. You'll take away： • The shift from reporting to operating: a practical framework for agentic growth • Where autonomy pays off across growth and operations, and where humans stay in the loop • Sovereign, compliant Al: acting on your data without giving up control [Source](https://www.youtube.com/watch?v=rhR3DBcPT70)

Source quotes

[0:08] Chris Han的儿子: 爸爸，你不用紧张，反正这是最后一场了，台下肯定一个人都没有。 [0:08] Chris Han: 至少我还有你来听。 [1:37] Chris Han: 人类在美索不达米亚发明的最早文字，既不是为了写诗、祈祷，也不是为了制定法律，而是纯粹为了记录粮食、牲畜和交易等商业数据。 [1:37] Chris Han: 如果一款游戏的 7 日留存率低于 30%，整个项目往往会被直接叫停。 [7:41] Chris Han: 目前仅约 24% 的企业能真正称得上是“数据驱动型公司”，而企业内部有近 80% 的数据仍沉睡在黑暗、冰冷的底层服务器中未被唤醒。 [11:49] Chris Han: 市面上许多 Agent 产品展示千篇一律：左侧配置技能、知识库和上下文，右侧提供对话框，但这种界面表象并非关键，真正的核心在于底层的架构支撑。 [11:49] 客户: 为什么大模型 Agent 算不对业务指标 [15:27] Chris Han: 目前 90% 的企业数据看板（Dashboard）都是无用的。 [15:27] 顾客: 这辈子再也不住这家酒店

Source quotes

[0:08] Chris Han's son: Dad, you don't have to be nervous. You don't have to worry about it because this is your final session and nobody will be there. [0:08] Chris Han: Agentic Growth: From Analytics That Report to Agents That Act [0:08] Chris Han: How can I grow my business? [4:14] Chris Han: very dark and deep servers [11:49] Chris Han: 90% of dashboards are useless [11:49] Chris Han: atomic opportunities [17:21] Chris Han: everyone is a leader right now

Original transcript

[0:08] Thank you. Thank you for the introduction. Uh I  believe this is the final session of the 3 days [0:15] conference and uh there's still a lot of people  here to listen. Actually before this presentation, [0:23] my 10 yearear-old son is sitting there. He  said to me, "Dad, you don't have to be nervous. [0:29] you don't have to worry about it because  this is your final session and nobody will [0:33] be there. [laughter] I said to him like at  least we I have you to listen. Okay. And [0:40] uh I'm co-founder of thinking AI and uh we have a  pretty interesting name thinking AI. Actually this [0:48] April we just finished the rebranding. You know,  we were syncing data before and uh we already [0:55] been in the industry for 11 years and most of our  clients in the game industry. As you guys know, [1:03] gaming industry is probably the fastest industry  uh to change the live campaigns to update the [1:11] product and basically every week. So we learned  a lot of experience from the game industry and [1:20] today's topic agentic growth and from analysts  that report to agents that acts definitely we [1:28] want to untap the agentic technology how this  new power can help us to grow the business so [1:37] that's the topic I want to discuss today I mean  everybody in this room right now. In some ways, [1:48] you got to do something to be responsible  for your business growth. No matter you are [1:54] working on the human resource, no matter you  are working on the product or the revenue, [2:00] I think probably every morning you wake up the  top question from in your mind should be how can [2:08] I grow my business? How can I grow my you know  work? And probably for your for every personal [2:15] I think there should be something we got to do  to grow personal things actually you know for [2:22] the past 5,000 years this the you know very  interesting story I love and uh the writing [2:32] and the human invented writing it's not for the  first thing human wrote is not a poem it was not [2:40] a prayer It was not a law. It was like business  records in Mesopotamia and uh to record the [2:52] grains to record the livestocks and to record the  business. So writing was invented for business and [3:03] uh we got we already doing this for 5,000 years.  It's just for most of the time we doing this we [3:12] grow this by humans gut and humans experience and  just like Las Vegas it just did something you know [3:23] like no one ever's done before about casino about  gambling and we done something crazy here and uh [3:30] people clapped here and doing some fun things if  you were brave enough before I mean you have the [3:37] chance to make success luckily if you have that  you know your your moment of life but after you [3:47] know the b you know the technology updated every  new things come up and all of everything we want [3:56] to do change totally the question is there how to  grow the business but the ways to untap it and the [4:04] ways to solve This question is totally changed. So  let's take a like a quick look about the history [4:14] how we grow the business before around 2000 I  believe there were a lot of you know websites [4:22] and just like agents today and websites everywhere  you know back then 20 years ago everyone says like [4:31] they were internet companies but we have and after  that we have the data and we can monitor every [4:39] session, every scroll like every interactions we  have a lot of data and we can see what is what was [4:47] happening and after that we have a big data you  know I believe everybody may familiar with Hadoop [4:57] spark you know a lot of fantastic open-source  framework and right now they still popular and [5:04] still solving a lot of questions in terms of  real time communication and and that dashboard [5:11] visualization, machine learning libraries and  suddenly not only we can see what was happening [5:17] but also we can see why we understand and we  can see a lot of insights from that but what [5:26] after that you know because of you know gaming  industry is pretty young industry and it happens [5:33] so fast you have to adjust your product strategy  probably daily. So from the insights to action and [5:43] uh we learned a lot from our clients and one thing  because retention and monetization probably are [5:50] two most important metrics in the game industry  like day seven retention. If you don't have like [5:57] above 30% of day seven retention again probably  you're going to give up the whole project. So a [6:04] lot of insights from there we have we helped our  clients to apply those insights to their product [6:11] strategies to like their live life live campaign  strategies so we can see the improvement and what [6:20] about right now with the agents technology you  know before that we finished the whole look most [6:26] of the job is like by human but what can we do  that right now And uh with agentic technology [6:35] I still believe you know there are so huge gap  between the LM and of your agents. Everybody can [6:43] build agents in a second. Everybody can you know  speak a lot of agents you know daily but in terms [6:51] of the business or enterprise application we still  have a lot of gap. I'll I'll speak about that. [6:58] So there's this pattern I really you know liked  about it. You know in every era we have seems like [7:05] we have a very powerful solution but there was  always a bigger challenge behind it. When we have [7:13] the data you know we are so drawing in the data  we don't know which metric is important. When we [7:21] have the insights and we don't know how to apply  to your business, when we have the next step, [7:27] we want to do it faster and much more efficiently.  We are in this era agent era probably we are [7:34] facing the most powerful tool in the human  history. how we can untap that and uh let's [7:41] make me let me make an example even we talk about  data data driven you know like growth yeah I think [7:47] everybody is familiar with that uh conception I  talked to some insurance real estate and education [7:55] industry people these days they are still stuck in  the datadriven era like 20 24% of the company they [8:04] can say that like we datadriven business company  and uh about 80% of the data they're still sitting [8:14] in a very very dark and uh deep servers and we  don't know how to use that. I just want to make [8:21] those numbers as example. Even we are in the  agentic era even with we are in the AI era we [8:29] still like you know cannot fully untap the data.  What I mean is data you know if you have to figure [8:37] out the fundamental things you have to reach there  like step by step data is still the foundation of [8:44] the you know AI era. So when facing the inflection  point I want to you know talk about those three [8:52] challenges the first one and the tools. Yeah. And  uh think about your current you know enterprise [9:02] applications the tools in terms of attribution.  You may have your attribution software and you [9:09] may have your user behavior analytical tools and  you may have the monetization tools and you may [9:16] have another middleware data tools different  team working different tools and people you [9:23] know team across it's really hard to communicate  to you know transfer those insights across the [9:29] teams and people may think hey let's just add  an agent layer and we can solve the questions [9:36] But you know among the agents how we can align  the agents works on the same context and how we [9:43] can make sure the agents have the same skill sets  as they should have that's that's not a that's [9:50] a huge gap between the LM to your agents level  application you know your security your context [9:58] your knowledge base your skill we definitely need  to create a whole agents run time OS and you can [10:06] figure that out. And in terms of the like the  process once we build up appropriate tools and how [10:14] you apply those agents to your current process.  I think there are basically two perspectives and [10:24] if either like you can apply the agents to your  current process or you should reshape or reync [10:32] and shift to your totally current process.  And finally and you figured out the tools, [10:38] you figured out the process. I mean the management  and the team culture like for like 2B company like [10:47] us, we believe in you know customer value should  be always the top priority. But once the agents [10:54] joined in the like whole loop, how I can make sure  the agents follow the same value we have for the [11:03] past 100 years from tailor scientific management  to the OKR probably everyone nowadays apply OKR [11:12] to a organization and the scrum like the like  very very you know agile organization all the [11:19] philosophies we develop those theory based on  the how to manage people. But now with agents [11:28] join us nowadays agents but maybe future not very  far future we have a robotics with agents joining [11:37] in the company. How we can make sure everything  follow the same value. No one has the answer [11:42] yet and no one has a material philosophy and  methodology for that. We need to think about that. [11:49] So basically there are three challenges. So based  on those three challenges and how I think about us [11:57] and how thinking AI and done something about  that. Firstly the tools not you know I saw a [12:06] lot of demos probably every demo just looks like  the same. You can see you know on the you can see [12:13] at the left panel we have skill we have knowledge  base we have the context all the things and right [12:19] you have just come from to your agent but that's  not the most important thing the interface is [12:26] not important thing I think behind the interface  behind the product that's most important so that's [12:34] the gap I just talked about it and uh right  this is our architecture agent architecture [12:42] we have the agent like application layer and  you can have your personal agent you have your [12:48] business agent one thing I've been working on  like uh I want to change the totally meeting [12:55] flow nowadays I probably have for 20 years and in  a very very routine like update meeting everybody [13:05] just update your work but you know for probably  Most of team are fully remote for the one half [13:11] an hours. How that's a very valuable time. How we  can make sure those one half an hours to make the [13:18] most value. One thing I've been I've been trying  is everyone should have your like personal work [13:24] work log ple. So before the meeting we let our  personal agent to read everyone's updates and [13:34] come ups with the questions for this person. And  during the meeting we challenge each other and uh [13:41] we push each others to the limit and to challenge  the real value from your personal work. So you [13:47] don't have to just update your routine things  during the valuable one and a half hours. That's [13:53] I just want to make an example about the personal  agent. So in terms to building the real agentic [14:00] enterprise platform not only the application layer  but also the personal personal agent how we how [14:07] the agents can communicate each other behind  that there is the agent OS control panel and [14:14] uh you got to figure out the security issue and  you have to build a sandbox you got to figure out [14:20] the context so every agents can work on the you  know same context you got to work out the like [14:27] the knowledge base. So one thing you know I got  a lot of complaints to our from our clients why [14:33] should I my agents you know cannot calculate the  right metric. I think that's not the problem of [14:41] a big models and you should build your like your  unified knowledge base to define every metric and [14:49] uh I think agents should based on that to work out  also you have to you know to fill in this agents [14:57] with the right skills the skills because we have  done this business for over 10 years and we got [15:04] every skills we should have in terms of analytics  and live ops and we once you you know untap the [15:12] whole platform you have the skill set right away  which means your agents are ready to go and with [15:19] the skills so this is like the agent OS layer  we've been building and next one the process and [15:27] uh this is a processing I've been pitching to  our clients you know we are sitting on pies of [15:33] dashboards nowadays you have a lot of dashboards  you know in whatever whatever data visualization [15:39] tools you are using but few very few insights you  can generate from them I believe 90% of dashboards [15:47] are useless you don't have to calc actually  you don't have even to calculate to you know [15:52] waste those hardware resource to calculate  those metrics every every day and you just [15:58] have to care about the the most important to your  business growth that's that's you know you know I [16:05] uh I created I I said this you concept a lot of  times um like atomic opportunities with in your [16:13] app and in your business. I believe there are  a lot of atomic opportunities we can you know [16:19] improve. Uh just for example when I check in the  hotel um I won't say the name there are like so [16:29] long the line and I just wait in the line for  two hours if I am the manager of the hotel I [16:36] should send more people and to just adjust those  atomic opportunities. Think about that. I will [16:43] never you know register that hotel again in my  life and uh if if if the manager has the like [16:50] the atomic opportunity mindset to adjust something  and they make can change a lot of people's mindset [16:57] of this hotel. So all I want to say is you know  you need to build your insights to action this [17:06] whole business growth loop in in in one time. you  you cannot just hey this is a insight team this [17:12] is a VI team and this is a like live ops team  this a product team and it cannot be isolated [17:21] and f and I think I mentioned this I'll skip this  one and uh finally and uh I think nobody has the [17:29] mature answer for what is AI native organization  I believe everybody you know heard this a lot of [17:36] times I don't have the answer yet but the agents,  the AI change our like routine work gradually and [17:44] everyone is like the leader right now please  basically every morning when I wake up I speak [17:50] I use some like dictation tools I speak to my  agent every agents just you know my task tell [17:58] them and I'll check maybe one hour later and  everyone is a is a leader right now and the [18:03] agents has a roles has your boundaries and but you  know we still cannot figure out the responsibility [18:10] problem and that accountability stays at a human  uh today and uh once we figure out the skills and [18:18] knowledge base I think agents has some kind of  level of trust. So yeah I believe there is still [18:25] a long way to go to build a real AI and native  organization. Yeah, that's all that's that's [18:31] my presentation and finally I think I thinking as  vision or vision we want to enable everybody have [18:39] a very powerful like agent team help you guys  to reach real like agency growth. Thank you.