Dcard Cuts Workflow Handoffs With AI Agents, Spins Out Platform GNTC

Turning a 20M-MAU Company into Agent-Native With GNTC — Ai4

Ai4

Enterprise AI investments fail to generate positive returns because organizations layer tools onto legacy pipelines instead of redesigning workflows to eliminate cross-department handoffs.

AI-assisted summary of Turning a 20M-MAU Company into Agent-Native With GNTC — original source. Check the original for full context.

Source published: 2026-08-20

Detailed notes

## Background Published by Ai4 on YouTube on August 20, 2026, this presentation features Kytu Lin, founder and CEO of Dcard and enterprise AI platform GNTC. Lin shares how Dcard bridged the "Agent Gap" to transform a 20-million-user social platform into an agent-native organization where non-engineering teams built hundreds of internal agents, turning those operational playbooks into an enterprise platform.

## TL;DR Addressing the industry reality that 95% of enterprise AI investments fail to generate positive ROI, Lin argues that the problem stems from inserting AI into legacy workflows plagued by cross-departmental handoffs. At Dcard, restructuring operations around an agent-native architecture led to over 100 production agents, a 90% employee adoption rate, and over 1,000 internal tools built across HR, finance, marketing, and legal within months. By compressing multi-tier value chains and using semantic data layers to drastically reduce new-hire time-to-value from months to days, companies can break the ROI deadlock. Lin concludes that the ultimate bottleneck in enterprise AI is not model capability, but human mindset and prompt architecture skills, making continuous workforce upskilling essential.

## Atmosphere & style The presentation has the pragmatic, authoritative rhythm of a seasoned founder-operator sharing hard-earned operational playbooks from the enterprise trenches. Lin speaks with crisp clarity and strategic urgency, stripping away superficial AI hype in favor of concrete workflow redesigns and structural bottlenecks. It feels like an executive masterclass that balances high-level organizational theory with tactical engineering architecture, leaving listeners focused on the human side of technological transformation.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) An MIT report indicates that 95% of enterprise AI investments fail to generate positive ROI due to superficial pilots rather than deep workflow redesign. - [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) Dcard restructured its operations around AI agents, deploying over 100 production agents with a 90% workforce adoption rate. - [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) Lin founded GNTC to bring this agent-native playbook and platform to enterprise clients across banking, telecoms, and manufacturing. - [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) The primary cause of AI failure is layering tools on top of legacy workflows without addressing handoff friction, status meetings, and multi-tier approvals. - [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) Companies face an adoption trap where leadership demands proven ROI before overhauling workflows, but ROI cannot materialize without redesigning the workflow first. - [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) An agent-native architecture relies on a semantic data layer that captures institutional context, company jargon, and operational ontologies. - [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) Enterprise AI success should be measured by time-to-value rather than token consumption or trailing financial metrics. - [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) Agentic onboarding shortens new-hire ramp-up time from months to days by dynamically generating tailored tutorials and company context on demand. - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) Agentic tooling compresses entire multi-role value chains—such as marketing, design, and engineering—into a single domain expert assisted by agents. - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) Dcard's non-engineering departments built over 1,000 internal applications within three to four months, fostering a widespread builder culture. - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) As technical barriers fall, human capability and prompt architecture become the primary constraints, making human upskilling more critical than agent development. - [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) Becoming an agent-native enterprise requires three core pillars: redesigning workflows, optimizing for time-to-value, and prioritizing continuous human development.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) Solving the Enterprise AI ROI Problem at Dcard and GNTC

Kytu Lin, founder and CEO of Dcard (a social platform serving over 20 million monthly active users) and enterprise AI platform GNTC, begins by addressing a sobering industry reality: an MIT report highlights that 95% of enterprise AI investments fail to generate positive ROI. Lin argues that this problem is entirely solvable. Rather than running superficial pilot tests, Dcard restructured its core enterprise operations around AI agents over the past year, successfully deploying over 100 production agents and achieving a 90% adoption rate across all employees.

Following the success of this internal transformation, Lin launched GNTC to help traditional enterprises—particularly banks, telecommunications providers, and manufacturing conglomerates across Asia—replicate this playbook. GNTC combines platform software and hands-on deployment services to guide organizations through the technical and cultural shifts needed to become truly "agent-native."

- **Scale of Operation:** Dcard supports over 20 million active users. - **Enterprise Adoption Metrics:** Shipped over 100 production agents with a 90% workforce adoption rate. - **GNTC Target Sectors:** Financial services, telecoms, and industrial manufacturers across the Asia-Pacific region.

---

[01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) The Agent Gap and the Deadlock of Workflow Redesign

Lin identifies the primary reason why standard AI implementations fail to move the needle: organizations simply layer AI tools on top of unchanged legacy workflows. The bulk of organizational drag does not stem from individual execution speed, but from "handoffs"—cross-functional status meetings, coordinating tasks between departments, and waiting for multi-tiered management approvals. Simply inserting an AI tool into an existing pipeline leaves these friction points intact, resulting in negligible ROI.

```mermaid flowchart TD accTitle: Workflow Redesign vs Traditional Handoffs subgraph Legacy["Traditional Pipeline (High Friction)"] A["Task Initiation"] --> B["Cross-Team Meetings"] B --> C["Awaiting Approvals"] C --> D["Delayed Output"] end

subgraph AgentNative["Agent-Native Pipeline (Shortened)"] E["Domain Owner"] --> F["Agent Orchestration & Data Layer"] F --> G["Direct Value Delivery"] end ```

True gains require radically redesigning and compressing the workflow itself. However, organizations often fall into an adoption trap: leadership requires proven ROI before committing to workflow overhauls, but meaningful ROI cannot occur without redesigning workflows first. This creates a cultural deadlock where initial pilot projects fail to generate enough momentum to break the loop.

> "The trap is not only technical, but it's also cultural. Real value comes from redesigning workflow, not just adding AI."

---

[02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) Technical Architecture: Onboarding Agents and Time-to-Value

To demonstrate how an agent-native architecture transforms operations, Lin shares the universal challenge of employee onboarding. Traditional onboarding is inherently high-friction: senior staff lose significant bandwidth mentoring recruits, coaching quality varies widely, and institutional knowledge is constantly lost when key personnel leave. In an agent-native company, agents become the primary interface for institutional context. Because the platform possesses a comprehensive semantic understanding of the company's internal jargon, workflows, and documentation, it dynamically synthesizes custom tutorials tailored to any new hire's role on demand.

This approach fundamentally shifts how companies measure the success of their AI initiatives. Instead of tracking token consumption or waiting for trailing financial metrics, Lin advises enterprise leaders to focus on "time-to-value"—the speed at which a new hire or initiative delivers tangible business output. While major tech companies historically relied on two-month onboarding bootcamps, agent-assisted new hires can now deliver production value in days.

```mermaid flowchart LR accTitle: Agent-Native Enterprise Architecture subgraph Context["Data Foundation"] A["Company Data & Ontology"] --> B["Semantic Data Layer"] end

subgraph Platform["Core Platform"] B --> C["Agent Orchestration"] C --> D["Internal App Deployment"] end

subgraph EndUsers["Execution"] D --> E["Cross-Functional Builders"] end ```

- **Core Technical Pillars:** - **Semantic Data Layer:** Maps internal institutional knowledge, specialized vocabularies, and enterprise ontologies. - **Agent Orchestration Layer:** Coordinates agent reasoning, tool usage, and process execution. - **Internal Deployment Infrastructure:** Enables rapid self-service creation and publishing of internal tools. - **Key Performance Indicator:** Shortening time-to-value from months to days acts as the most reliable leading indicator for long-term ROI.

---

[05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) Redesigning Value Chains and Cultivating a Builder Culture

Examining internal value chains reveals massive opportunities for team consolidation. In traditional web development or landing page creation, delivery requires a sequential chain of marketing managers, product managers, UI/UX designers, engineering managers, and software engineers. With agentic tooling, a single marketer paired with an orchestration agent can design, generate, and deploy production pages independently. The same compression applies to financial modeling, business intelligence queries, and legal reviews.

```mermaid flowchart LR accTitle: Value Chain Compression subgraph Traditional["Traditional Value Chain"] direction LR A["Product Manager"] --> B["Designer"] B --> C["Engineer"] C --> D["Production Deploy"] end

subgraph Compressed["Agent-Native Value Chain"] direction LR E["Domain Marketer / Builder"] --> F["Web Agent"] F --> G["Immediate Production Release"] end ```

Lin emphasizes that the future of work belongs to a "builder culture" where domain experts across HR, legal, finance, marketing, and sales directly ship functional software features. As the technical barriers to software creation collapse, human capability becomes the governing constraint. Enterprises must invest heavily in upskilling employees on prompt architecture, systems thinking, and product evaluation based on market feedback.

> "The agent possibility is infinity right now. But the mindset and the brain will limit how you prompt the agents. Human development is much more important than agent development."

- **Internal Velocity:** Dcard built over 1,000 internal applications across non-engineering departments within three to four months. - **Productization:** The underlying infrastructure powering these workflows has been spun out as a standalone enterprise SaaS platform.

---

[08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) Summary: Workflow Redesign, Time to Value, and Human Development

Lin concludes with a succinct summary of the core principles required to make an enterprise agent-native:

1. **Redesign Workflows:** Eliminating cross-department handoffs and restructuring end-to-end value chains is essential to breaking the ROI deadlock. 2. **Optimize for Time-to-Value:** Evaluate AI tooling by how rapidly it accelerates business delivery rather than tracking token usage or theoretical cost savings. 3. **Prioritize Human Development:** Provide continuous training and learning loops to help employees master agent prompting and product design, ensuring the workforce's ambition keeps pace with model capabilities.

## Notable quotes > Everyone now has PhD-level AI — so why do most companies still work exactly the same way? — Kytu Lin

> the people who understand the problems best are rarely the ones allowed to build the fixes. — Kytu Lin

> [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) The trap is not only technical, but it's also cultural. Real value comes from redesigning workflow, not just adding AI. — Kytu Lin

> [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) The agent possibility is infinity right now. But the mindset and the brain will limit how you prompt the agents. Human development is much more important than agent development. — Kytu Lin

> [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) Redesign Workflows: Eliminating cross-department handoffs and restructuring end-to-end value chains is essential to breaking the ROI deadlock. — Kytu Lin

> [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) Optimize for Time-to-Value: Evaluate AI tooling by how rapidly it accelerates business delivery rather than tracking token usage or theoretical cost savings. — Kytu Lin

> [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) Prioritize Human Development: Provide continuous training and learning loops to help employees master agent prompting and product design, ensuring the workforce's ambition keeps pace with model capabilities. — Kytu Lin

> [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) 95% of enterprise AI investments fail to generate positive ROI. — Kytu Lin

## Mentioned Kytu Lin, Dcard, GNTC, MIT, Ai4

## Video description Kytu Lin - Founder & CEO, GNTC

Everyone now has PhD-level AI — so why do most companies still work exactly the same way? I'm Kytu Lin, founder and CEO of Dcard, a 20-million-user social platform, and I think the reason is what I call the Agent Gap: the people who understand the problems best are rarely the ones allowed to build the fixes. To close it I became my own company's forward-deployed engineer — the result was 600+ internal agents built by HR, legal, finance, and sales, not just engineers. In this talk I'll walk through the three hardest problems we had to solve: making scattered company context usable by agents, governing what agents are allowed to do, and turning first wins into an adoption flywheel. If you're trying to move AI from impressive pilots to real, company-wide adoption, you'll leave with the playbook — and the story of how these lessons became the enterprise agent platform we now build for other companies.

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

中文简报

Dcard借AI Agent减少工作流交接，分拆出GNTC平台企业AI投资未能产生正向回报，是因为机构只是在传统流程上叠加工具，而非通过重塑工作流来消除跨部门交接。针对95%的企业AI投资未能产生正向ROI的行业现状，Lin指出，问题根源在于将AI直接套入充斥着跨部门交接的传统工作流中。在Dcard，围绕Agent原生架构重塑运营促成了100多个生产级Agent的部署与90%的员工采用率，并在数月内跨人力、财务、市场和法务等部门构建了1000多个内部工具。通过压缩多层价值链，并利用语义数据层将新员工创造价值所需的时间从数月大幅缩短至数天，企业得以打破ROI僵局。Lin总结认为，企业AI的终极瓶颈并非模型能力，而是人类的思维模式与提示词架构技能，因此持续提升员工技能至关重要。 MIT报告指出95%的企业AI投资因仅停留在浅层试点而未能产生正向ROI；Dcard通过围绕AI Agent重塑核心业务流程，成功上线超100个生产级Agent并实现90%的全员采用率。 (0:04) AI落地的核心障碍在于将工具直接叠加在传统流程之上，未能消除跨部门同步与多层审批等交接摩擦，从而陷入需要先证明ROI才肯重构流程的落地陷阱。 (1:35) 企业衡量AI成效应聚焦于“价值实现时间”（Time-to-Value）而非Token消耗量，例如通过语义数据层构建定制化培训，将新员工从入职到交付价值的周期从数月缩短至数天。 (2:43) Dcard推动非技术部门构建了上千个内部应用，表明端到端价值链正被Agent高度压缩，未来企业的核心瓶颈将从技术门槛转向员工的Prompt设计与系统思维能力。 (5:44)

Reading summary

Dcard and GNTC founder Kytu Lin argues that enterprise AI investments fail primarily because organizations layer tools onto unchanged legacy workflows rather than eliminating structural handoffs. To unlock positive ROI, Lin contends that enterprises must compress their operating value chains, benchmark performance on time-to-value, and upskill employees to build their own tools. Resolving the Enterprise AI Adoption Deadlock (1:35) Citing an MIT report indicating that 95% of enterprise AI investments fail to generate positive ROI, Lin claims the shortfall stems from organizational friction rather than individual execution speed. Traditional enterprise pipelines lose momentum through cross-functional coordination, status meetings, and multi-tier management approvals. Simply inserting AI tools into unchanged pipelines leaves these handoffs intact, producing negligible economic returns. Lin identifies a cultural deadlock where corporate leadership demands proven ROI before authorizing process overhauls, yet meaningful gains cannot occur without redesigning workflows first. At Dcard, a social platform with over 20 million monthly active users, breaking this loop involved restructuring enterprise operations around agents, resulting in over 100 production agents deployed and a 90% workforce adoption rate. Lin launched GNTC to help traditional enterprises—including banks, telecom operators, and industrial manufacturers in the Asia-Pacific region—replicate this operational model. Architecture Grounded in Time-to-Value (2:43) Lin uses employee onboarding to demonstrate how an agent-native architecture replaces legacy friction. Standard onboarding relies heavily on senior staff mentorship, yielding inconsistent coaching and institutional knowledge loss during employee turnover. In an agent-native setup, a semantic data layer maps enterprise ontologies, documentation, and internal jargon, enabling autonomous agents to synthesize tailored, role-specific tutorials dynamically. Under this model, Lin advises enterprise leaders to replace metrics such as token consumption or lagging financial returns with 'time-to-value'—the speed at which an initiative or new hire delivers tangible business output. While conventional tech onboarding programs historically required up to two months, Lin claims dynamic agent guidance enables recruits to deploy production value within days, serving as a reliable leading indicator of long-term operational returns. Compressing Value Chains into Builder Cultures (5:44) Agentic infrastructure enables organizations to consolidate sequential, multi-department production chains into single-operator workflows. In standard software development, launching a landing page requires coordination across product managers, UI/UX designers, engineering leads, and developers. Lin notes that an orchestration agent allows an individual marketer to design, generate, and deploy production pages independently, with similar compression possible in financial modeling, legal review, and business intelligence queries. As technical barriers to software creation collapse, Lin asserts that human conceptual capability becomes the governing operational bottleneck. Within three to four months, Dcard's non-engineering personnel built over 1,000 internal applications, prompting the company to spin out the underlying platform as a commercial enterprise SaaS product. Lin stresses that sustaining this velocity requires continuous investment in employee training on prompt architecture, systems thinking, and product evaluation to ensure workforce ambition keeps pace with model capabilities. Lin concludes that capturing real economic value from AI requires treating workflow restructuring, rapid time-to-value, and human upskilling as interdependent priorities. However, this strategy relies on the assumption that organizational playbooks proven in digital-native environments like Dcard can be replicated in heavily regulated legacy sectors such as banking and manufacturing.

精炼阅读摘要

Dcard 创始人兼首席执行官林裕钦（Kytu Lin）指出，企业 AI 投资回报率低的主要原因在于简单地将 AI 叠加于陈旧工作流之上；企业必须通过重构端到端流程、打破跨部门交接瓶颈并建立自主构建文化，才能实现“智能体原生”（Agent-Native）转型。企业AI面临的ROI困境与采用陷阱 (0:04) 林裕钦引用麻省理工学院（MIT）的一份报告指出，95% 的企业 AI 投资未能产生正向 ROI。他认为这一问题完全可以通过工作流转型来解决。企业应用 AI 见效甚微的核心根源，在于管理层往往仅将 AI 工具叠加在原封不动的传统流程之上，但组织的实际内耗主要来自跨部门状态会议、交接等待以及多层级审批等中间摩擦，单点加速员工个人执行力无法消除这些瓶颈。由此企业容易陷入一种文化与技术兼具的“死结”陷阱：领导层往往要求在看到明确 ROI 证据后才批准流程改造，但若不预先彻底重构工作流，企业根本无法获得具有实质意义的业务回报，导致试点项目因缺乏动力而停滞不前。 Dcard落地实践与GNTC服务扩展 (0:04) 面对企业普遍面临的 AI 落地困境，拥有超过 2000 万月活跃用户（MAU）的社交平台 Dcard 在过去一年中重构了内部核心运营体系。该团队正式部署了超过 100 个生产级智能体，在全体员工中达成了 90% 的采用率，并在三到四个月内由非技术部门自主搭建了逾 1000 个内部应用程序。在完成内部验证后，林裕钦推出了面向外部的 GNTC 平台，将这套方法论底层的基础架构产品化为独立企业 SaaS 服务。GNTC 结合平台软件与上门交付服务，主要面向亚洲地区的传统行业巨头，包括银行金融服务、电信运营商以及工业制造集团，协助他们推进技术与组织文化变革。技术三要素与以价值实现时间为衡量指标 (2:43) 智能体原生的企业技术架构包含三大技术支柱：映射企业内部专业词汇和业务本体的“语义数据层”、负责协调推理和工具调用的“智能体编排层”，以及支撑全员快速发布内部工具的“部署基础设施”。这种结构能将静态的企业背景知识转化为可直接交互的操作界面。在成效衡量标准上，林裕钦主张以“价值实现时间”（Time-to-Value）替代传统的 Token 消耗量或滞后的财务数据。以新员工入职培训为例，传统大型科技公司通常依赖长达两个月的培训营，且资深员工传帮带成本高、质量参差；而在接入能动态生成角色教程的智能体后，新员工可在数天内交付生产级业务成果，这种周期压缩构成了判断长期 ROI 的关键先行指标。价值链极限压缩与全员构建者文化 (5:44) 智能体编排对企业价值链产生了直接整合效应。在传统的网页开发等流程中，需求交付往往需要依次经过营销经理、产品经理、UI/UX 设计师、工程主管和软件工程师等多道环节。而在智能体支持下，单名业务营销人员配合网页智能体即可独立完成设计、生成并直接发布上线。相似的端到端压缩同样适用于财务建模、商业智能查询以及法务合规审查。林裕钦强调，当软件开发的工程门槛大幅降低后，人的认知与思维模式将成为核心约束。未来的工作模式属于“构建者文化”，即人力资源、法务、财务和销售等领域的业务专家直接交付可用软件功能。因此，企业对员工的投资重心必须转向 Prompt 架构设计、系统性思维以及基于市场反馈的产品评估能力。林裕钦总结认为，企业实现智能体原生的关键在于打破跨部门移交、以缩短交付周期为考核核心，并将人力发展置于智能体技术演进之上。然而，传统企业在组织惯性阻碍下如何平衡前期试错成本与激进的流程裁撤，仍是有待实践检验的现实挑战。

Detailed notes — zh

## Background 本视频由 YouTube 频道 Ai4 于 2026 年 8 月 20 日发布，主讲人为社交平台 Dcard 及企业级 Agent 平台 GNTC 的创始人兼 CEO Kytu Lin。演讲围绕如何跨越“Agent 鸿沟”展开，分享了他亲自担任前线部署工程师、带领全员开发数百个内部 Agent，并将这套实战经验沉淀为 GNTC 企业平台以驱动全公司乃至外部企业实现 Agent 原生化转型的完整方法论。 ## TL;DR 针对 MIT 报告中“95% 企业 AI 投资缺乏实质 ROI”的行业困境，Kytu Lin 指出单纯嵌入 AI 聊天机器人无法见效，必须彻底重构端到端的工作链路以打破组织死锁。在 Dcard 实践中，团队依托 Agent 编排层与语义数据层等底座，由非技术员工主导构建了超过 1,000 个内部业务应用，并以“价值交付时间”（Time to Value）取代传统的 Token 消耗量作为核心牵引指标。最终，企业转型的关键不仅在于技术构建，更在于打破部门孤岛、推行“人人皆为 Builder”的敏捷文化与对员工认知学习闭环的长期投资。 ## Atmosphere & style 整场演讲呈现出务实、笃定且充满实战洞察的技术布道氛围。Kytu Lin 抛弃了浮夸的行业概念，以一线操盘手的视角直击企业 AI 落地的真实症结与组织阻力。叙事节奏紧凑洗练，从剖析 AI 投资回报率低的普遍痛点，逐步层层推进到底层架构拆解与组织文化重塑，给听众带来强烈而清晰的认知冲击与破局信心。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) MIT 报告显示高达 95% 的企业 AI 投资缺乏实质性 ROI，症结在于多数企业仅局部嵌入 AI 插件，而非系统性重塑为 Agent 原生组织。 - [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) Dcard 探索出 Agent 原生落地路径，在生产环境部署超 100 个自主 Agent，实现了全员 90% 的日常 AI 采纳率，并孵化出 GNTC 赋能更多企业客户。 - [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) 传统组织效率损耗绝大部分源于跨部门交接、拉会对齐与审批等待，单纯给旧流程增加 AI 工具无法带来生产力跃升，必须大刀阔斧重构端到端工作流。 - [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) 企业转型面临“组织死锁陷阱”：缺乏成功案例导致缺乏信任，进而不敢打破现状重组流程，这本质上是一场深层的组织文化挑战。 - [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) 放弃 Token 消耗等表面数据，企业应将“价值交付时间”（Time to Value, TTV）作为衡量 Agent 原生成效的核心牵引指标。 - [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) Agent 原生架构依赖三大技术基石：Agent 编排层、基于本体（Ontology）的语义数据层以及内部业务应用快速部署系统。 - [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) 借助理解全公司上下文与行业黑话的 Agent，新员工入职培训可生成高度个性化的即时指引，将新人交付业务价值的周期从数月压缩到以天计算。 - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 传统冗长产研链路被打破，非技术人员（如市场、销售）也能借助 Agent 独立完成端到端的业务功能开发与上线。 - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 未来组织分工将从精细细分的螺丝钉岗位转向全面的“Builder（构建者）文化”，要求员工兼具市场洞察、设计逻辑与交付能力。 - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 在底层基座支撑下，Dcard 各职能部门在短短三至四个月内自发创建并交付了超过 1,040 个内部业务应用。 - [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 决定 AI 效能上限的是员工的认知与提问能力，因此企业对人才技能与学习闭环的投资远比单纯构建 Agent 更为关键。 - [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) Agent 原生化落地的三大抓手：打破死锁重构工作流、以价值交付时效为度量准绳、持续推动员工认知与技能升级。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) 从拥有 2000 万用户的 Dcard 到 GNTC：破解 95% 企业 AI 无 ROI 的困局作为拥有 2000 万月活跃用户的社交平台 Dcard 以及企业级 Agent 平台 GNTC 的创始人兼 CEO，Kytu Lin 开门见山地指出了当前企业 AI 落地面临的严峻现实：根据 MIT 的研究报告，高达 95% 的企业在 AI 上的投资根本没有带来预期的投资回报率（ROI）。这种“只砸钱不见效”的现象已经成为整个行业的普遍困境。然而，Kytu 结合自己公司的激进变革给出了截然不同的答案——这一难题不仅完全可以解决，而且团队已经在过去一年多的时间里，将这套机制彻底打造成型。他们并非只是局部测试或试水概念验证，而是系统性地将整个公司的工作流程全部重塑为“Agent 原生”（Agent-Native）。在 Dcard 内部，他们已经向生产环境交付了超过 100 个自主 Agent，全公司有 90% 的员工将 AI 深度融入日常工作流。正因验证了这套打法的可行性，他们创立了新公司 GNTC，通过“平台产品加配套服务”的双轨模式，专门帮助亚洲范围内的银行、电信运营商以及大型制造业客户完成 Agent 原生化转型。Kytu 强调，这次分享的核心目的正是拆解这段转型旅程：揭示阻碍 AI 产生实际价值的“Agent 鸿沟”（Agent Gap），剖析实现组织进化所必需的底层技术架构，以及探讨比算法更关键的组织文化变革。 - **行业现状**：MIT 报告显示 95% 的企业 AI 投资缺乏实质性 ROI。 - **内部实践成效**：生产环境部署 100 多个业务 Agent，员工日常工作流的 AI 采纳率高达 90%。 - **GNTC 业务布局**：以“软件平台 + 实施服务”模式，重点赋能亚洲地区的银行、电信与制造业龙头。 --- ### [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) 工作流陷阱与 Agent 鸿沟：光加 AI 无法提升效率，必须重构协作流许多企业之所以在引入 AI 后毫无起色，根源在于他们掉入了一个致命的认知误区：单纯将大模型插件或聊天机器人塞进原有的业务流程中，却从未动手重构整个工作链路。Kytu 指出，传统组织中绝大部分的时间和精力并不是消耗在“具体任务的执行”上，而是无休止地浪费在跨部门交接（handoffs）、拉会对齐、反复澄清需求以及等待各级领导审批的停滞时间里。如果现有的冗长链路纹丝不动，仅仅让某个环节的员工用 AI 写一段草稿，对整体产出效率的提升微乎其微。真正的突破在于敢于大刀阔斧地“重新设计并缩短工作流”，才能看到 ROI 的飞跃性倍增。“真正的价值源于对工作流的重构，而非单纯嵌入 AI。”然而，这就引出了更为棘手的“组织死锁陷阱”：团队由于看不到成功的业务范例，因而无法建立起对 AI 的深度信任；因为缺乏信任，管理者就绝不敢轻易打破现状去重组流程；而流程不改，就永远无法孕育出颠覆性的业务成果。这种恶性循环证明，企业迈向 Agent 原生时遇到的阻碍从来不只是技术选型问题，本质上是一场深层的组织文化挑战。 ```mermaid flowchart TD accTitle: 传统工作流与重构后 Agent 原生工作流的对比 subgraph Traditional["传统线性工作流：损耗严重"] A1["业务需求发起"] --> A2["反复开会与对齐"] A2 --> A3["等待多层人工审批"] A3 --> A4["多职能人工交付"] end subgraph AgentNative["重构后的 Agent 原生流：端到端闭环"] B1["业务人员定义目标"] --> B2["Agent 读取公司上下文"] B2 --> B3["自主生成 / 调度执行"] B3 --> B4["敏捷上线与反馈"] end ``` --- ### [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) 技术落地案例：用 Agent 重塑新人入职，以“Time to Value”取代 Token 消耗指标为了具象化展示 Agent 原生组织的技术形态，Kytu 以各家公司都会遇到的高频痛点——“新员工入职培训”（Onboarding）为例进行了深入剖析。在传统模式下，新人入职往往伴随着极高的摩擦成本：老员工需要耗费大量本职工作时间进行带教，导致团队短期生产力明显下滑；不同导师的带教水准参差不齐；更为严重的是，随着关键核心人员离职，组织内的宝贵智慧和隐性知识极易断档。在 Agent 原生企业中，这一逻辑被完全逆转：新员工的入职培训直接由深度理解公司运转逻辑的 Agent 来接管。任何公司内部都充斥着大量独有的术语、黑话（jargons）和隐性规则，正是这些壁垒阻碍了新人快速上手。现在的模式是，掌握全公司知识底座的 Agent 会主动检索内部上下文，并为每位新人量身定制专属的动态学习教程。以往老员工需要花几天手写 Wiki、准备课件，如今系统能直接根据新人的背景与当前课题即时生成针对性指引。基于这一转变，Kytu 提出了衡量 AI 价值的关键指标革新：放弃那些浮于表面的 Token 消耗量统计，甚至不要在早期死扣滞后的财务财务公式，企业应当紧盯的核心牵引指标是“价值交付时间”（Time to Value, TTV）。像 Google、Facebook 这样的硅谷大厂，通常需要为新工程师安排长达两个月的集中训练营（Boot Camp）才能让他们交付有效产出；而在 Agent 的赋能下，新人交付实质业务价值的周期被压缩到了“以天计算”。只要价值交付周期被剧烈压缩，长期的商业回报就是必然结果。在底层架构支撑上，Agent 原生体系需要构建三大基石模块： 1. **Agent 编排层（Orchestration Layer）**：负责复杂的任务规划与多 Agent 协同。 2. **基于本体（Ontology）的语义数据层（Semantic Data Layer）**：统一存储公司的所有结构化与非结构化数据，准确刻画公司业务的真实运作机制。 3. **内部应用快速部署系统**：让非技术人员在创建内部业务 App 后能够一键交付投产。 --- ### [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 组织文化重塑：消除职能孤岛，从传统价值链转向“人人皆为 Builder” 转向组织文化层面，Kytu 展示了如何通过重写价值链来打破传统的岗位分工。以最常见的网页开发（如落地页或基础展示页）为例，过去标准流程必须串联市场营销、产品经理、UI 设计师和前端工程师等整条链路；但在今天，一名市场人员完全可以借助专用的 Web Agent 独立端到端完成整个页面的搭建与上线，不仅响应时间极速缩短，沟通成本也近乎归零。这种价值链重塑同样发生在其商业分析、财务预测等各个部门。对于互联网公司而言，传统的产研价值链往往最为昂贵：业务单元（BU）负责人、运营、商业分析师、PM、设计师、研发经理和程序员层层嵌套，为了各司其职，企业不得不高薪雇佣大量垂直细分的专职人员。但在强大的 Agent 工具支撑下，这套僵化体系正在瓦解。只要底层基础设施足够稳固，哪怕是市场或销售人员，也能直接“交付功能产品”。 Kytu 坚信，未来的组织将不再需要那么多高度细分的螺丝钉岗位，取而代之的是全面的“Builder（构建者）文化”。在这一范式下，每个人都必须同时理解市场痛点、设计逻辑与产品交付，因为检验产品好坏的唯一标准只有真实的市场反馈。这也对企业的人才策略提出了更高要求：当前的 Agent 技术能力几乎是无限的，但员工脑中的固有思维和提示词（Prompting）能力却直接决定了上限。因此，“Agent 的开发固然重要，但对人的培养和学习闭环的建设要重要得多”。 - **研发链条扁平化**：打破 BU 经理、PM、设计与研发的漫长串行链条，让业务一线直接面向市场交付。 - **内部落地战果**：借助自身构建的 Agent 基础设施，全公司各职能部门在短短三到四个月内自发构建了超过 1,040 个业务应用。 - **产品化输出**：团队将内部验证成熟的底座提炼为名为 Viho 的 SaaS 产品，直接向外界开放体验。 --- ### [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) 核心总结：重构工作流、追踪落地时效与投资员工成长在演讲的结尾，Kytu Lin 对整场分享的核心心法进行了精炼总结，为志在完成 Agent 原生化改造的团队梳理了关键抓手：首先，在业务模式上，要想真正看到 AI 带来的高额投资回报，唯一的出路就是打破原有的“流程死锁”，坚决进行端到端的工作流重构，而不是在旧齿轮上涂抹润滑油。其次，在技术与度量层面上，切忌将“模型调用频次”或“Token 消耗量”当作成功的表征；团队必须坚定不移地以“价值交付时间”（Time to Value）作为核心牵引指标，只要从需求萌芽到创造实际价值的时间被大幅压缩，企业便已经走在正确的复利轨道上。最后，在团队与文化维度上，企业领导者必须将重心从单纯的“技术与 Agent 构建”向后退一步，投向“员工自身的认知与技能升级”。只有投资于人的成长、建立持续迭代的学习闭环，组织才能真正跨越鸿沟，在 Agent 时代蜕变为具备极强生命力的敏捷共同体。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) 根据 MIT 的研究报告，高达 95% 的企业在 AI 上的投资根本没有带来预期的投资回报率（ROI）。 — Kytu Lin > [00:04](https://www.youtube.com/watch?v=PIopKTVGwoU&t=4s) 在 Dcard 内部，他们已经向生产环境交付了超过 100 个自主 Agent，全公司有 90% 的员工将 AI 深度融入日常工作流。 — Kytu Lin > [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) 传统组织中绝大部分的时间和精力并不是消耗在‘具体任务的执行’上，而是无休止地浪费在跨部门交接（handoffs）、拉会对齐、反复澄清需求以及等待各级领导审批的停滞时间里。 — Kytu Lin > [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) 真正的价值源于对工作流的重构，而非单纯嵌入 AI。 — Kytu Lin > [01:35](https://www.youtube.com/watch?v=PIopKTVGwoU&t=95s) 企业迈向 Agent 原生时遇到的阻碍从来不只是技术选型问题，本质上是一场深层的组织文化挑战。 — Kytu Lin > [02:43](https://www.youtube.com/watch?v=PIopKTVGwoU&t=163s) 放弃那些浮于表面的 Token 消耗量统计，甚至不要在早期死扣滞后的财务公式，企业应当紧盯的核心牵引指标是‘价值交付时间’（Time to Value, TTV）。 — Kytu Lin > [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 未来的组织将不再需要那么多高度细分的螺丝钉岗位，取而代之的是全面的‘Builder（构建者）文化’。 — Kytu Lin > [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) Agent 的开发固然重要，但对人的培养和学习闭环的建设要重要得多。 — Kytu Lin > [05:44](https://www.youtube.com/watch?v=PIopKTVGwoU&t=344s) 借助自身构建的 Agent 基础设施，全公司各职能部门在短短三到四个月内自发构建了超过 1,040 个业务应用。 — Kytu Lin > [08:34](https://www.youtube.com/watch?v=PIopKTVGwoU&t=514s) 只有投资于人的成长、建立持续迭代的学习闭环，组织才能真正跨越鸿沟，在 Agent 时代蜕变为具备极强生命力的敏捷共同体。 — Kytu Lin ## Mentioned Kytu Lin, Dcard, GNTC, MIT, Google, Facebook, Viho, Time to Value (TTV) ## Video description Kytu Lin - Founder & CEO, GNTC Everyone now has PhD-level AI — so why do most companies still work exactly the same way? I'm Kytu Lin, founder and CEO of Dcard, a 20-million-user social platform, and I think the reason is what I call the Agent Gap: the people who understand the problems best are rarely the ones allowed to build the fixes. To close it I became my own company's forward-deployed engineer — the result was 600+ internal agents built by HR, legal, finance, and sales, not just engineers. In this talk I'll walk through the three hardest problems we had to solve: making scattered company context usable by agents, governing what agents are allowed to do, and turning first wins into an adoption flywheel. If you're trying to move AI from impressive pilots to real, company-wide adoption, you'll leave with the playbook — and the story of how these lessons became the enterprise agent platform we now build for other companies. [Source](https://www.youtube.com/watch?v=PIopKTVGwoU)

Source quotes

[0:04] Kytu Lin: 根据 MIT 的研究报告，高达 95% 的企业在 AI 上的投资根本没有带来预期的投资回报率（ROI）。 [0:04] Kytu Lin: 在 Dcard 内部，他们已经向生产环境交付了超过 100 个自主 Agent，全公司有 90% 的员工将 AI 深度融入日常工作流。 [1:35] Kytu Lin: 传统组织中绝大部分的时间和精力并不是消耗在‘具体任务的执行’上，而是无休止地浪费在跨部门交接（handoffs）、拉会对齐、反复澄清需求以及等待各级领导审批的停滞时间里。 [1:35] Kytu Lin: 真正的价值源于对工作流的重构，而非单纯嵌入 AI。 [1:35] Kytu Lin: 企业迈向 Agent 原生时遇到的阻碍从来不只是技术选型问题，本质上是一场深层的组织文化挑战。 [2:43] Kytu Lin: 放弃那些浮于表面的 Token 消耗量统计，甚至不要在早期死扣滞后的财务公式，企业应当紧盯的核心牵引指标是‘价值交付时间’（Time to Value, TTV）。 [5:44] Kytu Lin: 未来的组织将不再需要那么多高度细分的螺丝钉岗位，取而代之的是全面的‘Builder（构建者）文化’。 [5:44] Kytu Lin: Agent 的开发固然重要，但对人的培养和学习闭环的建设要重要得多。 [5:44] Kytu Lin: 借助自身构建的 Agent 基础设施，全公司各职能部门在短短三到四个月内自发构建了超过 1,040 个业务应用。 [8:34] Kytu Lin: 只有投资于人的成长、建立持续迭代的学习闭环，组织才能真正跨越鸿沟，在 Agent 时代蜕变为具备极强生命力的敏捷共同体。

Source quotes

Kytu Lin: Everyone now has PhD-level AI — so why do most companies still work exactly the same way? Kytu Lin: the people who understand the problems best are rarely the ones allowed to build the fixes. [1:35] Kytu Lin: The trap is not only technical, but it's also cultural. Real value comes from redesigning workflow, not just adding AI. [5:44] Kytu Lin: The agent possibility is infinity right now. But the mindset and the brain will limit how you prompt the agents. Human development is much more important than agent development. [8:34] Kytu Lin: Redesign Workflows: Eliminating cross-department handoffs and restructuring end-to-end value chains is essential to breaking the ROI deadlock. [8:34] Kytu Lin: Optimize for Time-to-Value: Evaluate AI tooling by how rapidly it accelerates business delivery rather than tracking token usage or theoretical cost savings. [8:34] Kytu Lin: Prioritize Human Development: Provide continuous training and learning loops to help employees master agent prompting and product design, ensuring the workforce's ambition keeps pace with model capabilities. [0:04] Kytu Lin: 95% of enterprise AI investments fail to generate positive ROI.

Original transcript

[0:04] Uh hi everyone. My name is Keta Lin and I'm founder and CEO of two company. One [0:09] is called GMTC and another is T-car. And we have like 20 million active users. [0:14] So today I going to share about our experience how we transfer this company into agent native. [0:21] Okay, so as we knows, uh according to MIT report, 95% of the enterprise their [0:27] AI investments doesn't have ROI. And so yeah, actually this is a huge problem, right? [0:34] And based on our own experience, I think this problem is solvable. Uh we saw that uh we didn't [0:39] just test AI agents, we built a whole company uh AI workflow based on this. And in the past years, [0:46] we shipped over 100 agents in productions and we have 90% of employees adopt all the AI into their [0:53] workflows. So I going to show you how we did it. And and based on this, because of this experience, [0:59] we launched another company called GMTC to help more people uh to help more enterprises become [1:05] agent native. And this is our approach. We deliver the platform and we provide services at the same [1:10] time and our main focus is about the banks, telecoms, and manufacturers across the Asia. [1:17] So today I going to share about few things. I will describe about the agent gap what we found [1:23] and then the technology and the culture behind to shift the company into agent native. And then [1:28] the last I will do the brief recap. Okay, so let's talk about the gap. [1:35] Actually, based on our observation, why many company fail and what works is this chart. [1:41] If you use the AI into your existing workflow, but you didn't redesign your workflow, [1:46] actually you didn't see dramatically increase in the ROI because the major effort of the [1:53] how we get work done is about the handoff. You talk to another about the meetings, [1:57] about the job, how how to get done, and you have waited for someone else's approval. [2:02] This is the major cause of uh time waste. But, what really works is that you redesign [2:06] we decide the workflow and shorten the workflow, and that will see the increase uh increase [2:12] the ROI dramatically. So, real value comes from the redesigning workflow, not just adding AI. [2:18] But, here comes another problem that is not only a gap, but also the trap. Because if you don't have [2:24] a successful, sorry, you don't believe and you won't redesign. And again, you doesn't see more [2:30] successful story. So, you become a dead lock. So, a a brief uh a brief probably can start a trial, [2:36] but it cannot break the loop. So, the trap is not only technical, but it's also cultural. [2:43] So, let me describe about what a agentic company will look like in tech perspectives. [2:49] So, I will start from the case study. So, we have over 100 uh agents on production. But, [2:55] I will start from the one who every company will encounter is the the onboarding new hires. [3:01] I guess everyone onboard a new hires probably in the last one year one year, right? And onboarding [3:07] new hires is is a has high frictions. First, it will drop the productivity because the senior [3:13] have to spend time on uh new hires. And actually, the coaching is not consistent. [3:18] And someone coach better, someone doesn't. And the continuity is the key people leave, [3:23] they bring in the knowledge wisdom. So, this is the one of the biggest topic in the company. [3:29] For a uh agentic company, they will think another way. How can we make sure that [3:35] each people onboarding that through agents, not through people. So, in the in the center, [3:40] the agents you will build upon the agents, but they they will know your company very well, [3:45] and each people will ask the agents to understand your people. Uh to understand your company. So, [3:51] it will look like this. Every company have less company-specific terms, jargons, only inside of [3:57] those. That's what block the new hires to learn about the company. But, when you ask agent [4:04] agent native platform, you ask them because they understand company so well, they will look into [4:08] the company and directly create a tutorial for these new hires. So, this is totally different, [4:14] right? In the In the past, even your senior, they have to prepare some tutorial to teach the new [4:20] hires. But, right now, they can directly directly ask those system and create a tutorial for them. [4:28] And we found this the most important metrics is about time to values. Because in the past, [4:34] like like big company like Facebook, Google, they probably have to spend 2 months on a boot camp to [4:39] to train those new new hires to develop to deliver values. But, right now, based on our observation, [4:44] new hires can shift value probably in days, not months. So, this is what we we will [4:52] our suggestions. In the agent eras, leading matches is not about the finance matches you [4:57] calculate on the ROI. You have to measure the time to value. If the time to value is decrease a lot, [5:04] shorten a lot, it means it must bring in long-term ROI. And it's is a bit much better [5:11] measurement than token consumption. So, this is basically our our tech stack. I I I won't go deep, [5:18] but I will brief introduce. So, you must Each company must have a agent orchestration layer [5:24] and also semantic data layer that store all the company data you base on ontology and then and how [5:30] you build and describe the company's how it works. And probably you will have an internal deploy [5:35] our system. And then when people deploy the create the app for internal use, they can deploy easily. [5:44] Okay, so the next stop I want to to about the culture side. So, this is another way to draw the [5:53] redesign workflow. So, if you rewrite the value chain of each company, like web development, [5:59] some like landing site, some basic sites, you used to have marketing PM, designer, engineer [6:04] to complete the job. But right now, a huge proportion of the jobs can be done directly [6:11] by the marketer with the web agent. So, you can kill through later financial improvement and [6:17] a response time very easy. It applies to business inside apply to finance analyst. And actually, [6:25] in each company, you can draw this kind of a value chain probably 10 10,000 over 100. Yeah. [6:33] And let me give you a deeper example. In many company, like for those we are internet company, [6:39] this is the most expensive value chains for BU managers, operator, business inside, [6:45] PM, designer, engineer manager, and engineers. And you will have different vertical [6:51] different type. So, we have to hire a bunch of very vertical people to complete the job. [6:56] But right now, because you have lots of agent tools, actually everyone can become builders. [7:02] Even the marketer or sales can ship features if you have very robust infrastructure. [7:08] So, I I believe in the futures, we don't we won't have so many different jobs, but we will embrace [7:14] the builder culture. It means that everyone they have to understand market, design, and product. [7:20] Because the only way to to tell a product engineer a product is good is based on market feedback. [7:29] And I think this is very important that not only thinking about the agent building, [7:33] but also thinking about how you train your people. Because the agent possibility is [7:38] infinity right now. But the mindset and the brain will limit you how you prompt the agents. [7:45] So it's very important for each company to create a learning loop to train each [7:49] employees to how to use it. Yeah, so I would say for the culture side actually agent development is [7:55] important but the human development is much more important for the company to become agent native. [8:01] So like for our example of our SDK example we can call in the car engineers, PM designers, [8:08] and HR, legal, finance, PR domain Electra. So and so to do that we build the infrastructure. [8:16] Uh it Right now probably only three or four months over a 1,040 apps was built [8:22] and we launched this as a SaaS so it's easy you can try it online. It's called Viho. [8:27] And it tremendously increased our productivity across the company. Okay, so that's it. [8:34] Uh so simply a recap. So the problem actually to see the ROI you have to redesign the workflow and [8:42] you have to break the loop. Otherwise your company won't change. And the solution for technical side [8:47] every technical tech technology introduced you have to measure the time to value. If [8:52] you increase you are on the right path not the token consumption. And then for the culture wise [8:57] you have to make sure you invest more on human development over agent development. So that's [9:02] it. Thank you. If you want to the style connect talk to more you can scan the QR code. Thank you.