PromptQL Pushes Codified Context Layers to Fix Enterprise AI Failures

Your AI Is Only as Smart as Your Team With PromptQL — Ai4

Ai4

Enterprise AI initiatives collapse because foundation models reset after every session, making human-curated, portable business context the true proprietary asset rather than swappable underlying LLMs.

AI-assisted summary of Your AI Is Only as Smart as Your Team With PromptQL — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 20, 2026, this presentation features Suku Krishnaraj, COO and President of PromptQL. According to the description, the session addresses how knowledge work depends on scattered, rapidly changing business context that AI inherently lacks, outlining a framework to turn team expertise into a permanent, computable asset.

## TL;DR Suku Krishnaraj argues that enterprise AI initiatives stall not due to underlying model capabilities, but because of unshared context, isolated single-user chat sessions, and an inability to scale across shifting enterprise rules. Using a live go-to-market war room demo, he illustrates how cross-functional teams—spanning marketing, sales, and engineering—can collaboratively teach an AI operational nuances that are saved directly into a centralized wiki. He advocates for establishing a portable 'company brain' as proprietary intellectual property, allowing organizations to swap frontier models freely while compounding institutional human expertise across departments.

## Atmosphere & style The presentation has the crisp, pragmatic vibe of an enterprise conference talk, trading speculative tech hype for grounded operational realism. Krishnaraj speaks with steady executive clarity, illustrating familiar corporate pain points like multi-day support backlogs and chaotic incident war rooms before pivoting into a collaborative live demonstration. The tone feels structured, forward-thinking, and reassuring, framing knowledge workers not as casualties of automation, but as essential domain experts actively shaping their tools.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) Enterprise knowledge bottlenecks often stem from operational friction and fragmented systems rather than question complexity, forcing analysts to repeatedly start inquiries from scratch. - [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) Enterprise AI initiatives fail not because of foundation model limitations, but due to three core structural gaps: lack of shared context, isolated single-user collaboration, and an inability to scale complex rules. - [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) Current AI chat interfaces create knowledge silos where context taught by an analyst evaporates the moment a session closes, preventing peers from inheriting those learnings. - [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) AI models break down at scale when they cannot dynamically adapt to intricate geographic and policy nuances, such as localized salary bands or varying state regulations. - [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) Cross-functional collaboration in a shared environment allows diverse teams—such as sales, marketing, and engineering—to simultaneously train AI systems on live operational context. - [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) Human judgment remains crucial in enterprise AI workflows, allowing team members to place explicit manual holds on strategic accounts rather than relying on full automation. - [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) Advanced AI systems must identify their own boundary conditions by surfacing what they know, what they lack, and what specific business rules they need humans to teach them. - [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) Persisting collaborative inputs directly into a shared enterprise wiki transforms fleeting human conversations into permanent, computable organizational knowledge. - [09:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=544s) A codified 'company brain' or semantic layer enables a 'teach it once, everyone gets the skill' paradigm that drives 10x operational velocity over minor efficiency gains. - [09:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=544s) Because frontier models evolve continuously, enterprises must decouple the underlying swappable LLM from their proprietary, portable context layer and knowledge graphs. - [12:24](https://www.youtube.com/watch?v=JjukIuNqpSM&t=744s) Without systems that compound organizational knowledge over time, teams across industries repeatedly rehash identical investigative steps during operational incidents. - [12:24](https://www.youtube.com/watch?v=JjukIuNqpSM&t=744s) Sustainable enterprise AI strategies focus on elevating human expertise by embedding evolving institutional context directly into existing daily communication channels.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) The Business Context Problem: A Case Study in Enterprise Support

A major global human resources and payroll company faced a staggering operational bottleneck: managing over 160,000 customer support requests each year with a staff of 200 to 250 support analysts. Despite the size of the team, each individual support interaction dragged out to an average duration of 40 to 50 minutes. The root cause was not the complexity of the questions themselves, but the friction required to answer even standard inquiries.

For example, when a customer asked a straightforward question like, "What is my time off balance?", an analyst had to navigate through six disparate software systems managed by six separate internal teams. The analyst then had to manually compile and curate a custom report, which often took four to five days to finally deliver back to the customer. This breakdown occurs because every single customer interaction effectively "starts from scratch." Without a centralized mechanism to capture, persist, and distribute critical business context across the organization, institutional knowledge remains locked in silos, forcing repetitive manual work for every incoming request.

- **Support Volume:** Over 160,000 support requests handled annually by 200–250 support analysts. - **Handling Time:** Averaging 40–50 minutes per call, with curated reporting turnarounds taking 4–5 business days. - **System Fragmentation:** Answering simple inquiries required manual data consolidation across 6 separate systems and 6 different internal teams.

---

[01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) Three Fundamental Reasons AI Initiatives Fail

AI initiatives across enterprises frequently fail to deliver meaningful business value. The breakdown rarely stems from the underlying foundation models themselves—as the models are generally capable—but rather from three structural gaps in how AI systems are deployed in knowledge work:

```mermaid flowchart TD accTitle: Three Core Failure Modes of Enterprise AI subgraph Failures["Core AI Failure Modes"] A["1. Lack of Shared Context<br/>(AI resets to zero each session; human brain knowledge lost)"] B["2. Isolated Collaboration<br/>(Single-user silo; peer analysts cannot inherit learnings)"] C["3. Failure to Scale<br/>(Breaks under complex, shifting enterprise policies)"] end Failures --> D["Operational Stagnation & Inaccurate Results"] ```

1. **Lack of Shared Context:** Every new interaction with an AI tool starts from ground zero. Current systems fail to capture what is stored inside the "human brain" and leave valuable institutional knowledge unrecorded and undistributed. Without shared context, AI cannot reliably deliver accurate, domain-aware results. 2. **Absence of Team Collaboration:** When a skilled analyst spends time teaching a chatbot the specific context it needs during an active session, that knowledge exists solely within that temporary thread. The moment the chat window closes, the acquired understanding evaporates. The analyst sitting in the next chair learns nothing from the exchange, and there is no collaborative mechanism to invite team members in to enrich and expand that shared context. 3. **Inability to Scale:** A model might handle a single narrow query—such as retrieving a time-off balance—but collapses when tasked with broader enterprise operations. Scaling requires managing intricate regional nuances, such as mapping recruitment salary bands across specific counties, districts, and states, or adapting to shifting state policies like California's minimum wage versus federal standards.

---

[03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) Live GTM War Room: Collaborative Context in Action

To demonstrate how cross-functional collaboration enriches AI, an operational go-to-market event war room was assembled with five key functions: the Head of Marketing (Zach), the Head of Sales (Mike), a Business Development Representative (BDR), a Forward Deployed Engineer (Adam), and Go-to-Market leadership (Suku Krishnaraj). The goal was to take live event data—such as conference booth badge scans and hallway conversations—and build an actionable operational execution plan in real time.

```mermaid flowchart TD accTitle: Collaborative Context Capture Pipeline subgraph Roles["Cross-Functional Team Inputs"] MKT["Marketing (Zach)<br/>Defines hot/cold scans & dashboards"] SLS["Sales (Mike)<br/>Flags manual hold for Moody's"] FDE["Eng (Adam)<br/>Routes technical pilots for Cisco"] BDR["BDR<br/>Updates lead routing playbook"] GTM["GTM Lead (Suku)<br/>Reserves VIP dinner accounts"] end subgraph AI_Engine["PromptQL Collaborative Engine"] AI["AI Ingests Real-Time Context<br/>Identifies knowledge gaps & rules"] end subgraph Output["Codified Knowledge"] WIKI["Shared Enterprise Wiki / Company Brain"] end

Roles --> AI AI --> WIKI ```

The workflow progressed transparently in a shared environment where all team members could observe the AI's evolving logic:

- **Marketing Integration:** Zach taught the AI how to categorize raw badge scans into "hot" versus "cold" classifications, how those scans convert into qualified leads, and how to prioritize them before generating a unified campaign dashboard visible to all participants. - **Sales Intelligence:** Mike reviewed captured accounts and layered in private conversational context from recent text and Slack exchanges. He explicitly instructed the AI regarding Moody's: "When it comes to Moody's, I need to be in the loop. I need to be able to actually respond back. I don't want any of this automated." - **Technical Deployment Nuance:** Adam intervened regarding key enterprise accounts such as Cisco, noting that booth visitors had raised specific technical requirements and had already been sent specialized pilot information packets, preventing redundant or misaligned outreach. - **Playbook Generation:** The BDR inputted updated routing rules from the previous evening and asked the AI to generate a revised outreach playbook synthesizing the inputs from marketing, sales, and engineering. - **Executive Carve-Outs:** Suku flagged three strategic accounts slated for private executive dinners, directing the AI to isolate them from standard automation playbooks and generate tailored briefing packets for direct executive follow-up.

When prompted on what it needed to execute future events even more effectively, the AI surfaced its own boundary conditions: "Here are the things that I know. Here are the things that I don't. I need you to teach me the routing rules. I need you to teach me how to prioritize the hot accounts." With the click of a button, this newly synthesized context was saved directly into an enterprise wiki, converting fleeting human conversations into computable organizational knowledge.

---

[09:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=544s) Building the "Company Brain" as Core Intellectual Property

Whether industry professionals refer to it as a "knowledge layer," "context," or a "semantic layer," the core objective remains the same: codifying the dynamic operational knowledge of an enterprise. Rather than continuously re-prompting models from scratch, organizations need a framework where subject matter experts "teach it once, everyone gets the skill."

Operational parameters change constantly in high-growth environments: a CFO may alter revenue recognition rules, a sales organization might redefine market segments (e.g., no longer categorizing accounts under 1,000 employees as "commercial"), or executive leadership might introduce higher gross margin targets. A dedicated "company brain" captures these evolving business rules dynamically, enabling organizations to achieve 10x operational velocity rather than settling for incremental 10% efficiency gains.

```mermaid flowchart LR accTitle: Modular Architecture with Shared Context IP subgraph Swappable["Swappable Frontier Models"] LLM1["Model A (Week 1)"] LLM2["Model B (Week 2)"] end subgraph ProprietaryIP["Enterprise Proprietary Asset"] Brain["Company Brain / Knowledge Graphs<br/>(Shared Context & Wiki)"] end Swappable <--> Brain ```

A critical architectural consideration is vendor lock-in. Because frontier foundation models evolve and change weekly, enterprises must maintain a clean separation between the underlying model and their proprietary knowledge. The true intellectual property of the enterprise is the portable shared context, the knowledge graphs, and the operational wiki—not the underlying vendor's LLM.

- **Reusability:** Institutional context is codified once by domain experts and distributed across the entire team automatically. - **Dynamic Adaptability:** Real-time updates reflect strategic shifts, such as modified revenue definitions, new margin goals, or updated ICP thresholds (e.g., accounts with $\le$ 1,000 employees). - **Model Portability:** Frontier models should remain modular and swappable while preserving the underlying proprietary context layer as enterprise IP.

---

[12:24](https://www.youtube.com/watch?v=JjukIuNqpSM&t=744s) Enterprise Compounding and Human-Centric AI Strategy

The necessity of a shared context layer spans diverse industries, including retail, e-commerce, human resources, software, and financial services. In financial services, for example, firms processing millions of dollars in daily trade settlements frequently assemble ad-hoc emergency war rooms of 100 people when settlement failures occur. These teams often spend hours rehashing the exact same historical failure modes and investigative steps because they lack the connective tissue to compound their learnings across incidents.

Sustainable enterprise AI strategy hinges on elevating, rather than replacing, the human knowledge base. Successful deployments focus on building an evolving, collaborative company brain integrated directly into everyday workflows—whether natively within Slack, Microsoft Teams, or dedicated platforms. Turning human institutional expertise into an interactive, computable context layer enables organizations to scale beyond isolated use cases and drive compounding operational performance.

## Notable quotes > AI has gotten remarkably good at software. But most knowledge work runs on something AI still doesn't have: business context. — Suku Krishnaraj

> [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) What is my time off balance?

> [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) When it comes to Moody's, I need to be in the loop. I need to be able to actually respond back. I don't want any of this automated. — Mike

> [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) Here are the things that I know. Here are the things that I don't. I need you to teach me the routing rules. I need you to teach me how to prioritize the hot accounts.

> [09:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=544s) teach it once, everyone gets the skill. — Suku Krishnaraj

> The organizations winning with AI analytics aren't just deploying smarter tools. They're building new operating models where teams continuously teach AI their business context, so every analysis gets sharper over time and human judgment stays in the loop. — Suku Krishnaraj

> [09:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=544s) The true intellectual property of the enterprise is the portable shared context, the knowledge graphs, and the operational wiki—not the underlying vendor's LLM. — Suku Krishnaraj

> Knowledge roles aren't going away. They're being redesigned. — Suku Krishnaraj

## Mentioned Suku Krishnaraj, PromptQL, Ai4, Zach, Mike, Adam, Moody's, Cisco, Slack, Microsoft Teams

## Video description Suku Krishnaraj - COO and President, PromptQL

AI has gotten remarkably good at software. But most knowledge work runs on something AI still doesn't have: business context. It lives in people's heads, it's scattered across teams, and it changes constantly. So every time you hand a question off to an AI tool, something gets lost. A definition. An assumption. The thing you actually meant.

The organizations winning with AI analytics aren't just deploying smarter tools. They're building new operating models where teams continuously teach AI their business context, so every analysis gets sharper over time and human judgment stays in the loop.

In this session, Suku Krishnaraj, President and COO of PromptQL, makes the case for why AI analytics is a team sport and what it looks like when your team's expertise becomes a permanent, computable asset. You'll walk away with a practical framework for turning shared context into a real competitive advantage and making decisions that are not just fast, but ones you can actually stand behind.

Knowledge roles aren't going away. They're being redesigned.

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

中文简报

PromptQL推动代码化上下文层以化解企业AI失败困局 企业AI项目之所以停滞，是因为基础模型在每次会话后都会重置，这使得经人工整理且可移植的业务上下文成为真正的专有资产，而非可随意替换的底层LLM。 Suku Krishnaraj认为，企业AI项目停滞不前并非因为底层模型能力不足，而是由于缺乏共享上下文、孤立的单用户聊天会话，以及无法适应不断变化的企业规则进行扩展。通过一场实时的入市（GTM）作战室演示，他展示了涵盖市场营销、销售和工程的跨职能团队如何协同向AI传授运营细节，并将其直接保存到中心化wiki中。他主张建立一个可移植的“公司大脑”作为专有知识产权，使企业能够自由更换前沿模型，同时在各部门之间持续沉淀机构的人类专业经验。 企业的知识瓶颈通常源于运营摩擦和零散的系统，而非问题本身的复杂性，这迫使分析师不得不屡次从头开始发起查询。 (0:04) 企业AI项目的失败并非由于基础模型的局限，而是源于三大核心结构性缺陷：缺乏共享上下文、孤立的单用户协作，以及无法扩展复杂规则。 (1:33) 当前的AI聊天界面造成了知识孤岛，分析师传授的上下文在会话结束时便随之消失，导致其他同事无法继承这些经验。 (1:33) 当AI模型无法动态适应复杂的地域和政策细微差异（例如本地化薪酬范围或各州不同的法规）时，其在大规模应用中便会崩溃。 (1:33) 在共享环境中开展跨职能协作，使销售、营销和工程等不同团队能够结合实时运营上下文同步训练AI系统。 (3:33)

Reading summary

Enterprise AI initiatives frequently underperform because institutional knowledge remains trapped in human silos and ephemeral chat threads rather than being captured in a shared operational context. Decoupling proprietary business rules from rapidly changing foundation models allows organizations to codify expertise once and compound operational intelligence across teams. Structural Bottlenecks in Disconnected Enterprise AI (0:04) Enterprise operations often stall because workflows start from scratch with each task. In a cited human resources and payroll case study, 200 to 250 support analysts handling over 160,000 annual customer requests faced average handling times of 40 to 50 minutes, with reporting taking four to five days due to manual data consolidation across six fragmented software systems and internal teams. This breakdown reflects three core structural failure modes in enterprise AI adoption: models reset to zero without access to unrecorded institutional knowledge, single-user chat sessions trap acquired context instead of sharing it across peer analysts, and systems collapse when tasked with scaling across nuanced operational rules, such as local recruitment salary bands or shifting state minimum wage policies. Real-Time Cross-Functional Context Capture (3:33) To demonstrate collaborative context capture, a live go-to-market war room brought together marketing, sales, engineering, business development, and executive leadership to process event badge scans and hallway discussions. Marketing lead Zach taught the AI how to categorize scans into hot or cold leads, while sales head Mike layered in private conversational context to halt automated messaging for Moody's. Simultaneously, forward deployed engineer Adam flagged prior pilot distributions for Cisco, and GTM lead Suku Krishnaraj isolated strategic accounts for executive dinners. Rather than assuming complete autonomy, the AI identified its own boundary conditions by asking the team to specify lead routing and account prioritization criteria. The synthesized logic was then committed directly to an enterprise wiki, converting fleeting operational conversations into codified organizational knowledge visible to all participants. Codifying the Company Brain as Enterprise IP (9:04) The presentation argues that an enterprise's true intellectual property is not the external foundation model, but rather its proprietary context layer, semantic layer, or 'company brain.' Because frontier large language models evolve weekly, organizations must maintain an architectural separation between modular, swappable models and internal knowledge graphs. This context layer captures evolving business rules that models cannot deduce independently, such as changes in revenue recognition, updated gross margin targets, or revised market segments (such as adjusting commercial criteria to exclude accounts with 1,000 or fewer employees). By codifying rules once through domain experts, organizations distribute new capabilities instantly across all users to drive significant operational velocity. Compounding Operational Knowledge Across Industries (12:24) The failure to compound institutional context recurs across diverse sectors, including retail, software, and financial services. In trade settlement operations, for example, institutions processing millions of dollars in daily transactions frequently convene 100-person emergency war rooms that spend hours re-investigating identical historical settlement failures because earlier troubleshooting steps were never preserved. Sustainable enterprise deployment requires human-centric strategies that embed interactive context layers directly into everyday collaboration environments such as Slack or Microsoft Teams. Translating expert judgment into computable, structured knowledge allows organizations to move beyond isolated, one-off AI interactions to achieve compounding productivity gains. The presentation asserts that enterprise operational velocity depends on codifying live human collaboration into a portable context layer rather than repeatedly re-prompting raw frontier models. However, the session focuses primarily on conceptual workflows and GTM event scenarios, leaving unresolved the technical governance and automated conflict-resolution mechanisms needed when cross-functional business rules directly contradict one another.

精炼阅读摘要

Ai4 的专题内容指出，企业级 AI 项目未达预期通常并非底层基础模型能力不足，而是因为缺乏共享业务上下文、缺乏团队协同以及无法适配复杂的业务规模。演讲者认为，企业的核心资产应当是独立于底层模型的“企业大脑”上下文层，通过领域专家单次沉淀让全员继承能力，从而推动业务效率的复合增长。 系统割裂与上下文缺失导致的支持瓶颈 (0:04) 节目列举了一家大型全球人力资源与薪酬公司的运营案例。该公司拥有 200 至 250 名支持分析师，每年需处理超过 160,000 起客户请求。尽管团队规模可观，单次支持交互的平均耗时仍长达 40 至 50 分钟。效率低下的症结不在于咨询问题的复杂性，而在于获取基础信息所伴随的巨大摩擦。例如，当客户询问“假期余额”这一常规问题时，分析师必须在 6 个不同内部团队维护的 6 套独立软件系统间手动查询整合，交付定制报告往往耗时 4 至 5 个工作日。 这种流程阻滞的根源在于客户交互缺乏传承机制，几乎每次服务都需“从零开始”。由于组织内部缺乏集中捕捉、存储和分发业务上下文的工具，机构知识被深锁在信息孤岛中，迫使支持团队在面对重复请求时不断进行低效的手工劳动。 企业级 AI 部署失效的三大结构性缺陷 (1:33) 演讲者分析认为，当前企业 AI 应用未能产生实际商业回报，主要源于知识工作流程中的三大结构性缺陷：第一是缺乏共享上下文，AI 工具在每个新会话中都会重置为初始状态，无法持续捕捉和沉淀人类脑海中的隐性经验；第二是单点会话缺乏团队协作，分析师在独立聊天窗口中对机器人进行的微调与纠偏，会随着窗口关闭而彻底蒸发，邻近同事无法继承或复用其训练成果；第三是无法支撑复杂业务规模的扩展，模型处理孤立简单查询尚可，但在面对复杂多变的跨区域政策（如加利福尼亚州最低工资标准与联邦法规的差异）或细分到县市维度的招聘薪酬带规则时，极易因逻辑冲突而失效。 跨职能作战室中的协同上下文捕获 (3:33) 为验证跨部门协作对 AI 效能的提升，团队展示了一个走向市场（GTM）活动作战室的实操案例。该作战室汇集了市场主管 Zach、销售主管 Mike、业务拓展代表（BDR）、前线部署工程师 Adam 以及 GTM 负责人 Suku Krishnaraj，旨在实时将展会徽章扫描与走廊对话转化为可落地的执行方案。在此流程中，市场部门定义了有效线索的冷热判别规则；销售主管 Mike 根据即时通讯获得的私有背景，明确指示 AI 对 Moody's 账户暂停自动化分流并由人工介入；工程师 Adam 补充了 Cisco 的专业技术试点诉求；BDR 与 GTM 负责人则分别录入了最新的线索分配手册并隔离出高管晚宴专属账户。 在协同输入后，AI 能够清晰反馈其已知规则与知识盲区（例如主动要求输入分配规则与热点账户排序逻辑）。经由团队确认，整套业务逻辑一键沉淀至企业 Wiki 中，完成了从零散口头对话到可计算组织知识的转化。 构建与模型解耦的“企业大脑”核心资产 (9:04) 演讲者指出，无论是被称为知识层、上下文层还是语义层，核心诉求均是将动态变化的运营规则结构化沉淀。在快速发展的企业环境中，业务参数频繁变更：首席财务官可能调整收入确认规则、销售部门可能重新界定规模在 1,000 人以下的企业不属于“商业客户”，管理层亦会提出更高的毛利率目标。通过构建动态的“企业大脑”，领域专家只需“教学一次”，全团队即可共享该能力，帮助企业实现 10 倍速的运营跨越，而非局限于 10% 的微小边际改善。 此外，由于前沿基础模型演进迭代迅速，企业在架构设计上必须保持谨慎以防范供应商锁定。演讲者主张将底层大模型与企业专有知识进行明确解耦：模型作为可随时替换的即插即用组件，而沉淀在知识图谱和可移植上下文层中的专属业务经验，才是企业真正的核心知识产权。 演讲者总结指出，构建共享上下文层不仅适用于营销与人力支持，也是金融、零售和软件等行业的普遍需求；例如日均处理数百万美元交易的金融结算部门，常因缺乏知识沉淀而陷入百人应急作战室反复排查同类故障的困境。可持续的企业 AI 实施策略在于赋能而非取代人类专家，通过将机构经验深度嵌入日常协作工作流，实现组织智能的长期复利。

Detailed notes — zh

## Background 本内容由 Ai4 于 2026 年 8 月 20 日发布在 YouTube 平台上，演讲者为 PromptQL 总裁兼首席运营官 Suku Krishnaraj。页面描述指出，AI 在软件层面进展显著，但缺乏企业赖以运转的业务上下文，本场演讲探讨了如何通过新型协作运营模型将团队专业经验沉淀为可计算的持久企业资产。 ## TL;DR 演讲首先剖析了传统大型企业因缺乏共享上下文而在数据支持上陷入的低效泥潭，指出 AI 项目落地失败的核心在于上下文缺失、缺乏协作机制与难以规模化。随后，演讲者通过五人 GTM 展会战情室的现场实战，演示了市场、销售、工程师等跨角色如何实时向 AI 注入动态隐性知识，并一键沉淀为企业 Wiki。最后，演讲呼吁企业构建独立于底层大模型的“可移植知识层”与企业大脑，将组织智慧转化为核心 IP，以实现 10 倍速的效能跃升。 ## Atmosphere & style 整场演讲呈现出务实、干练且极具说服力的企业级技术布道风格。演讲者通过真实的业务痛点案例迅速建立共鸣，随后以紧凑流畅的现场多角色实战演练带给观众极强的代入感与确定性。整体节奏条理清晰、层层递进，既有对当前 AI 浅层应用局限性的敏锐剖析，又洋溢着以人为核心、重塑知识工作形态的前瞻信心。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) 企业 AI 项目的痛点不在于基础大模型的能力缺陷，而在于业务上下文的缺失与知识孤岛问题。 - [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) 缺乏上下文连续性会导致团队陷入每次任务都“从零开始”的低效泥潭，极大地拉长响应周期与协作成本。 - [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) AI 落地失败主要源于三大系统性瓶颈：共享上下文缺失、缺乏团队协作机制，以及无法应对复杂的规模化业务边界。 - [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) 单人私有会话训练出的 AI 无法沉淀为组织资产，会话关闭后知识即刻流失，难以在团队内部共享复用。 - [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) AI 跨部门协同模式允许市场、销售、工程师等多角色向统一 AI 注入分散的业务信息，消除跨部门信息不对称。 - [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) 团队协作中的即时对话与业务上下文能够一键归档为企业共享 Wiki，将临时交互沉淀为持久的组织资产。 - [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 人类专家才是发挥 AI 价值的关键，企业最核心的决策信息往往散落在日常聊天与隐性经验中，需要被主动捕获。 - [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 企业大脑需要实现“教一次，全员掌握”，动态跟进收入确认准则、客户分段等不断演变的业务运营规则。 - [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 企业不应满足于带来 10% 边际改善的浅层工具，而应依托可计算的知识图谱实现 10 倍速的效能跨越。 - [11:40](https://www.youtube.com/watch?v=JjukIuNqpSM&t=700s) 企业的真正核心知识产权是自主掌控的知识层与业务图谱，而非外部频繁更迭的底层大模型算法。 - [11:40](https://www.youtube.com/watch?v=JjukIuNqpSM&t=700s) 技术架构必须保持高度的可移植性，使企业能够热插拔底层前沿大模型，同时牢牢锁定自身业务数据与规则资产。 - [11:40](https://www.youtube.com/watch?v=JjukIuNqpSM&t=700s) 引入具备连接性的知识层是打破金融结算、客户支持等跨职能复杂业务反复故障推演的关键解法。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) 痛点案例：全球人力资源巨头的知识孤岛与支持危机 演讲开场分享了一家全球顶尖的人力资源与薪酬管理（HR & Payroll）公司的真实遭遇。这家巨头每年要处理超过 16 万起客户支持工单，配备了 200 到 250 名专业支持分析师，然而平均每通支持电话的耗时高达 40 到 50 分钟。问题的荒谬之处在于，即便是一个看似最简单的问题——例如客户询问“我的带薪休假（Time off）余额是多少？”——支持人员都无法立刻给出答案。 为了回答这一个简单的问题，分析师必须跨越由 6 个不同团队各自掌管的 6 套孤立业务系统，手动抓取数据并拼凑成一份定制报告，往往需要耗费 4 到 5 天时间才能最终交付给客户。造成这种低效的根本原因，在于交互过程中严重缺失了“业务上下文”（Business Context）。每一次支持通话都像是一场从零开始的重启，那些深植于员工脑海中的业务认知和操作经验根本无法被沉淀、捕获并在全公司范围内复用。 - **工单规模与成本**：每年 160,000+ 起支持请求，由 200-250 名分析师支持，单次平均响应时长 40-50 分钟。 - **流程断层**：查询基础休假余额需要穿透 6 个系统、协调 6 个团队，客户等待周期长达 4-5 天。 - **核心症结**：每次支持任务缺乏上下文连续性，每次处理都沦为“从零开始”（Restart from scratch）。 --- ### [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) AI 项目失败的三大根源：上下文、协作与规模化 当前许多企业的 AI 落地项目频频折戟，但问题其实并不在底层基础大模型本身——“模型本身没有问题”，真正的瓶颈在于三个维度的系统性缺失：共享上下文、团队协作与规模化扩展能力。 ```mermaid flowchart TD accTitle: AI落地失败的三大核心原因 subgraph Bottlenecks["AI 落地三大阻碍"] B1["1. 共享上下文缺失<br/>(每次对话从零开始)"] B2["2. 缺乏团队协作<br/>(单点会话关闭即丢失)"] B3["3. 无法规模化延伸<br/>(复杂业务与政策边界)"] end Bottlenecks --> Consequence["AI 仅能应对单点浅层任务<br/>无法沉淀为企业级生产力"] ``` 首先是**共享上下文（Shared Context）的缺失**。每次人类员工与 AI 对话，AI 的认知基准都是零。系统无法自动捕获那些埋藏在专家大脑中的隐性经验与组织知识，导致 AI 始终无法给出真正精准、符合业务预期的结果。其次是**缺乏团队协作（Team Collaboration）机制**。一位极其出色的分析师可能在自己的聊天窗口里把 AI 训练得非常聪明，但一旦会话关闭，这些上下文立刻烟消云散；坐在她旁边的同事一无所获，团队成员之间无法共同参与、邀请协作并持续丰富这套知识库。最后是**规模化（Scale）难题**。让 AI 查询简单的带薪休假余额或许容易，但当业务延伸到全美各州、县、学区交织的薪资等级（Salary Bands），或是加利福尼亚州与联邦最低工资标准的政策差异时，AI 缺乏多层级复杂业务逻辑的支撑，根本无法准确拓展到更广泛的企业场景中。 --- ### [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) 实战演示：五人 GTM 战情室如何协同训练 AI 为了展示如何通过协作解决上述难题，PromptQL 团队进行了一场现场实战模拟：由主讲人牵头组织其 GTM（Go-To-Market）核心团队建立一个展会“战情室”（War Room），成员包括市场营销负责人 Zach、销售负责人 Mike、业务开发代表（BDR）以及现场前瞻部署工程师（Forward Deployed Engineer）Adam。团队的目标是全面复盘展会展台的扫描数据与现场对话，制定即时作战与跟进计划。 ```mermaid flowchart LR accTitle: GTM 战情室多角色协同上下文注入流程 subgraph TeamInput["多角色业务上下文实时注入"] MKT["市场负责人 (Zach)<br/>定义冷热线索与仪表盘"] SALES["销售负责人 (Mike)<br/>锁定重点客户 (如 Moody's)"] FDE["部署工程师 (Adam)<br/>标记技术对接 (如 Cisco)"] BDR["BDR 专员<br/>同步路由规则与最新 Playbook"] LEAD["GTM 负责人<br/>高管晚宴名单独立切分"] end subgraph PromptQL["PromptQL 企业大脑协作层"] SharedAI["统一认知与动态学习"] WikiBtn["一键归档至企业共享 Wiki"] end MKT --> SharedAI SALES --> SharedAI FDE --> SharedAI BDR --> SharedAI LEAD --> SharedAI SharedAI --> WikiBtn ``` 整个协作过程中，各部门实时向同一个 AI 输入分散的业务上下文，AI 则在多轮交互中持续学习演进： 1. **市场端注入标准**：Zach 率先教导 AI 什么是有效扫描、如何界定“高意向扫描”（Hot scan）与“低意向扫描”（Cold scan）、线索如何转化以及优先级规则，AI 随后生成全员可见的活动看板，消除了跨部门沟通的信息不对称。 2. **销售端接入动态**：Mike 结合现场与客户的 Slack 消息和短信互动，向 AI 明确指示：涉及穆迪（Moody's）等重点客户时，必须由销售人员介入直连，严禁任何自动化营销流程。 3. **技术端规避冲突**：Adam 提醒 AI，思科（Cisco）等老客户在展台提出了特殊技术诉求，他已私下安排了试点方案包，因此 AI 不得向其推送常规物料。 4. **BDR 统一战术**：面对前一晚刚调整的线索路由规则，BDR 结合市场、销售和技术专家已注入的信息，要求 AI 实时生成最新的跟进工作手册（Playbook）。 5. **管理层全局统筹**：主讲人要求 AI 将他在展会晚宴上亲自接待的 3 家核心大客户剥离出通用流程，生成专属的高管对接交接包。 在任务完成时，主讲人向 AI 提问：“为了打好下一场展会，我们还需要补充什么？”AI 主动梳理出知识盲区并给出提示。主讲人只需轻点蓝色按钮，所有交互中沉淀的业务上下文便瞬间转化为结构化的企业 Wiki，成为全员共享的持久资产。 --- ### [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 打造“企业大脑”：将人类智慧沉淀为可复用的核心资产 展会战情室的案例揭示了一个核心现实：企业中海量的关键决策信息其实散落在短信、Slack 讨论以及走廊茶水间的闲聊中，AI 无法凭空获取这些隐性知识，"人类才是让 AI 真正发挥价值的关键"。许多 AI 项目之所以失败，正是因为它们脱离了企业赖以运转的组织智慧。 行业里人们将其称为“知识层”（Knowledge Layer）、“上下文”（Context）或“语义层”（Semantic Layer），但本质上它就是企业的**运营知识（Operational Knowledge）**。企业需要一套动态机制来形式化、代码化这些知识：“教一次，全员掌握，无需重复传授”。 ```mermaid flowchart TD accTitle: 企业大脑知识持续演进模型 subgraph DynamicInputs["日常动态业务变更"] D1["CFO 调整收入确认准则"] D2["重新定义客户分段 (如商业客户门槛)"] D3["毛利率与经营目标更新"] end DynamicInputs --> Brain["企业大脑 (Company Brain) / 知识图谱"] Brain --> Impact["赋能日常决策与全员 AI 交互<br/>实现 10 倍速跨越 (10x not 10%)"] ``` 在日常运营中，业务规则是高度动态变化的——CFO 可能在今天修改收入确认的口径，上周公司可能刚将商业客户（Commercial）的定义从“1000 人以下”进行调整，或者管理层提出了全新的利润率指标。如果这些变动没有被持续捕捉，AI 就会输出脱节的错误分析。企业需要构建一个始终伴随业务运转的“企业大脑”（Company Brain），将人脑中的知识沉淀为可计算的知识图谱，从而驱动企业实现 10 倍（10x）而不是 10% 的微小效能跃升。 - **核心价值主张**：“教它一次，全员获益”（Teach it once, everyone gets the skill）。 - **动态知识范例**：收入确认口径变更、目标客群规模重新定义、边际利润目标调整。 - **目标设定**：告别仅带来 10% 边际改善的单点工具，通过共享上下文构建带来 10 倍级质变的企业大脑。 --- ### [11:39](https://www.youtube.com/watch?v=JjukIuNqpSM&t=699s) 跨行业落地与可移植的 IP 架构：追求 10 倍效能飞跃 企业的真正核心知识产权（IP）不是外部频繁迭代的算法模型，而是团队共同沉淀的**共享上下文与业务知识层（Wiki / 知识图谱）**。前沿大模型（Frontier Models）几乎每周都在推陈出新，因此企业在选择技术架构与供应商时，必须确保知识层具备高度的“可移植性”（Portable），能够无缝热插拔底层大模型，同时牢牢将业务知识资产掌控在自己手中。PromptQL 自身也是深度践行“吃自家狗粮”（Dogfooding），其内部运营完全构建在自身的知识图谱与上下文系统之上。 这种架构已经在零售、电商、人力资源、SaaS 软件及金融服务等多个行业广泛落地： - **金融结算故障排查案例**：某大型金融服务机构每天处理数百万美元的交易结算，一旦结算失败，往往需要临时召集上百名业务专家组成应急战情室。尽管团队成员极度聪明且早已使用多种 AI 工具，但由于缺乏将业务上下文串联起来的“连接组织”（Connective Tissue），团队不得不在每次故障中反复推导相同的基础问题。 ```mermaid flowchart LR accTitle: 可移植知识层架构 subgraph CoreIP["企业专属核心 IP (自身掌控)"] Context["动态共享上下文<br/>(Shared Context & Wiki)"] KG["企业运营知识图谱<br/>(Knowledge Graph)"] end subgraph SwappableModels["可自由替换的底层模型"] M1["Frontier Model A"] M2["Frontier Model B"] M3["未来新模型..."] end Context <--> KG KG --> SwappableModels ``` 演讲最后呼吁企业管理者：绝不要低估团队成员脑海中的知识库价值。应当选择支持多渠道协作（无论是在 Slack、Microsoft Teams 还是专属工作台）的平台，让业务专家与 AI 持续互动协作，将零散经验快速编码为具备生命力、不断进化的企业大脑，在风云变幻的 AI 浪潮中构建真正坚固的竞争壁垒。 ## Notable quotes > [01:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=93s) 模型本身没有问题 — Suku Krishnaraj > [00:04](https://www.youtube.com/watch?v=JjukIuNqpSM&t=4s) 从零开始 — Suku Krishnaraj > [03:33](https://www.youtube.com/watch?v=JjukIuNqpSM&t=213s) 为了打好下一场展会，我们还需要补充什么？ — Suku Krishnaraj > [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 人类才是让 AI 真正发挥价值的关键 — Suku Krishnaraj > [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 教一次，全员掌握，无需重复传授 — Suku Krishnaraj > [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) Teach it once, everyone gets the skill — Suku Krishnaraj > [08:19](https://www.youtube.com/watch?v=JjukIuNqpSM&t=499s) 10x not 10% — Suku Krishnaraj > [11:39](https://www.youtube.com/watch?v=JjukIuNqpSM&t=699s) 吃自家狗粮 — Suku Krishnaraj ## Mentioned Suku Krishnaraj, PromptQL, Zach, Mike, Adam, Moody's, Cisco, Slack, Microsoft Teams, Ai4 ## Video description Suku Krishnaraj - COO and President, PromptQL AI has gotten remarkably good at software. But most knowledge work runs on something AI still doesn't have: business context. It lives in people's heads, it's scattered across teams, and it changes constantly. So every time you hand a question off to an AI tool, something gets lost. A definition. An assumption. The thing you actually meant. The organizations winning with AI analytics aren't just deploying smarter tools. They're building new operating models where teams continuously teach AI their business context, so every analysis gets sharper over time and human judgment stays in the loop. In this session, Suku Krishnaraj, President and COO of PromptQL, makes the case for why AI analytics is a team sport and what it looks like when your team's expertise becomes a permanent, computable asset. You'll walk away with a practical framework for turning shared context into a real competitive advantage and making decisions that are not just fast, but ones you can actually stand behind. Knowledge roles aren't going away. They're being redesigned. [Source](https://www.youtube.com/watch?v=JjukIuNqpSM)

Source quotes

[1:33] Suku Krishnaraj: 模型本身没有问题 [0:04] Suku Krishnaraj: 从零开始 [3:33] Suku Krishnaraj: 为了打好下一场展会，我们还需要补充什么？ [8:19] Suku Krishnaraj: 人类才是让 AI 真正发挥价值的关键 [8:19] Suku Krishnaraj: 教一次，全员掌握，无需重复传授 [8:19] Suku Krishnaraj: Teach it once, everyone gets the skill [8:19] Suku Krishnaraj: 10x not 10% [11:39] Suku Krishnaraj: 吃自家狗粮

Source quotes

Suku Krishnaraj: AI has gotten remarkably good at software. But most knowledge work runs on something AI still doesn't have: business context. [0:04] What is my time off balance? [3:33] Mike: When it comes to Moody's, I need to be in the loop. I need to be able to actually respond back. I don't want any of this automated. [3:33] Here are the things that I know. Here are the things that I don't. I need you to teach me the routing rules. I need you to teach me how to prioritize the hot accounts. [9:04] Suku Krishnaraj: teach it once, everyone gets the skill. Suku Krishnaraj: The organizations winning with AI analytics aren't just deploying smarter tools. They're building new operating models where teams continuously teach AI their business context, so every analysis gets sharper over time and human judgment stays in the loop. [9:04] Suku Krishnaraj: The true intellectual property of the enterprise is the portable shared context, the knowledge graphs, and the operational wiki—not the underlying vendor's LLM. Suku Krishnaraj: Knowledge roles aren't going away. They're being redesigned.

Original transcript

[0:04] Let me begin with a story. One of the  largest global HR and payroll company [0:14] came to us and they have over 160,000 support  requests a year supported by 200 250 support [0:28] analysts. Each of these support calls takes  about an average 40 to 50 minutes. Case in point, [0:40] it's as simple as a question like, "What is my  time off balance?" Takes these support people to [0:48] go through six different systems owned by six  different teams to build a curated report and [0:56] deliver the report back to the customer.  sometimes four, sometimes 5 days later. [1:03] What is going on? The business context is the  knowledge that's missing in these interactions. [1:14] Every support call is a restart. It start  from the scratch. the business context and [1:21] the knowledge that's there is not there is no way  to capture and use it across your organization. [1:33] Let's look at deeply why the AI initiatives are  failing. There are three fundamental reasons. [1:42] It's not the model. Models are fine. What we've  learned first and foremost is the shared context. [1:50] Every time you engage with an AI, it starts  from zero. The system is not capturing what [1:59] you have in human brain, human head. All of the  institutional knowledge that's buried in people [2:06] is not getting captured and distributed.  So that's one of the fundamental reasons [2:13] why shared context is critical to get  AI to deliver accurate results for you. [2:20] Second is team collaboration. If you look at a  brilliant analyst that teaches the AI what AI [2:30] needs to know in a given session and in a given  chatbot, the AI learns it, but then you close [2:40] the chat session, it's gone. The analyst sitting  next to her hasn't learned anything. There is no [2:47] way for this analyst to teach the team, invite the  team over and collaborate with them to enrich the [2:54] shared context. That's the second reason we  saw why AI initiatives are failing. Third is [3:02] scale. It's not sufficient for your AI to work  for time off balance. What about recruitment? [3:10] What about salary bands across a particular  county which belongs to a district which belongs [3:16] to a state? What if certain policies were changed?  What about the minimum wage for the country versus [3:24] California? Where are all these information  coming from? How do you scale accurately? [3:33] What we did is we did a live example very  simple example just to make the point I run [3:43] go to market and I brought in the team let's  first talk about what who is in this room and [3:51] who is in this team my head of marketing my  head of sales my BDR and my forward deployed [3:59] engineer they're all we invited all of them  into my room and a simple example of can we [4:06] create a war room to work this event. Can  we actually look at who actually stopped by [4:13] the booth? What interactions and conversations  have we had? And can we build an attack plan, [4:18] an operational plan that allows us to go execute? [4:25] Let me walk you through some examples. A  headup marketing came in. He taught the [4:30] AI what is a scan, what's a hot scan, what's a  cold scan, how did these scans turn to leads, [4:40] what should be prioritized and what should  not be. And then he asked the AI to build [4:44] a campaign dashboard that everyone in  the room remember all five functions [4:51] are in the room including myself as Zach is  working with AI. Everyone else can also see [5:01] there is no what is this and why is it this  way. None of that work is just happening. [5:09] Mike, who runs sales, looks at these, looks  at the accounts that have been captured, [5:16] and he's just had some Slack and text  conversation with certain accounts that [5:23] actually stopped by the booth. He teaches  the AI, hey, when it comes to Moody's, [5:30] I need to be in the loop. I need to be able  to actually respond back. I don't want any [5:35] of this automated because A, B, and C. If you  think about it, AI is always learning. It was [5:41] learning in the previous thread. It's learning  now. It's constantly absorbing what's going [5:49] on. Then my forward deployed engineer comes  in and says, listen, some of our customers, [5:56] Cisco and others are actually stopped by. I don't  want you to send them anything because the people [6:03] that stopped by had A, B, and C problems  from a technical requirement. I've already [6:07] arranged for to send them a pilot information  packet. Adam teaches AI. So AI has learning. [6:24] BDR comes in. Last night there was a change  in the routing of leads and there were some [6:32] automation rules that were set up that he needed  to come in and teach the AI and he's asking based [6:40] on all the information you've learned from  marketing sales and deployment engineer give [6:47] me the new playbook. The AI comes in and gives a  new playbook based on everything it's learning. [6:56] Then I come in and say, "Listen, there are  three accounts that are very particularly [7:01] that I've I have dinner with at this event  and I don't want them going through the [7:07] playbook. I need you to carve them out for  a specific handoff, executive handoff." [7:17] And it gives me a packet that allows me to  go touch on these accounts specifically. [7:25] And then I also ask AI, okay, you learned all  of this. You created the campaign dashboard. [7:30] You created a new routing playbook. You you  gave direction to sales. Um you now we have [7:37] a new playbook for VDR. What is it that we  should be doing better to run the next event [7:44] in a in a place where we actually have learned  all of this stay start with the next next level. [7:51] The AI comes in and says, "Here are the things  that I know. Here are the things that I don't. I [7:57] need you to teach me the routing rules. I need you  to teach me how to prioritize the hot accounts, [8:03] etc." The AI is surfacing all of this. And  all I have to do is click that blue button. [8:12] We capture shared context in the form  of wiki. It goes straight into the wiki. [8:19] So the reason for this example is a very  simple example of war room for an event. But [8:26] look at the number of nuances that is in people's  heads. Text conversations, Slack conversations, [8:32] hallway conversations. Not all of this is getting  captured by AI automatically. Humans are the ones [8:40] that are making AI useful. This is what we've  learned by being in this business. And I'll [8:48] show you some of the customer examples. So a lot  of the AIS fail because they're failing to touch [8:55] on the human institutional knowledge which is  super critical for AI to become productive. Zoom [9:04] back out for a second. As you think about your own  company in this ballroom, I'm sure there's a lot [9:12] of institutional knowledge floating around. And  of all the conversations I've had at this event, [9:20] everyone comes in, they call it knowledge layer,  they call it context, they call it semantic layer, [9:26] doesn't matter. It's it's really about the  operational knowledge of your company. How [9:32] do you codify that? And how do you instruct AI to  use that information to incrementally get better? [9:42] teach it once, everyone gets the skill. There's  no need to retach it. How do you move forward? [9:51] By improving the base. The the base is improved  by constantly teaching. The way you want to move [9:57] fast is by codifying the human knowledge into  AI. Use the power of your team collaboration. [10:05] Find those subject matter experts. Have them  teach your AI what they have in their head. [10:12] So you don't have to teach every single  time you you teach it once. CFO changed [10:19] the definition of revenue recognition. Last  week we changed our segment definition. We no [10:26] longer look at thousand employees  and below accounts as commercial. [10:33] We have margin goals that are different today  than yesterday. Who is capturing all of this? [10:41] So that is why it's critical to have a system  that allows you to capture this in an ongoing [10:49] basis in a dynamic fashion. We call this the  company brain. Can you have a company brain [10:58] that is always with you that is always guiding  you with every interaction that allowing you to [11:02] move 10x not 10% better? Anybody can do that.  The global HR company was using AI tools. They [11:11] are. They were, but it wasn't helping. Is  10% improvement any good? It's not. You I [11:21] urge you to think about 10x and your IP  intellectual property is really the shared [11:30] context and the knowledge layer, the wicki that  you built. That's your intellectual property. [11:39] Find a vendor that allows you to swap out the  frontier models which is changing every week. [11:46] So you want to be able to swap them out. You  want to be able to retain the shared context. [11:53] That's your IP. It's not vendor's IP. Make sure  you look for a vendor that has a portable shared [12:00] context and a knowledge layer. And what you see  there is that our own company brain we use only [12:08] our our own dog food and it's the context  that we've built. If you look at all the [12:14] knowledge graphs that's there, it's all about  the knowledge of our operational knowledge. [12:24] Now as I transition and show you  some of the customers that have [12:28] interested us that have come come  for come looking for how to build [12:32] the company brain you can see accounts  across retail e-commerce human resources [12:41] software across the board they all have  different challenges financial services [12:49] one of the financial services company  runs millions of dollars in settlements [12:55] and settlements fail and they have an army.  They have a ad hoc war room of 100 people [13:01] trying to figure out what went wrong.  They're rehashing the same topics over [13:08] and over. They're using AI. They're very  smart people. But what's missing is that [13:14] connective tissue that allows you to compound.  I'm not talking about 1x. I'm talking about 10x. [13:27] Just to leave you with a thought. Do not  underestimate the human knowledge base. [13:37] Find a way to codify, find a way to capture that  and scale that across the organization. A lot of [13:46] AIs fail because they work for one use case,  not across the board. Think of a vendor that [13:52] allows you to build a company brain. That's  living, breathing, constantly evolving think. [14:00] And as AI landscape continues to change,  the context is what you have as your IP [14:09] and provide a way. Find a vendor that  allows your teams to collaborate. [14:17] But be it on Slack, be it on Teams, be it  on the vendor's own platform. Get the teams [14:24] working with each other constantly. That's the  fastest way to capture and scale the context. [14:34] We can show you live. Come by, stop by our booths [14:39] and we can go into depth. Would love  to hear from all of you. Thank you.