## Background
Published on YouTube by the Ai4 channel on August 21, 2026, this presentation features Jacob Schofield, Enterprise AI GTM lead at Google Cloud. According to the page description, the session explores common roadblocks stalling enterprise AI initiatives and introduces a practical framework for connecting AI capabilities directly to bottom-line business value.
## TL;DR
Enterprise AI is moving past superficial chatbots into a mature 'second inning' that demands comprehensive organizational redesign rather than mere software seat licensing. Drawing on historical parallels like the productivity paradox of the Industrial Revolution, Schofield presents Google Cloud's AI Fluency Framework to help leaders unify siloed data, cultivate curious generalists, and orchestrate autonomous multi-agent workflows. As financial scrutiny shifts attention away from unbridled 'token maxing' toward durable P&L impact, companies must balance bold transformational bets with iterative, muscle-building use cases. Ultimately, successfully scaling AI requires managing agent sprawl, modernizing legacy systems, and grounding executive expectations in realistic operational timelines.
## Atmosphere & style
The session carries a grounded, candid, and reassuring tone as Schofield delivers an end-of-conference keynote anchored in front-line enterprise reality rather than academic hype. Using evocative analogies—from steam engines and assembly lines to gym bodybuilding and stepping into the ring with Mike Tyson—he demystifies executive anxieties with pragmatic clarity. The delivery feels like an empathetic insider briefing, balancing patient organizational empathy with an urgent call for disciplined execution.
## Key takeaways
- [02:19](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=139s) Enterprise AI is transitioning from surface-level automations like chatbots into a 'second inning' focused on deep workflow redesign and bottom-line business value.
- [03:53](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=233s) The productivity paradox illustrates that AI value lags initial adoption because organizational processes must be structurally redesigned, mirroring the historical shift from centralized steam engines to distributed assembly lines.
- [07:18](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=438s) Google Cloud's AI Fluency Framework outlines five critical layers for transformation: unified data foundations, AI literacy, evolving employee profiles, scalable agentic platforms, and progressing from automation to fluid human-AI collaboration.
- [10:24](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=624s) Agentic workflows augment go-to-market teams by synthesizing SEC filings and executive media into custom value hypotheses, eliminating generic discovery questions in executive meetings.
- [12:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=768s) With natural language acting as a universal coding layer, enterprise hiring is shifting away from rigid technical specialists toward curious, adaptable generalists.
- [13:58](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=838s) Pursuing isolated point-solution widgets creates tool sprawl and security liabilities, necessitating a platform-driven approach to cross-functional agent orchestration.
- [18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) High failure rates in early AI proofs of concept (POCs) are essential for building the organizational, cultural, and technical muscle needed for complex deployments.
- [20:57](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1257s) The rapid proliferation of vendor-specific agents requires IT leadership to implement centralized governance frameworks to prevent fragmented agent sprawl.
- [21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) Enterprises are moving past an unconstrained 'token maxing' mindset toward strict CFO scrutiny, demanding direct P&L alignment and defensible ROI on all AI compute spend.
- [23:59](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1439s) Practicing on smaller, single-function and multi-agent workflows builds the baseline competencies required to safely execute high-impact flagship transformations.
- [25:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1548s) Technology leaders must temper board expectations of immediate '10x magic' by communicating realistic infrastructure timelines for integrating legacy enterprise data.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=4s) Introduction: Realities of Enterprise AI from the Front Lines
Opening the final session on the last day of the Ai4 conference, Jacob Schofield commends the audience's stamina before introducing his role leading Google Cloud's enterprise AI go-to-market team across the Western United States and the semiconductor vertical. Rather than styling himself as an academic theorist or a visionary scientist among the brilliant minds at the event, Schofield anchors his perspective in day-to-day enterprise realities. Working directly with corporate leadership teams affords him an unvarnished look at how companies actually grapple with AI across their workforce, their product roadmaps, their competitive positioning, and the acute friction points created by the rapid velocity of modern technology. Even inside Alphabet, teams face the exact same disruptive shifts and organizational growing pains that external enterprises experience.
Schofield emphasizes that enterprise leaders must navigate an overwhelming amount of public noise and extreme polarities. The media oscillates wildly between utopian promises of automated abundance and universal basic income, and dystopian scenarios where robots displace humanity. In truth, anyone claiming to predict where enterprise AI will stand in six, twelve, or twenty-four months is disingenuous. Because the landscape is so fundamentally volatile and confusing, Schofield encourages executives to grant themselves and their teams a degree of grace while working through the transition. The broader directive is to cultivate patience and internal breathing room, yet execute with deliberate urgency in the face of inevitable industry disruption.
[02:19](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=139s) Entering the Second Inning: Beyond Chatbots and Surface Automation
Enterprise AI is moving past its introductory phase and entering what Schofield describes as the "second inning of the game." In this phase, tactical time-savings metrics—such as deploying an internal chatbot to summarize a document, draft an email, or generate slides for a presentation—are no longer considered sufficient victories. Boards, CEOs, and executive committees are exerting intense downward pressure on Chief Information Officers, Chief Technology Officers, and Chief Product Officers to discover the catalytic "spark" that will reinvent the business, instill a "10x mentality," eliminate decade-old manual processes, and unlock explosive revenue growth.
Basic automation alone cannot deliver that magnitude of corporate transformation. While drafting a document faster or having a virtual agent answer simple employee questions is convenient, it merely shaves off minutes rather than repositioning an organization within its competitive market. Leaders are increasingly impatient with superficial automation narratives; they need clear architectural pathways that translate generative AI capabilities into measurable operational leverage and bottom-line expansion.
[03:53](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=233s) The Productivity Paradox: From Steam Engines to the Modern Assembly Line
To contextualize why immediate corporate windfalls remain elusive, Schofield turns to economic history and the "productivity paradox"—a concept popularized in late-1990s economic literature illustrating that the true economic value of technological revolutions routinely lags decades behind the initial breakthrough. Following the late-2023 and early-2024 boom, popular sentiment assumed a linear relationship between AI adoption and enterprise productivity: simply inserting AI into existing workflows would yield instant transformation. In reality, adoption curves show a brief initial bump from surface-level automation followed by an extended plateau, because organizational workflows must be entirely restructured before the technology can achieve its full potential.
```mermaid
flowchart TD
accTitle: The Industrial Energy Evolution vs AI Adoption
subgraph Centralized["Phase 1: Direct Substitution"]
direction TB
A["Central Steam Engine<br/>(1800s Factories)"] -->|Replaced By| B["Central Electric Motor<br/>(Same Pulleys & Shafts)"]
E["Buying AI Tools<br/>(Copilot / Gemini Add-ons)"] -->|Superimposed On| F["Legacy Workflows<br/>(Incremental Muscle)"]
end
subgraph Decentralized["Phase 2: Architectural Redesign"]
direction TB
B -.->|30-Year Structural Lag| C["Decentralized Power<br/>at Workstations"]
C --> D["Modern Assembly Line<br/>Massive Manufacturing Scale"]
F -.->|Workflow Restructuring| G["Agentic Services & Data Fabric"]
G --> H["Fluid Hybrid Intelligence<br/>Transformational Enterprise Value"]
end
```
Schofield draws an extended analogy to the transition from steam power to electricity during the Industrial Revolution. In the 1800s, factories operated around a single centralized steam engine that powered an intricate, shop-wide network of spinning shafts, gears, and pulleys. When electric motors were first introduced, factory owners merely swapped out the central steam engine for an electric generator, leaving the legacy mechanical distribution system untouched. It took roughly thirty years for manufacturers to decentralize electric power directly to individual workstations along the floor, a structural redesign that finally catalyzed the modern assembly line and the second Industrial Revolution.
Purchasing off-the-shelf software tools such as Google Workspace with Gemini, Microsoft Copilot, or Cloud Code builds baseline organizational muscle, but Schofield warns that buying seat licenses is the modern equivalent of swapping the steam engine while keeping the old pulleys in place. It does not reinvent operations. Leaders frequently wrestle with arbitrary board mandates—such as demanding a 50% operational efficiency gain or a 10x output multiplier—while confronting severe real-world headwinds:
- Deeply fragmented data spread across hundreds of disparate SaaS platforms, public clouds, and legacy on-premises data centers.
- Internal cultural polarization, where aggressively AI-forward employees run up against deeply AI-averse colleagues resistant to changing their day-to-day habits.
- The absence of a systematic blueprint to connect raw foundational models to end-to-end operational workflows.
[07:18](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=438s) The AI Fluency Framework: Data, Culture, and the Shift to Hybrid Intelligence
To help enterprises maneuver through these hurdles, Google Cloud established the internal "AI Fluency Framework." Schofield points out that Google faces these market pressures itself—such as pervasive public commentary questioning whether emerging conversational AI tools will cannibalize Alphabet’s core Search business. The framework outlines five foundational layers necessary to move an organization from simple tool adoption to structural transformation:
```mermaid
flowchart LR
accTitle: The AI Fluency Framework Continuum
subgraph Foundations["Foundational Layers"]
direction TB
D["1. Data Foundation<br/>(Unify SaaS, Cloud & On-Prem)"]
C["2. Culture & Literacy<br/>(Instill AI-First Mindset)"]
D --> C
end
subgraph Execution["Organizational Capabilities"]
direction TB
P["3. Evolving Employee Profile<br/>(Skills-Based to Generalist)"]
T["4. Scalable Tooling & Platforms<br/>(Agentic Infra for Tech Teams)"]
O["5. Organizational Design<br/>(Automate -> Augment -> Transform)"]
P --> O
T --> O
end
Foundations ==> Execution
```
- **Data Foundation**: An AI model or agent is fundamentally constrained by the quality and accessibility of the data it can query. Transforming operations requires breaking through data silos across AWS, Microsoft Azure, on-premises environments, and hundreds of SaaS tools.
- **Culture and Literacy**: Handing employees a software license does not teach them how to rethink their jobs. Organizations must actively train personnel to build confidence and literacy in AI-native problem-solving.
- **Shifting the Employee Profile**: Traditional corporate hiring focused strictly on siloed technical skill sets, such as hiring a dedicated SQL programmer, a financial analyst, or an accountant. AI shifts value toward generalist capabilities. Natural language now functions as a universal programming layer: an employee without technical coding experience can query databases directly in plain English because the model handles the code generation.
- **Scalable Technology Infrastructure**: Scalability requires more than end-user desktop copilots. Technical teams need enterprise-grade development platforms to deploy, govern, and monitor autonomous agentic services that interact safely with backend corporate applications.
- **Organizational Design Continuum**: Google structures workflow modernization across three distinct operational phases:
1. *Automate*: Eliminating repetitive, time-intensive operational bottlenecks.
2. *Augment*: Constructing collaborative workflows where automated systems run in parallel with human oversight.
3. *Transform*: Achieving a fluid, hybrid operating environment where humans and autonomous AI systems collaborate seamlessly with real-time accuracy, speed, security, and strict enterprise governance.
[10:24](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=624s) Augmenting Go-to-Market Teams with Agentic Workflows
Reflecting on his personal mission leading a go-to-market organization, Schofield notes that every sales leader shares the same ultimate wish: for their entire team to sell ten times their quota. Achieving that level of scale requires moving far beyond basic initial automations, such as using AI to draft email responses, answer RFPs, or pull Salesforce and BI data into account plans. While those automations served as a solid first step, the real leap forward happens during the augmentation phase through agentic workflows. Instead of manually parsing accounts, sales teams can assign tasks to autonomous agents that evaluate strategic accounts, synthesize 10-K filings, ingest every public video of a target company's CEO from YouTube into an indexed data store, and extract deep leadership insights to build a custom, consulting-style value hypothesis.
```mermaid
flowchart TD
accTitle: Go-to-Market AI Evolution
subgraph S1["1. Automation Phase"]
A["Basic Tools"] --> B["Automated RFPs & Emails"]
A --> C["Salesforce & BI Account Plans"]
end
subgraph S2["2. Augmentation Phase"]
D["Agentic Workflows"] --> E["Analyze 10-Ks & Filings"]
D --> F["Ingest CEO Media & Talks"]
E & F --> G["Consulting-Style Value Hypothesis"]
end
subgraph S3["3. Transformation Phase"]
G --> H["Executive-Tailored Conversations (CFO, CIO, CTO)"]
end
```
This agentic capability elevates customer engagement into a genuinely transformational mission. Sales reps no longer need to walk into an executive meeting and ask basic discovery questions like "what keeps you up at night," because that information is already publicly available across earnings reports, SEC filings, and internet interviews. With autonomous agents handling the contextualization of massive datasets, account teams arrive at client meetings with intimate knowledge of the company's financials, board priorities, and executive vision, delivering targeted perspectives tailored specifically to a CFO, CIO, CTO, or VP of Product.
- **Initial Automation:** Routine tasks such as email responses, RFP answers, and CRM/BI account plan population.
- **Agentic Augmentation:** Ingesting 10-Ks, earnings reports, and executive multimedia to generate structured value hypotheses.
- **Strategic Impact:** Eliminating generic discovery questions by arriving at executive meetings with pre-synthesized insights tailored to specific C-suite personas.
---
[12:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=768s) The Evolving Employee Profile: From Rigid Specialists to Curious Generalists
The widespread integration of agents is fundamentally altering organizational hiring and the standard profile of successful employees. Previously, enterprises sought out rigid, hyper-specialized talent—such as SQL developers dedicated solely to querying databases, data analysts building reports, finance specialists executing repetitive models, or enterprise salespeople relying exclusively on traditional relationship-based playbooks. With AI agents directly mapped to enterprise data sources, an individual no longer needs to write SQL code to surface critical operational insights.
This technical abstraction is driving a broader shift toward hiring curious generalists who thrive in ambiguity and possess a discovery mentality. Human resources and people operations teams are increasingly prioritizing candidates who demonstrate strong adaptability, high curiosity, and the willingness to harness modern AI tools over those who only bring static, hyper-focused technical capabilities.
- **Legacy Workforce Model:** Narrow technical specializations (e.g., dedicated SQL coders, manual data extractors).
- **Emerging Workforce Model:** Adaptive generalists with deep domain context, high intellectual curiosity, and fluency in agent orchestration.
---
[13:58](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=838s) Moving from Tool Sprawl to the Platform-Driven Agentic Enterprise
Navigating the AI landscape has become overwhelming due to the sheer volume of startups and tech giants claiming to offer a "magic wand" for every business problem. For enterprise CIOs and IT administrators, chasing isolated widgets leads to severe tool sprawl, heightened security vulnerabilities, and significant ongoing maintenance overhead. Instead of acquiring single-point solutions based on novelty or viral demos, enterprises must adopt a scalable platform approach designed to solve systemic workflow bottlenecks across lines of business.
```mermaid
flowchart LR
accTitle: Three Phases of Enterprise AI Maturity
subgraph P1["Phase 1: Tool Adoption"]
T1["Point Tools & Widgets"]
T2["Time Savings KPIs (e.g., 20% faster campaigns)"]
end
subgraph P2["Phase 2: Process Decoupling"]
P3["Line of Business Redesign"]
P4["Removal of Operational Bottlenecks"]
end
subgraph P3["Phase 3: Agentic Enterprise"]
A1["Autonomous Cross-Functional Agents"]
A2["Velocity & Strategic Decision KPIs"]
end
P1 --> P2 --> P3
```
In a fully realized agentic enterprise, AI agents run across marketing, HR, finance, and engineering operations to orchestrate complex business logic. In the semiconductor industry, for instance, chip manufacturers leverage large language models and multi-agent R&D workflows to analyze massive datasets and accelerate design cycles to outpace competitors. Similar agentic ecosystems automate hyper-personalized creative production in marketing, streamline talent workflows in HR, and unlock real-time supply chain forecasting and demand planning in finance. As organizations progress through these maturity stages, their success metrics shift from basic time savings (such as saving 20% of the time required to build a marketing campaign) to strategic operational KPIs centered on decision velocity and business agility.
- **Phase 1 (Tool Utilization):** Fragmented point solutions focused on tactical, individual-level time savings.
- **Phase 2 (Process Decoupling):** Targeted workflow optimization designed to remove operational friction across lines of business.
- **Phase 3 (Agentic Enterprise):** Interconnected autonomous agents interacting seamlessly with humans, enterprise applications, and underlying data stores.
---
[18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) The "Spreadsheet of 100 Use Cases" and Building Muscle Through Failed POCs
Almost every enterprise begins its AI journey with "the sheet"—a sprawling spreadsheet cataloging 100 to 150 potential use cases gathered from various business units. Organizations typically plot these ideas onto a classic 2x2 matrix (mapping complexity against business value) and attempt to start with low-effort, high-impact quadrant initiatives. While categorizing use cases is a necessary first step, Schofield challenges the common marketing claim that modern AI proofs of concept (POCs) will rarely fail, arguing instead that failure is practically guaranteed and essential to success.
Schofield compares enterprise AI adoption to physical bodybuilding: "You're never going to get to be Mr. Olympia and Arnold Schwarzenegger on stage by going to the gym for a week. You have to build a muscle." Organizations must build operational, cultural, and technical data muscles by failing fast and embracing rapid iteration. Even if a team spends six months developing a capability only to have a newer foundation model render that work obsolete overnight, the institutional learning and agility gained during the process are what ultimately enable the enterprise to tackle and execute the next generation of business use cases.
- **The Use Case Dilemma:** Managing backlogs of 100+ proposed AI applications mapped across effort and impact matrices.
- **The Reality of POCs:** Accepting high failure rates as an unavoidable mechanism for building organizational resilience and technical competence.
- **Organizational Fitness:** Cultivating data foundations, cultural adaptability, and process maturity through continuous experimentation.
[20:57](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1257s) Managing Agent Sprawl and Enterprise-Grade Scale
Operating AI at enterprise scale introduces unprecedented organizational complexity. As adoption accelerates, enterprises will inevitably face an explosion of autonomous agents—potentially scaling into the hundreds or thousands across disparate business functions. Managing this environment presents technology leaders, CIOs, and CTOs with an uncharted and "not well-lit path" of operational governance.
```mermaid
flowchart TD
accTitle: Enterprise AI Agent Ecosystem Complexity
subgraph SAA["Disparate Agent Ecosystems"]
A1["Salesforce Agentforce"]
A2["ServiceNow Service Agents"]
A3["Data Agents (BigQuery / Databricks)"]
A4["Custom & Domain Agents"]
end
subgraph GOV["Enterprise Oversight & Control"]
B["IT & Executive Governance\n(Quality, Cost & Security)"]
end
A1 --> B
A2 --> B
A3 --> B
A4 --> B
```
The proliferation of vendor-specific agents—from Salesforce Agentforce to ServiceNow workflows and data-layer agents in Databricks or Google BigQuery—creates an overwhelming orchestration challenge. The core obstacle for technical leadership is determining how to weave these autonomous systems together to ensure enterprise-grade quality, reliability, and security without drowning in agent sprawl.
---
[21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) The CFO Call, ROI, and Moving Past "Token Maxing"
Enterprise AI initiatives are facing a critical turning point as financial scrutiny sets in. Many organizations are beginning to receive anxious inquiries from the CFO regarding spiraling operational expenditures. For months, development teams operated under an unbridled "token maxing mentality," prioritizing deployment speed and experimentation regardless of compute or inference expenses. However, burning through AI tokens faster than annual data center budgets without direct, measurable returns is fundamentally unsustainable.
```mermaid
flowchart LR
accTitle: Shift from Token Maxing to Durable Value
subgraph Phase1["Past Phase: Unbridled Speed"]
A["'Token Maxing'\nMentality"] --> B["High Token Spend\nLow P&L Tracking"]
end
subgraph Phase2["Current Phase: Durable ROI"]
C["P&L Alignment"] --> D["Top-Line Growth\n& Bottom-Line Efficiency"]
end
Phase1 ==>|"Pendulum Swing"| Phase2
```
The pendulum is rapidly swinging back toward financial accountability and durable value creation. AI leaders must clearly tie every initiative to the profit and loss (P&L) statement, showing tangible impact on either top-line revenue growth or bottom-line operational efficiency. When evaluating high-value versus low-effort project matrices, teams must prepare defensible value narratives for executive leadership.
- **Enterprise Culture:** Foster top-down organizational support alongside bottom-up grassroots innovation, ensuring teams have the freedom to fail fast.
- **Data Readiness:** Centralize and structure business data so AI models can immediately extract actionable insights.
- **Durable ROI:** Transition away from vanity AI experiments to projects mapped directly to measurable balance-sheet outcomes.
---
[23:59](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1439s) Q&A: Big Transformational Bets vs. Building Muscle on Small Use Cases
When asked whether organizations should bypass small, incremental use cases in favor of high-cost, high-value flagship projects, Schofield cautions against an all-or-nothing approach. While transformative use cases should remain the primary focus—analogous to making sure you are not "eating too much bread before the meal"—attempting to tackle massive organizational bottlenecks without prior operational experience poses significant downside risk.
```mermaid
flowchart LR
accTitle: Dual Track Strategy for Enterprise AI
subgraph Execution["Iterative Foundation"]
E1["Single Function Agents"] --> E2["Multi-Agent Workflows"]
E2 --> E3["Secure Data Connectors"]
end
subgraph Strategy["Transformational Target"]
T1["Core Flagship Use Case\n(Major P&L Impact)"]
end
Execution -->|"Builds capability to de-risk"| Strategy
```
Attempting a massive AI overhaul without foundational experience is like stepping into the ring against Mike Tyson—it is an avoidable loss. Organizations must build baseline competencies by developing single-function agents, orchestrating multi-agent workflows, and establishing secure data connections. Practicing on smaller, less intimidating problems builds the organizational muscle and technical guardrails required to de-risk and successfully execute large-scale, high-impact transformations.
---
[25:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1548s) Q&A: Legacy Data Readiness and Resetting Executive Expectations
Addressing the challenge of enterprise data readiness, Schofield emphasizes that there is no universal blueprint. Most established enterprises operate legacy application stacks—such as 25-year-old SAP deployments from the 1990s—and struggle with whether to modernize the application, centralize into an enterprise data warehouse, or leverage modern protocols like the Model Context Protocol (MCP) and intelligent APIs that allow models to interface directly with existing data stores.
```mermaid
flowchart TD
accTitle: Aligning Executive AI Expectations with Reality
subgraph Exp["Board & Executive Perception"]
E["Expectation of Immediate '10x Spark'\n& Magic Turnkey Solutions"]
end
subgraph Reality["Operational Reality"]
R1["Historical Paradigm:\nSteam Engine to Distributed Power"]
R2["Iterative Technical Buildout\n& Infrastructure Groundwork"]
end
Exp -.->|"Ground with clear roadmaps"| Reality
```
On the subject of unrealistic senior leadership KPIs, Schofield highlights the need to ground executives in reality. Leadership often expects an instant "10x spark" or magic turnkey solution. Drawing a historical parallel, Schofield compares the evolution of AI to the transition from steam engines to distributed electrical power grids—a profound shift that required significant time and infrastructure before delivering its full promise. While AI holds transformative potential for humanity, leaders must balance enthusiasm with clear project visibility and realistic execution roadmaps.
## Notable quotes
> [02:19](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=139s) second inning of the game — Jacob Schofield
> [02:19](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=139s) 10x mentality — Jacob Schofield
> [03:53](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=233s) productivity paradox — Jacob Schofield
> [10:24](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=624s) what keeps you up at night — Jacob Schofield
> [13:58](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=838s) magic wand — Jacob Schofield
> [18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) the sheet — Jacob Schofield
> [18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) You're never going to get to be Mr. Olympia and Arnold Schwarzenegger on stage by going to the gym for a week. You have to build a muscle. — Jacob Schofield
> [20:57](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1257s) not well-lit path — Jacob Schofield
> [21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) token maxing mentality — Jacob Schofield
> [23:59](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1439s) eating too much bread before the meal — Jacob Schofield
> [25:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1548s) 10x spark — Jacob Schofield
## Mentioned
Jacob Schofield, Google Cloud, Ai4, Alphabet, Google, Google Workspace, Gemini, Microsoft Copilot, Cloud Code, AWS, Microsoft Azure, Salesforce, YouTube, Salesforce Agentforce, ServiceNow, BigQuery, Databricks, Arnold Schwarzenegger, Mike Tyson, SAP, Model Context Protocol (MCP)
## Video description
Jacob Schofield - Enterprise AI GTM, Google Cloud
While AI technology is advancing rapidly, many organizations still struggle to translate these innovations into meaningful, transformational business value. This session explores the common roadblocks stalling enterprise AI initiatives and introduces a practical framework for connecting AI capabilities directly to bottom-line results.
Jacob Schofield of Google Cloud argues that enterprise AI is entering a second phase where surface-level automation yields diminishing returns, requiring organizations to restructure data architectures and workflows to unlock durable economic value.
The Productivity Paradox and Structural Workflow Redesign (3:53)
Schofield observes that corporate leadership is applying intense pressure on technical executives to deliver ten-fold productivity leaps and eliminate legacy manual processes. However, basic automations—such as deploying internal chatbots for document summarization or slide generation—only shave off isolated minutes rather than repositioning an enterprise competitively. Simply buying software seat licenses like Google Workspace with Gemini or Microsoft Copilot risks superimposing new tools onto obsolete operational habits.
To explain this lag, Schofield cites the economic "productivity paradox," comparing modern AI adoption to the Industrial Revolution's transition from steam to electricity. Early factories merely replaced central steam engines with electric generators while retaining legacy pulleys and mechanical shafts, taking roughly thirty years to decentralize power to individual workstations and unlock modern assembly-line efficiency. Realizing transformational AI value similarly requires overcoming fragmented SaaS and on-premises data silos, internal cultural resistance, and the absence of end-to-end operational blueprints.
The AI Fluency Framework and Generalist Talent (7:18)
To navigate organizational transformation, Google Cloud uses a five-layer "AI Fluency Framework" spanning unified data foundations, cultural literacy, evolving talent profiles, scalable technical platforms, and operational redesign. Schofield asserts that organizational design must progress across a three-stage continuum: automating repetitive bottlenecks, augmenting human workflows with parallel automated systems, and ultimately transforming into a hybrid operating environment governed by enterprise-grade security and accuracy controls.
This structural evolution is reshaping enterprise hiring. Because natural language increasingly operates as a universal programming layer that enables non-technical staff to query data directly without writing SQL, demand for narrow technical specialists is receding. Schofield reports that organizations are shifting their talent strategies toward curious, adaptable generalists who possess strong domain context and demonstrate an aptitude for orchestrating agentic tools within ambiguous environments.
Transitioning from Tool Sprawl to Agentic Workflows (13:58)
Schofield warns enterprise technology leaders against chasing disconnected point solutions and viral vendor demos, which generate tool sprawl, integration overhead, and security liabilities. Instead, organizations should advance from tactical tool adoption to process decoupling and finally to a platform-driven agentic enterprise. In this mature phase, interconnected autonomous agents execute complex cross-functional business logic—such as accelerating design cycles in semiconductor manufacturing, automating creative assets in marketing, or modeling supply-chain forecasts in finance.
Within go-to-market teams, for example, agentic augmentation moves far beyond automated email drafting or basic CRM entry. Autonomous agents can ingest SEC 10-K filings, earnings calls, and multimedia executive interviews to generate customized value hypotheses. This allows sales professionals to bypass generic discovery questions and arrive at executive meetings equipped with pre-synthesized strategic insights tailored directly to specific C-suite priorities.
Balancing POC Muscle Building with CFO Accountability (18:03)
Addressing the tendency of enterprises to catalog backlogs of 100 or more AI use cases on effort-versus-impact matrices, Schofield challenges vendor claims that proofs of concept rarely fail. He compares AI adoption to physical bodybuilding, asserting that failing fast through iterative experimentation builds the requisite organizational, data, and cultural muscles. Even when rapid foundation model advancements render months of initial development obsolete, the institutional competency gained prepares the organization to tackle higher-value initiatives.
At the same time, enterprise initiatives face mounting financial scrutiny as CFOs push back against undisciplined "token maxing" that inflates compute costs without verifiable returns. Technical leaders must tie AI roadmaps directly to the profit and loss statement through top-line growth or bottom-line efficiency. Schofield advises adopting a dual-track strategy: practicing on single-function agents and secure multi-agent workflows to build operational guardrails before committing resources to high-risk, flagship transformations.
Schofield emphasizes that leaders must actively reset board expectations away from immediate turnkey miracles and toward realistic multi-year roadmaps. Modernizing decades-old enterprise application stacks requires persistent infrastructure groundwork, cross-vendor agent governance, and disciplined fiscal execution to translate raw model capabilities into sustainable business advantage.
精炼阅读摘要
Google Cloud 企业 AI 市场拓展负责人 Jacob Schofield 认为，企业 AI 正在迈入超越表层自动化与聊天机器人的“第二局”，企业不能仅靠采购席位软件实现跃升，而必须彻底重构业务流程与数据架构，经历类似从蒸汽动力向分布式电网过渡的长期结构性转型。
打破生产力悖论与表层自动化局限 (2:19)
企业董事会和高管层当前正对技术领袖施加巨大压力，要求实现颠覆性增长与效率跃迁，但简单的草拟邮件、摘要文档或回答基础问答等表层自动化工具无法带来竞争格局的重构。Schofield 援引 1990 年代后期的“生产力悖论”以及工业革命的历史指出，早期工厂引入电动机时，只是机械地替换了中央蒸汽机，却沿用了原有的轴承与滑轮网络，直到三十年后将动力分散到各个工作台建立现代装配线，才真正释放了电气化的经济潜力。在现有遗留流程上直接堆砌 Google Workspace with Gemini 或 Microsoft Copilot 席位，就像是用电动机替换蒸汽机而不改动传输滑轮，无法带来质的飞跃。
AI 流畅度框架与新型员工素养 (7:18)
为推动企业从被动采用工具走向实质性变革，Google Cloud 总结出涵盖五个层面的“AI 流畅度框架”：打破跨 AWS、Microsoft Azure 和 SaaS 平台的孤岛以奠定数据基础；构建企业文化与技术素养；重构员工能力结构；搭建支持自主智能体（Agent）的可扩展技术基础设施；以及沿着“自动化、增强、变革”三个阶段进行组织设计。随着自然语言成为通用编程接口，企业以往对编写 SQL 等狭隘硬技能人才的刚性需求正在减弱，具备好奇心、能应对模糊性并善于调度智能体协同的通识型人才正成为企业招聘的核心重点。
从单点工具泛滥走向平台化智能体生态 (10:24)
盲目追逐市场上层出不穷的单点小工具，正给企业 IT 部门带来严重的工具泛滥（Tool Sprawl）、安全漏洞以及高额维护负担。企业需要转向平台化架构，在半导体研发、市场营销、人力资源和财务等核心业务中部署协同运作的多智能体工作流。例如在市场拓展团队中，智能体能够自动抓取 SEC 10-K 申报文件并解析高管公开访谈，在拜访客户高管前便生成咨询级的商业价值假设，从而消除了泛泛的背景询问。随着组织成熟度演进，考核指标也从单纯节省个别工时的局部指标，转向以业务决策速度和敏捷性为主的战略性 KPI。
财务审视回归与通过失败构建组织肌肉 (18:03)
企业 AI 项目正迎来严格的财务审视时刻，以往不计推理与算力开支的“代币最大化（Token Maxing）”粗放实验模式已不可持续，CFO 开始要求所有方案明确挂钩损益表（P&L）的营收增量或成本节约。针对企业梳理出上百个用例却面临 POC（概念验证）频繁失败的现状，Schofield 认为试错是组织构建数据与文化“肌肉”不可避免的必经过程。面对遗留 IT 系统（如 25 年前的老旧 SAP 系统）与数百上千个智能体共存的管理挑战，企业应当在推进具有重大 P&L 影响力的核心用例的同时，从小规模单功能智能体逐步积累集成与安全治理经验，而非在缺乏底座能力时孤注一掷。
Schofield 强调，任何宣称能精准预测未来 6 至 24 个月企业 AI 走向的言论都是不切实际的；技术领袖必须设法重置董事会对即时“十倍增长奇迹”的幻想，在应对智能体激增与老旧数据治理复杂性的过程中，保持战略定力并建立务实的落地路线图。
Detailed notes — zh
## Background
该视频由频道 Ai4 于 2026 年 8 月 21 日发布在 YouTube 平台。内容由 Google Cloud 企业级 AI GTM 负责人 Jacob Schofield 主讲，旨在探讨企业在将前沿 AI 技术转化为切实商业价值时遇到的普遍阻碍，并分享如何将 AI 能力与业务底层成果直接挂钩的实操框架。
## TL;DR
演讲指出企业 AI 探索正从浅层的单点工具提效迈入深度重构业务的“第二局”，并通过工业电气化耗时 30 年的历史类比警示企业警惕“生产力悖论”与工具蔓延陷阱。Google Cloud 提出了包含数据底座、组织文化、拥抱模糊性的通才画像与平台化能力的“AI 流畅度框架”，推动业务从单纯自动化演进为多智能体协同的 Agentic 企业。同时，面对高达 99% 的 POC 失败率与 CFO 对 Token 成本的严苛审视，企业需正视试错价值以锤炼组织肌肉，让 AI 投资坚决回归损益表（P&L）与可衡量的财务 ROI。
## Atmosphere & style
整体基调务实、坦诚且极具一线实战感，主讲人摒弃了宏大的技术乌托邦叙事，以极其接地气的行业洞察直击高管层的焦虑与阵痛。通过“健身练肌肉”、“与泰森打拳击”以及“飞机座椅背后的独角兽广告”等生动诙谐的比喻，现场在直面现实残酷（如断言听众的 POC 同样会失败、模拟 CFO 的追责来电）与给予组织耐心包容之间形成了精妙张力，营造出一种穿透行业喧嚣、直面商业本质的清醒与共鸣。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=4s) 面对 AI 变革的技术不确定性与极端舆论噪音，企业领导者应秉持宽容与耐心，但必须带着紧迫感快速行动以应对已至的颠覆时刻。
- [02:19](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=139s) 企业 AI 探索进入“第二局”，管理层不再满足于文档生成、PPT 制作等浅层省时工具，亟需寻找能重塑业务流程与驱动 10 倍增长的实质性变革价值。
- [03:53](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=233s) 重温工业电气化历经 30 年从简单动力替换走向现代流水线的历史，单纯采购通用 AI 工具无法跨越生产力悖论，唯有重构生产组织形式才能释放质变效能。
- [07:18](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=438s) 谷歌的“AI 流畅度框架”以数据底座和组织文化为基石，推动企业沿“自动化、协同增强、人机共生变革”三阶段演进。
- [10:24](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=624s) 销售 GTM 正在由单点邮件自动化升级为 Agentic 咨询式销售，智能体自主分析 10-K 年报与高管公开言论以构建定制化商业价值假设。
- [12:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=768s) 人才画像由狭窄的垂直硬技能专才（如纯 SQL 编写者）转向具备强烈好奇心、能拥抱模糊性并熟练调度智能体的复合型通才。
- [13:58](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=838s) 警惕单点工具采购导致的“工具蔓延”与安全维护黑洞，企业应基于平台化能力构建贯通核心业务线的 Agentic 企业架构。
- [18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) 正面接纳 99% 的 AI POC 失败并快速试错，每一次受挫都是在为企业沉淀不可或缺的数据底座与组织心智“肌肉”。
- [20:57](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1257s) 多平台智能体的大规模落地将引发“Agent 蔓延”挑战，技术管理者亟需建立跨异构环境的统一编排、治理与安全扩展机制。
- [21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) 告别盲目消耗资源的“Token 狂飙”心态，企业 AI 投资必须直接映射到损益表（P&L），以顶线增收与底线降本证明持久 ROI。
- [23:59](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1439s) 平衡大场景落地与能力演进：核心高价值大场景是“主菜”，但团队需在低风险用例中打磨多智能体工作流能力，避免盲目单挑巨型项目导致惨败。
- [25:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1548s) 面对老旧技术栈与管理层过高的脱离实际预期，需利用 API 与 MCP 实现灵活数据就绪，并建立透明规划以拉齐对渐进式工业演进周期的客观认知。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=4s) 峰会尾声的坦诚对话：喧嚣中的企业 AI 现实
作为整场大会最后一天的最后一个议题，Google Cloud 西部地区及半导体行业企业级 AI GTM 负责人 Jake Schofield 首先向全场能够坚持到最后的听众表达了由衷的敬佩。他谦逊地表示，大会期间登台的许多嘉宾都是业界顶尖的智慧大脑，而他自己并不以此自居；但他拥有一项独特的优势——每天置身于业务最前线，与大量企业客户深入交流。这让他得以一手洞见各行各业的领导者究竟在如何思考 AI、如何针对人才、产品及竞争格局进行战略布局，以及在技术狂飙突进时所遭遇的种种现实陷阱与困惑。
面对当前外界铺天盖地的噪音，Jake 直言企业领导者面临的认知环境极为复杂。媒体与行业舆论极端分化：一方面有人描绘全民基本收入（UBI）的技术乌托邦，另一方面又有人渲染机器人统治世界的末日图景。他坦言，如果有人声称自己确切知道未来 6 个月、12 个月或 24 个月后技术会走向何方，那绝对是在说谎。因此，在与企业客户沟通时，他的首要建议是给团队和自己保留一点宽容与耐心（"give yourself some grace"）。拥抱前所未有的技术变革固然令人兴奋且充满特权感，但整个过程注定伴随着困惑与阵痛。企业应当秉持一种平衡的心态：“保持耐心，宽容对待自己，但必须带着紧迫感行动”（"be patient, give yourself some grace but act with urgency"），因为对于很多企业而言，颠覆时刻已经到来。
- **行业一线视角**：负责 Google Cloud 美西与半导体 vertical 的企业级市场落地，每天直接对接企业决策层。
- **现实认知法则**：拒绝非黑即白的舆论噪音，承认未来 6 到 24 个月的不确定性，以审慎而敏捷的姿态应对变革。
---
### [02:19](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=139s) 告别工具化省时：高管层对“变革性火花”的执念
随着企业对生成式 AI 的探索从最初的尝鲜阶段步入“第二局”（second inning），高管们的关注点正在发生质的转变。诸如利用聊天机器人写文档、辅助生成演示文稿这类单纯节省时间的表面生产力指标，已经无法满足领导层的胃口。企业决策者真正渴望看到的是能够彻底改造业务、拉开竞争差距、驱动营收增长并显著提升运营效率的“实质性变革价值”。单靠自动化本身，从管理层与董事会的视角来看，根本算不上核心价值引擎。
如今，来自董事会、CEO 的自上而下的重压正直接传导给 CTO、CIO 和首席产品官（CPO）。他们被赋予了极高的期望，必须去发掘那个能够重塑整个公司的“火花”——带来所谓的“10 倍思维”（10x mentality）、推动收入呈爆发式增长、并将存在了十余年的陈旧流程彻底自动化。在这一背景下，如果技术团队仅仅给出一个“我们准备上线一个聊天机器人”的陈腐方案，已经完全无法交差。领导者需要的是业务范式的重构，而非简单的微小提效。
```mermaid
flowchart LR
accTitle: 企事业对 AI 价值诉求的演进阶段
subgraph S1["初期工具探索 (第一局)"]
A["局部自动化与效率"] --> B["时间节省指标<br/>(写文档/做 PPT/问答)"]
end
subgraph S2["深度变革阶段 (第二局)"]
C["重构业务底层逻辑"] --> D["变革性业务价值<br/>(10x 增长/流程颠覆)"]
end
S1 -. "无法满足董事会期望".-> S2
```
- **第二局转折点**：从“工具型提效”（问答助手、办公写作）转向“重构核心竞争力与营收”。
- **高管层的现实困境**：CTO、CIO、CPO 承载着自上而下的指标压力，亟需从底层业务逻辑中寻找 10 倍增长的破局点。
---
### [03:53](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=233s) 生产力悖论与百年镜像：从蒸汽动力到流水线的启示
回顾技术演进史，重大突破的出现与真正价值的释放之间往往存在漫长的滞后。Jake 援引了 1990 年代末著名的“生产力悖论”（Productivity Paradox）理论指出：技术创新初期的实际价值释放绝非线性上升。在 2023 年底至 2024 年初，行业内曾充斥着一种盲目乐观的狂热——“AI 能搞定那个”（"AI can do that"），似乎只要引入 AI 一切问题就能迎刃而解。然而两年过去，尽管 AI 做了很多，其创造的宏观商业价值却远未达到当初狂热预期的峰值。
为了清晰阐释这种滞后，Jake 提出了一个横跨百年的工厂动力演进类比：在 19 世纪，绝大多数工厂由中央蒸汽机驱动，工厂车间上方架设着复杂的齿轮、皮带与传动轴网络，将动力层层分发给各个工位。当电力技术刚刚诞生时，最早的工厂仅仅是将中央蒸汽机替换成了一台集中式电动机，而原有的皮带、滑轮和传动轴系统原封不动地保留着。这种简单的“动力替换”并没有带来生产力的质变。工厂整整花费了 30 年的时间，才彻底打破原有的中心化动力架构，将电动机分散小型化到每一个独立的工作台（Workstation）。正是这种生产组织形式的根本重构，才真正催生了第二次工业革命中标志性的现代流水线作业。
对照当下的生成式 AI 浪潮，很多企业陷入了同样的思维误区：以为给员工买几个现成的工具（如 Copilot、Gemini、Cloud Code、Cloud Co-work），就能自然而然地跨入现代化生产线时代。采购工具固然能够帮助团队建立起对 AI 的适应性与紧迫感肌肉，但单纯的工具采纳绝不等于变革。现实中，客户反复抛出的痛点往往是：董事会凭空定下了“10 倍思维”、“运营效率提升 50%”的目标，但技术栈与数据现状却支离破碎，既有对 AI 极其推崇的激进派，也有抗拒改变的保守员工，整个组织尚未做好迎接生产范式转移的准备。
```mermaid
flowchart TD
accTitle: 电气化历史与当前 AI 落地路径的演进对照
subgraph History["历史镜鉴：工业动力架构演进 (历时30年)"]
H1["中央蒸汽动力<br/>(复杂的滑轮与齿轮轴系)"] --> H2["集中式电动机替换<br/>(保留旧有机械传动结构)"]
H2 --> H3["去中心化独立工位动力<br/>(现代流水线诞生/效能质变)"]
end
subgraph EnterpriseAI["现实映射：现代企业 AI 演进路径"]
E1["人工与传统软件<br/>(陈旧割裂的业务流程)"] --> E2["采购通用 AI 工具<br/>(Copilot/Gemini 简单嵌入)"]
E2 --> E3["全链路重构与 Agentic 架构<br/>(业务流彻底重塑/变革价值)"]
end
H2 -. "仅替换动力源<br/>未改变组织形式".- E2
H3 -. "重构生产架构<br/>释放真正生产力".- E3
```
- **生产力悖论的核心机制**：突破性技术出现后，价值曲线会先经历短暂跃升，随后进入漫长的重构平台期。
- **电气化历史教训**：从集中式替换到工位去中心化耗费 30 年，单纯更换动力源无法产生流水线效应。
- **企业采购陷阱**：单纯采购 AI 软件授权只能“锻炼肌肉”，无法直接转化为业务质变。
---
### [07:18](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=438s) 谷歌的 AI 流畅度框架：重塑数据、组织与人机协同
面对技术颠覆带来的共同挑战，Google 自身同样无法置身事外。外界热议诸如“ChatGPT 问世将冲击传统搜索业务”等论调，母公司 Alphabet 不仅需要思考自身的战略航向，更在内部实践中提炼出了一套旨在提升变革速度的“AI 流畅度框架”（AI Fluency Framework）。该框架旨在帮助企业在混乱中梳理优先级，从底层的坚固基石逐步向上构建变革能力。
该框架由两大底层支撑与三大上层支柱构成。最底层的第一块基石是“数据底座”（Data Foundation），也是决定成败的关键：任何智能体（Agent）或基础模型的表现上限，完全取决于它能够访问的数据质量。现实中绝大多数企业不仅分散使用着数百个 SaaS 应用程序，数据还深度割裂在 AWS、Azure、Google Cloud 以及本地机房（on-premises）的数据孤岛中。第二块基石则是“组织文化与素养”（Culture & AI Literacy），仅仅分发账号无法改变日常工作范式，必须培育能够理解并主动重塑业务流程的文化土壤。
在此之上，框架进一步聚焦三大核心维度的演变：
1. **组织架构设计（Organizational Design）**：赋能各业务线（Line of Business）主管与一线员工，探索贴合自身职能的落地路径。
2. **人才画像重构（Employee Profile）**：从过往死板的“技能型”（例如纯 SQL 开发人员、财务、传统会计）转向拥有通用驾驭能力的复合型人才。例如以往需要专人编写 SQL 查询，如今员工只需使用自然语言提问，模型即可翻译并交互，人与数据的沟通门槛被彻底抹平。
3. **规模化工具体系（Tooling for Scale）**：不仅指端给最终用户的辅助工具，更指代提供给技术团队、用于构建并运行能够在大规模场景下自主与数据和应用交互的 Agent 服务（Agentic Services）的开发基建。
在具体的业务演进路径上，Google 将其划分为三个梯级：首先是针对耗时陈旧流程的“自动化”（Automate）；进而迈向构建能与人类员工并肩协作的“增强系统”（Augment）；最终迈入真正的“变革阶段”（Transform）——在保障准确性、响应速度、安全性与合规治理的前提下，让 AI 与人类在混合环境中全天候、高频且流畅地协同运作。
- **数据现实困境**：企业普遍面临跨多云（AWS、Azure、GCP）与本地私有数据中心的严重数据碎片化。
- **人才画像转移**：从窄域硬技能专才（如纯手动写 SQL 的数据分析师）向善于运用自然语言撬动系统能力的复合型通才演进。
- **演进三部曲**：自动化（Automate） $\rightarrow$ 协同增强（Augment） $\rightarrow$ 流畅的人机共生变革（Transform）。
### [10:24](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=624s) 销售 GTM 的演进：从任务自动化到 Agentic 深度洞察
谈及业务实践，Jacob Schofield 分享了自己主管市场拓展（GTM）团队的切身体会。作为销售领导者，他最核心的愿望就是让团队每个人都能卖出超出指标十倍的业绩。回顾企业借助 AI 赋能销售人员的过程，最初的第一步仅仅停留在单纯的自动化工具：自动起草电子邮件回复、自动化处理投标文件（RFP）或者是联动 Salesforce 与 BI 工具生成客户规划。这一阶段满足了基础的自动化需求，但组织很快就走向了第二步——如何真正增强（Augment）销售人员日常运作的效能。
为了达成深层赋能，团队开始构建 Agentic 工作流（智能体工作流）。企业不再只是让 AI 填表格，而是给智能体指派复杂的端到端任务：深入审查核心战略客户、研读其 10-K 年报、抓取客户 CEO 在 YouTube 上的所有公开发言与视频存入数据存储库，并深度解析其领导风格与战略重心。借由对海量数据的归纳与提炼，销售团队得以构建定制化的“价值假设”（Custom Value Hypothesis），转变为咨询式的销售模式。Schofield 强调，优秀的销售“永远不必跑去当面问客户 CEO 究竟因何事彻夜难眠”（"You should never have to go ask a CEO what keeps them up at night"）——这些信息全部公开在互联网、财报与业绩电话会中。借助 Agentic 架构，销售可以在见客户之前，依据对方具体职务（无论是 CFO、CIO、CTO 还是产品副总裁）精准框定其关切点，带着成熟的业务构想登门拜访。
```mermaid
flowchart TD
accTitle: 销售 GTM 团队的 AI 演进路径
subgraph S1["第一阶段：自动化 (Automation)"]
A["邮件回复 / RFP 处理"]
B["Salesforce 与 BI 数据连接"]
end
subgraph S2["第二阶段：增强与赋能 (Augment)"]
C["Agent 研读 10-K 年报与财报"]
D["抓取分析 CEO 视频与公开言论"]
E["形成定制化价值假设与咨询叙事"]
end
subgraph S3["第三阶段：业务转型 (Transformation)"]
F["依据 CFO/CIO/CTO 痛点精准对话"]
G["以智能体驱动商业决策与业绩突破"]
end
S1 --> S2 --> S3
```
- **初级自动化痛点**：早期的 RFP 自动生成和 CRM 关联虽然提效，但无法改变销售策略的深度与成单格局。
- **Agentic 咨询式销售**：智能体通过融合非结构化多媒体数据（如公开视频）与结构化财报，直接输出高阶的业务叙事。
- **高管对话框架**：根据目标客户管理层（如技术、财务或业务条线）的不同背景，由智能体自动生成针对性视角与应对策略。
---
### [12:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=768s) 人才画像重塑：从垂直技能专才到拥抱模糊性的通才
AI 与智能体体系的普及，对企业的人才选拔与员工画像（Employee Profile）带来了颠覆性的冲击。过去，企业招聘有着极其清晰且垂直的硬技能边界：招募专门编写复杂查询的 SQL 开发者、分析报表的数据分析师、依靠既往人脉的大客户销售，以及精通特定模型的财务分析师。然而随着智能体的介入，员工调用数据的门槛被彻底打碎——人们不再需要亲自手写 SQL 才能与数据对话，只需确保底层智能体准确对接了核心数据源，即可直接获取深度业务洞察。
这种技术更迭使得极其专精但狭隘的技能变得不再不可替代，取而代之的是一种新型人才画像：具有强烈好奇心、高度适应不确定性与模糊环境、具备探索精神并在某个领域有独到眼光的通才（Generalist）。人力资源与组织运营部门正在全面经历这场人才标准的重组：企业不再渴求按部就班执行固定流程的人，而是迫切需要能够自如驾驭 AI 智能体、快速拥抱变革并将前沿能力转化为实际商业价值的复合型人才。
- **技能门槛的去中心化**：SQL 编写等硬编码要求被自然语言驱动的智能体取代，业务人员可直接与底层数据交互。
- **核心软实力崛起**：组织选人标准全面向“强好奇心”、“在模糊性中生存（Thrive in Ambiguity）”以及“探索发现心态”倾斜。
- **HR 运营转向**：传统的人力资源技能树正被重构，员工的核心价值在于能否驱动 AI 工具创造边际增量。
---
### [13:58](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=838s) 摆脱工具迷思与工具蔓延：平台化构建 Agentic 企业
面对市场上扑面而来的 AI 浪潮，Schofield 坦言这是他最喜欢探讨的话题。眼下不管是初创公司，还是来自 Google、微软或亚马逊的销售，每天都在向企业高管鼓吹自己拥有“魔法棒”（"magic wand"）。这种纷乱的市场噪音让很多企业陷入了“工具蔓延”（Tool Sprawl）的严重困境：IT 管理员或 CIO 面临严峻的数据安全隐患与维护黑洞——组织不可能为每个小场景各采购一个独立的小工具，然后再为此付出高昂的长期维护代价。
Google Cloud 在与客户沟通时，始终强调必须将重点放在平台级能力（Platform Capability）上。企业必须超越某位员工“看了 YouTube 视频后想要尝鲜”的局部单点需求，以平台化思维消除跨部门的流程瓶颈，将 AI 映射到真正的业务线（LOB）重大挑战中。真正的“Agentic 企业”（Agentic Enterprise）意味着全方位的业务升级：在芯片制造半导体行业，企业利用大语言模型融合海量研发数据与复杂多智能体流程加速下一代芯片设计；在市场营销端，实现完全自动化且极度个性化的内容生成；在人力与财务端，智能体则自主负责人员调度、供应链推演与需求预测。
| 演进阶段 | 核心关注点 | 典型应用与瓶颈 | 衡量 KPI 变化 |
|:--- |:--- |:--- |:--- |
| **工具利用阶段 (Tool Utilization)** | 单点工具应用与尝鲜 | 自动化营销文案、孤立辅助软件；面临工具蔓延与安全隐患 | 以微观效率为主（如：“节省了 20% 活动筹备时间”） |
| **流程解耦阶段 (Process Decoupling)** | 业务线深度重构 | 打破流程堵点，针对跨部门大问题实施定制化重组 | 运营瓶颈的消除率与流程吞吐量改善 |
| **全面智能体化 (Agentic Enterprise)** | 平台化、自主协同体系 | 研发加速、超个性化营销、跨系统跨数据的自主决策智能体 | 宏观商业决策速度、敏捷预测能力与综合竞争优势 |
---
### [18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) 直面 99% 的 POC 失败：在试错中锤炼企业的组织肌肉
在谈及实施 AI 的现实阻力时，Schofield 幽默地询问台下听众是否手头都有一份“神秘表格”——几乎每个正在推进 AI 转型的管理层手里都攥着一张列满 100 到 150 个 Use Case 的庞大电子表格。大家依照咨询机构的高价值、低复杂度象限（Gartner 魔力象限式的划分法则），试图从中挑选出“低投入、高回报”的项目作为起点。然而现实极为骨感，会场展示板上往往写着“99% 的 AI POC 都会失败，但您的不会”，对此 Schofield 直截了当地提出异议：“我不同意，你的 POC 一样会失败”（"I disagree. Yours will."）。
在 Schofield 看来，AI 项目的持续受挫、反复启停、推翻重来是不可避免的客观规律，企业必须学会“快速失败”（Fail Fast）。他打了一个生动的比方：正如不可能去健身房练上一周就变成施瓦辛格夺得健美先生（Mr. Olympia）一样，企业绝不可能挥一挥“魔法棒”就能瞬间拥有现代化装配线般的 AI 能力。每一次 POC 失败，都是在为企业构建心智肌肉、文化肌肉以及底层的数据底座肌肉。即便团队苦心研发 6 个月的方案可能瞬间被供应商发布的新一代基础模型所覆盖，但只要保持敏捷并加速试错，团队就能在这个尚无现成范本的探索之路上抢得先机。当未来的技术形态与平台架构真正成熟时，那支已经在泥泞中练就了组织肌肉的团队，才能真正有能力将规模化解决方案落地生根。
- **用例表格的普遍陷阱**：企业普遍热衷于罗列上百个 AI 用例并在象限中寻找低垂果实，但往往缺乏落地韧性。
- **正面接纳 POC 失败**：失败是构建技术理解力与组织信心的必经之路，避开失败往往意味着避开了真正核心的系统化改造。
- **健身与肌肉理论**：把 AI 能力建设比作健美训练，强调没有捷径，必须经历长时间的底层数据梳理与文化磨炼。
### [20:57](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1257s) Agent 蔓延与企业级规模化管理的挑战
随着企业 AI 技术的落地演进，真正的难题在于如何实现大规模运行。如果未来的技术走势如预期发展，企业内部将会出现成百上千个 Agent 同时运作的局面。对于技术负责人、管理员、CIO 和 CTO 而言，如何统一管理这些分散的系统将是一条极其艰难且尚未被充分探索的道路。企业将面对各种工具与智能体并存的复杂环境：Salesforce 的 Agentforce 在运行、ServiceNow Agent 在处理工单、Databricks 或 BigQuery 中有专职数据 Agent，各业务线也都有独立的 Agent。这种多智能体交织的状态极易演变为管理灾难，如何在这种碎片化生态中维持企业级的品质、安全和可扩展性，是技术领导者必须攻克的核心课题。
```mermaid
flowchart TD
accTitle: 多智能体企业级管理架构挑战
subgraph S1["多智能体运行环境 (Agent Sprawl)"]
A["Salesforce Agentforce 智能体"]
B["ServiceNow 业务工单智能体"]
C["Databricks / BigQuery 数据智能体"]
D["定制化业务 Agent"]
end
subgraph S2["企业管理层治理"]
E["CIO / CTO / 技术管理者"]
end
subgraph S3["核心治理目标"]
F["统一治理与编排"]
G["企业级品质与可扩展性"]
H["ROI 与成本控制"]
end
A --> E
B --> E
C --> E
D --> E
E --> F
E --> G
E --> H
```
---
### [21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) CFO 的来电：告别“Token 狂飙”与回归 P&L 财务价值
企业内部的 AI 投资正在迎来财务层面的严厉审视。企业可能很快就会接到来自 CFO 的问责电话：“我们消耗的 Token 费用已经超过了 2025 年整个数据中心的预算，到底发生了什么？我们是在盲目追求 Token 消耗上限吗？你们到底构建了什么，能配得上耗资 1000 万美元的 LLM Token？”过去那种不计成本、盲目追求速度的开发模式正在迅速退潮。当时企业要求数百甚至上万名开发者全力狂飙，但这种“Token Maxing”心态并不具备长期可持续性。
任何无法直接作用于营收增长（Top-line）或效率提升（Bottom-line）的 AI 举措，本质上都是一次“假起跑”。在企业评估 AI 投资决策并绘制“高价值、低实施难度”矩阵时，必须直接将项目映射到损益表（P&L）上。来自高层的严苛质询已然不可避免，技术团队必须能够清晰汇报每一笔投资换取的真实价值与持久 ROI。
- **三大核心要点总结：**
- **建设组织文化**：让全员做好迎接技术浪潮的准备，鼓励自顶向下的快速试错容错机制，同时不压制自底向上的创新活力。
- **做好数据就绪**：将数据置于可充分利用的状态，确保在价值释放时能够顺畅调度。
- **聚焦持久价值与 ROI**：将技术方案与损益表的顶线增收与底线降本紧密挂钩，只做真正产生商业价值的决策。
---
### [23:59](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1439s) 问答环节：大场景落地与能力演进的平衡艺术
针对现场观众提出“业界常告诫不要做小场景，而应聚焦高成本、高价值的大规模用例”这一观点，演讲者指出大型用例确实应当放在桌子正中央作为“主菜”，避免在正餐前因为吃太多面包而本末倒置。然而，解决大型问题极其困难，直接挑战庞然大物就像在毫无准备的情况下去和迈克·泰森（Mike Tyson）打拳击一样，注定会遭遇惨败。
因此，企业必须在解决相对低风险、不那么可怕的较小问题过程中锻炼自身“肌肉”。通过构建特定功能的单个 Agent、打通多智能体工作流（Multi-agent Workflows）以及建立安全的数据连接通道，团队能够在不暴露于巨大业务风险的前提下沉淀实战能力。既不能忽视代表转型价值的核心大场景，也要给自己留出磨炼技能的缓冲空间，兼顾两者的协同推进。
---
### [25:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1548s) 问答环节：老旧架构下的数据就绪与高管预期管理
在被问及如何让数据真正做到“AI Ready”时，演讲者坦言由于各企业基础设施差异巨大，并不存在标准通用答案。大量客户仍然运行着 25 年前乃至 1990 年代的 SAP 老旧技术栈，面临着究竟是重构应用、还是将数据统一搬迁至中央数据仓库的抉择。如今的解决方案更加多样化：部分场景需要将数据集中存储，而在另一些场景中，AI 可以借助模型上下文协议（MCP, Model Context Protocol）和 API 集成直接接入现有系统。
对于如何应对高级管理层脱离实际的高预期（如盲目追求“10 倍爆发”），演讲者建议必须将他们拉回现实：
- **拒绝空中楼阁**：正如不能轻信“飞机座椅背后广告里承诺会有独角兽飞过来解决一切问题”，AI 的普及更像是从蒸汽机走向分布式电力的漫长工业演进。
- **正视能力边界与时间周期**：尽管 AI 在疾病攻克、气象预测与农业丰产方面展现出惊人的潜力，但变革绝不可能在一夜之间全部兑现。
- **平衡预期与路线透明度**：既要传达团队全情投入的态度，也要直面“预期与现实”之间的客观鸿沟，向管理层提供清晰、透明且脚踏实地的推进规划。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=4s) 保持耐心，宽容对待自己，但必须带着紧迫感行动（be patient, give yourself some grace but act with urgency）。 — Jacob Schofield
> [03:53](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=233s) 给员工买几个现成的工具并不等于进入了现代流水线时代。 — Jacob Schofield
> [07:18](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=438s) 任何智能体（Agent）或基础模型的表现上限，完全取决于它能够访问的数据质量。 — Jacob Schofield
> [10:24](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=624s) 优秀的销售永远不必跑去当面问客户 CEO 究竟因何事彻夜难眠（You should never have to go ask a CEO what keeps them up at night）。 — Jacob Schofield
> [12:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=768s) 企业不再渴求按部就班执行固定流程的人，而是迫切需要能够自如驾驭 AI 智能体、在模糊性中生存（Thrive in Ambiguity）的通才。 — Jacob Schofield
> [18:03](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1083s) 会场展示板上写着‘99% 的 AI POC 都会失败，但您的不会’，我不同意，你的 POC 一样会失败（I disagree. Yours will.）。 — Jacob Schofield
> [21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) 我们消耗的 Token 费用已经超过了 2025 年整个数据中心的预算，到底发生了什么？我们是在盲目追求 Token 消耗上限吗？你们到底构建了什么，能配得上耗资 1000 万美元的 LLM Token？ — Jacob Schofield
> [21:45](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1305s) 任何无法直接作用于营收增长（Top-line）或效率提升（Bottom-line）的 AI 举措，本质上都是一次‘假起跑’。 — Jacob Schofield
> [23:59](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1439s) 直接挑战庞然大物就像在毫无准备的情况下去和迈克·泰森（Mike Tyson）打拳击一样，注定会遭遇惨败。 — Jacob Schofield
> [25:48](https://www.youtube.com/watch?v=0DCgPDwJ0R0&t=1548s) 不能轻信‘飞机座椅背后广告里承诺会有独角兽飞过来解决一切问题’，AI 的普及更像是从蒸汽机走向分布式电力的漫长工业演进。 — Jacob Schofield
## Mentioned
Jacob Schofield, Google Cloud, Alphabet, Ai4, ChatGPT, Gemini, Copilot, Cloud Code, Cloud Co-work, AWS, Azure, Salesforce, Agentforce, ServiceNow, Databricks, BigQuery, SAP, Model Context Protocol (MCP), Gartner, Mike Tyson, Arnold Schwarzenegger
## Video description
Jacob Schofield - Enterprise AI GTM, Google Cloud
While AI technology is advancing rapidly, many organizations still struggle to translate these innovations into meaningful, transformational business value. This session explores the common roadblocks stalling enterprise AI initiatives and introduces a practical framework for connecting AI capabilities directly to bottom-line results.
[Source](https://www.youtube.com/watch?v=0DCgPDwJ0R0)
Source quotes
[0:04] Jacob Schofield: 保持耐心，宽容对待自己，但必须带着紧迫感行动（be patient, give yourself some grace but act with urgency）。
[3:53] Jacob Schofield: 给员工买几个现成的工具并不等于进入了现代流水线时代。
[7:18] Jacob Schofield: 任何智能体（Agent）或基础模型的表现上限，完全取决于它能够访问的数据质量。
[10:24] Jacob Schofield: 优秀的销售永远不必跑去当面问客户 CEO 究竟因何事彻夜难眠（You should never have to go ask a CEO what keeps them up at night）。
[12:48] Jacob Schofield: 企业不再渴求按部就班执行固定流程的人，而是迫切需要能够自如驾驭 AI 智能体、在模糊性中生存（Thrive in Ambiguity）的通才。
[18:03] Jacob Schofield: 会场展示板上写着‘99% 的 AI POC 都会失败，但您的不会’，我不同意，你的 POC 一样会失败（I disagree. Yours will.）。
[21:45] Jacob Schofield: 我们消耗的 Token 费用已经超过了 2025 年整个数据中心的预算，到底发生了什么？我们是在盲目追求 Token 消耗上限吗？你们到底构建了什么，能配得上耗资 1000 万美元的 LLM Token？
[21:45] Jacob Schofield: 任何无法直接作用于营收增长（Top-line）或效率提升（Bottom-line）的 AI 举措，本质上都是一次‘假起跑’。
[23:59] Jacob Schofield: 直接挑战庞然大物就像在毫无准备的情况下去和迈克·泰森（Mike Tyson）打拳击一样，注定会遭遇惨败。
[25:48] Jacob Schofield: 不能轻信‘飞机座椅背后广告里承诺会有独角兽飞过来解决一切问题’，AI 的普及更像是从蒸汽机走向分布式电力的漫长工业演进。
Source quotes
[2:19] Jacob Schofield: second inning of the game
[2:19] Jacob Schofield: 10x mentality
[3:53] Jacob Schofield: productivity paradox
[10:24] Jacob Schofield: what keeps you up at night
[13:58] Jacob Schofield: magic wand
[18:03] Jacob Schofield: the sheet
[18:03] Jacob Schofield: You're never going to get to be Mr. Olympia and Arnold Schwarzenegger on stage by going to the gym for a week. You have to build a muscle.
[20:57] Jacob Schofield: not well-lit path
[21:45] Jacob Schofield: token maxing mentality
[23:59] Jacob Schofield: eating too much bread before the meal
[25:48] Jacob Schofield: 10x spark
Original transcript
[0:04] Good afternoon everybody. Welcome to the session.
Last one on the last day of the conference. So I'm
[0:09] fully impressed with uh the perseverance on this
thing. Um as mentioned my name is Jake Scoffield.
[0:16] I lead Google Cloud's enterprise AI go to market
team for the Western US and the semiconductor
[0:21] vertical. um been in AI for a long time and
just looking at the the folks that I'm sure
[0:27] you've listened to throughout the show, there's a
lot of brilliant minds. I do not consider myself
[0:32] a brilliant mind. Uh, but I do get the advantage
of working with a with a lot of customers every
[0:37] single day and I get firsthand exposure
and experience into how people are thinking
[0:42] about AI in their organizations, how they're
approaching it for their people, their products,
[0:47] their competitive landscape, um, and some of
the pitfalls and the challenges as we start to
[0:51] move through the pace and the velocity that AI is
changing the enterprise and changing our world. So
[0:57] um I'm going to talk a little bit about how
we see organizations adapting to AI a little
[1:01] bit about the history and what we can learn from
you know past times of innovation and then also
[1:07] um you know how Google's doing it internally cuz
we also get a view at like how are we approaching
[1:11] this problem. We're not unique. You know we do
have a lot of technology that powers some of
[1:15] these systems but everybody's going through the
same change. So um you know it's hard and I think
[1:21] the first thing that I I recognize when I talk
to leaders in the field is um there's a lot of
[1:26] noise that comes from the media and the industry.
Um you know some people say that there's going to
[1:32] be universal income in this utopian future. Some
people say that it's going to be a dystopian you
[1:36] know robots are going to take the world. Um
it's really hard to navigate the reality of
[1:41] where we're going. And if somebody told you that
they knew with confidence where we're going to be
[1:44] in 6 12 24 months they'd be lying. So whenever I
work with our customers and leaders in the field,
[1:51] uh it's important to give yourself a little
grace as as you're going through this process.
[1:56] It's hard. It's super confusing. Um and you
know, it's it's fun at the same time. You know,
[2:01] we we get the privilege of going through a major
shift in how technology is going to interact with
[2:06] our worlds and our people and our companies. So
my philosophy is be patient, give yourself some
[2:12] grace but act with urgency because this is also a
moment of disruption for many. Okay. Um one thing
[2:19] that we notice when we talk to executives is you
know the time savings with AI like hey the chatbot
[2:25] or the tool that can help me do a PowerPoint
or you know whatever productivity metric that
[2:30] you're driving towards is not enough and and it's
starting to get into the second inning of the
[2:35] game here in this AI world. And people are looking
for real transformational value out of AI. Great,
[2:42] I can write a doc. Great, I can ask a question
to an agent, it'll give me the answer. That's
[2:48] time savings metrics. But what folks are really
looking for is how they can use AI to transform
[2:54] their organization, be more competitive in
their in their market, drive more revenue,
[2:58] create more operational efficiency, and really
realize the the transformational value that that
[3:03] the the technology is capable of. So automation
alone is not like a significant value driver
[3:10] from a leadership perspective. Everybody's looking
for the spark that's going to do something really
[3:15] special that's going to 10x our revenue that's
going to automate all these processes that are
[3:20] 10 years old. And it's up to leaders to try to
figure out how to make that so for the board,
[3:24] the CEO, you know, it falls on a CTO, it falls
on a CIO, it falls on a chief product officer.
[3:30] there's this downward pressure to go figure out
what that spark's going to be that's going to
[3:34] completely reinvent our company that's going to
give us 10x mentality that's going to drive our
[3:38] revenue that's going to make us explode into a
unicorn and this automation narrative of oh we're
[3:43] going to deploy a chatbot is not really the thing
that's going to like be the transformational value
[3:47] that leaders are looking for. What we also see
when we look back to you know previous iterations
[3:53] of major technology shift is that there's a
big lag. You know the productivity paradox is
[4:00] a paper that was written in the late 90s and
it says that the there's the true value after
[4:04] technology innovation is very much so delayed.
So when this whole thing started in late 2023,
[4:11] early 2024, there was a philosophy and I'm sure
everybody saw it of like AI can do that. You know,
[4:17] we're just going to have AI do that. And 2
years later, you know, AI is doing a lot,
[4:22] but it hasn't necessarily realized the value of
what was anticipated when it first came out. So
[4:28] the expectations associated with the productivity
paradox is when there's major transformation in
[4:33] technology, we're going to see a linear graph of
value realized as this starts to roll out. What
[4:40] actually happened is there's a a quick ramp, you
know, the the automation piece and then there's a
[4:46] curve and we can see this present through time and
and I like the analogy of when steam central power
[4:53] for factories moved into electricity. You know,
this the steam engine you can see is ran most
[5:00] of the factories in the 1800s. There's a bunch
of gears and pulleys and shafts that drive the
[5:05] factory efficiency throughout the whole shop and
it was like a central power station. Electricity
[5:10] came along and they replaced the central
power station with an electric power station,
[5:15] but they still drove all the same gears and the
pulleys and the shafts that ultimately distribute
[5:21] energy throughout a factory floor. It took 30
years before they decentralized that power source
[5:27] and brought electricity to an actual workstation
to enabled the industrial revolution, the phase
[5:32] two of the industrial revolution of a real massive
manufacturing capabilities that came through. So
[5:38] I know we probably hear a lot of comparing AI to
mobile or comparing AI to internet. I think it's a
[5:44] really good thing to look back and see and and to
understand that it takes time for these things to
[5:49] really be implemented at scale and to deliver
the value that everybody's thinking that it's
[5:54] going to do. It doesn't happen overnight. Steam
went to electricity. It took 30 years before it
[5:59] was decentralized to really drive the value of the
modern assembly line. Um we also see that a lot of
[6:05] folks in that same mentality of oh AI will do it.
we'll just adopt AI, you know, we'll buy Copilot,
[6:11] we'll buy Gemini, we'll buy Cloud Code, we'll
buy Cloud Co-work, we'll we'll we'll just buy
[6:16] the tool. And just adopting AI is not going to
really get you to the assembly line. Adopting
[6:24] AI is going to build a muscle and I think it's
very important to instill that sense of urgency
[6:29] in teams, but it's not going to get to the point
of transformational value. And our customers are
[6:35] asking us, how can I really do this? you know, I
I want our CEO and our board is say 10x mentality,
[6:41] 50% operational efficiency, 10, you know, all of
these metrics that are kind of like pulled out of
[6:46] the ether. How do we do that? And how can I get
ready? How can I get my people ready? How can I
[6:52] get my tech stack ready? How can I get my data
ready to really harness this wave that's coming?
[6:58] Um, and it's really, really hard. Like, there's no
question. There's data fragmentation. And there's
[7:04] resistance from end employees. There's folks
that are real AI forward. There's folks that are
[7:08] really AI averse. It's a it's it's a really hard
problem to have. But everybody from a leadership
[7:13] standpoint is asking how do I do this for my
business and how can I get ready for this shift
[7:18] as it's coming. So at Google um they developed
a framework called the AI fluency framework.
[7:26] Um and and this is I'll tell you we're in the same
position as everybody else in trying to increase
[7:33] velocity to transformation recognizing disruption.
I mean there's no question you probably see in
[7:39] the news like chatt launches search is going to
torpedo like how do we navigate this as alphabet
[7:44] and as a business but more importantly how do we
make sure that our our people our organization our
[7:49] employees and the tools that we're using
are getting us ready for this shift. Um,
[7:55] everybody talks about it. I'm sure it's not a
new concept. The data foundation is the most
[8:01] important piece that you're going to go through.
If your data is not AI ready, you will be left
[8:07] behind and you'll be caught trying to make sure
that those core functions, an agent or a model
[8:14] is only as good as the data it has access to.
And almost every organization has hundreds of
[8:19] SAS applications. They have multiple data silos.
They have things in Amazon. They have things in
[8:24] Microsoft. They think have things in an on-remise
data center. The foundation for this to realize
[8:29] real transformation is the data. The second is
really around culture. You need to instill an AI
[8:37] first culture and you need to help people grow
literacy. Whenever there's a new paradigm and
[8:43] a new thing like people don't know how to use
it and and just giving somebody a tool doesn't
[8:47] necessarily help them reinvent how they work
dayto-day. So it starts with data. you need to
[8:53] reinforce a culture. And then the top three, which
we're going to go through, you know, in a little
[8:56] bit more depth, is the organizational design, how
you're helping line of business leaders and line
[9:02] of business employees figure out how to use this
for their own functions. It's the change in the
[9:07] employee profile. You know, we used to hire that
are very skills-based employees. You know, you're
[9:12] a SQL developer, you're a salesperson, you're a
finance person, you're an accountant. With AI,
[9:17] we're seeing a major shift into more general
capabilities. Like I can write SQL now because
[9:24] it's English. I can just ask it a question.
There's going to be a translation and all of a
[9:27] sudden I can talk to data. So that shift in the uh
employee profile is happening. And then what tools
[9:33] you're going to give and provide for people to
scale. And those aren't necessarily tools just for
[9:38] end users, you know, the co-pilots, the co-works,
the Geminis. It's tools that you're going to give
[9:42] your technology teams to really build scale and
run agentic services that interact with your data
[9:48] and interact with your applications at scale.
So I think this is where Google is coming from
[9:52] a fluency framework and we're taking a philosophy
within the organizational design realm of how can
[9:57] we automate processes, build more efficiency into
things that take a long time today. How can we
[10:04] augment those processes and build systems that
work in parallel with humans? And then how can
[10:09] we move that into a transformational play where
humans and AI interact in a hybrid environment
[10:13] fluidly all day long with accuracy with speed
with security with governance and ultimately
[10:19] that is the transformational piece when you have
AI right alongside people right alongside systems
[10:24] and data that operate autonomously throughout
processes. Okay, one example that's kind of close
[10:31] to my heart leading a go to market team. We want
to get more out of our salespeople. Like if you
[10:36] ask me what my one wish is, it's like I hope that
all my whole team sells 10 times more than what
[10:41] their number is. How can we use AI to help them
do that? You know, just having a tool that can
[10:48] automate email response or automate RFPs. That was
the first the first step. Hey, I want you to build
[10:53] an account plan. We can use AI that integrates
with Salesforce and our, you know, BHI tools and
[10:57] all these other things. Great automation piece.
Check. Then we said, how can we really augment a
[11:03] salesperson's effectiveness in their day-to-day
operations? That's where we started building
[11:08] agentic flows. How can you give an agent a task
and it's going to go and actually perform a review
[11:14] of your most strategic accounts, analyze their
10Ks, come with a perspective, put all of the
[11:20] videos of their CEO that's on YouTube into a, you
know, data store, and then get insights into the
[11:25] CEO. Um, and that's really where we got into this
custom value hypothesis and more of a consulting
[11:31] approach with heavy amounts of data that gets
integrated into a value narrative. So, we got
[11:37] that. We're really in that augment phase now. And
now we have many agents that are available for go
[11:43] to market teams so they can interact and they
can get better insights into their customers.
[11:47] They can get in better insights into the data
in their business. And that's where we're seeing
[11:51] this push into a transformational new mission. I
want my teams to be able to go to their customer,
[11:57] have intimate knowledge of their business,
understand exactly what they're trying to do,
[12:02] understand the vision of their executive teams
and their CEO and their board, understand their
[12:06] financials, and be able to architect that using a
suite of in of agents that come with a value and a
[12:13] hypothesis. You should never have to go ask a CEO
what keeps them up at night. It's on the internet.
[12:20] It's in their 10K. It's in their earnings report.
All of that information lives somewhere and AI
[12:25] and agentic workflows have a very good ability
of contextualizing large amounts of data. And
[12:31] with the right skills and instruction giving you
a perspective on how you should talk to somebody,
[12:35] whether that's a CFO, a CIO, a CTO, a VP of
product, you should be able to frame the person,
[12:41] the company, and come with a perspective of how
you can help them in your organization. The second
[12:48] pillar we talked a little about is the employee
profile. You know, we used to, as I mentioned,
[12:53] we used to hire SQL developers and we used to hire
data analysts and we used to hire salespeople that
[12:58] were like really good because they'd sold to the
enterprise before. We used to hire finance roles
[13:03] and all these other things with AI and agents.
Now, people can do so much more. I don't have to
[13:10] write SQL to talk to data and to get the insights
that I want. I just need to make sure that that
[13:15] agents mapped to the right data source that I
so I can get the right insights. I don't need to
[13:19] necessarily have very skill-specific capabilities
to go do a job the right way when I have AI and
[13:26] agents. So, we're seeing this generalist role with
high curiosity, ability to thrive in ambiguity,
[13:34] discovery mentality and specializations in a
certain field sort of take hold. So you can see
[13:40] the shift from very focused skill sets to more
broad curiosity, ability to change, ability to
[13:47] adapt and want to harness the capabilities that
are available to drive more value. Um big shift
[13:52] and I think everybody from a human resource and
a people operations standpoint is seeing that.
[13:58] um with tool enablement. I mean, this is actually
like my favorite topic because every single day I
[14:05] hear about a new tool and somebody in a startup
or somebody that works at Google or somebody
[14:10] that works at Microsoft or Amazon or whomever
is coming and saying, "I have a magic wand for
[14:15] you." You know, it's like unbelievable the amount
of uh noise and uh you know, influence that folks
[14:23] that are trying to make these decisions get. The
tool piece is super important, but I think the
[14:29] tools are consistently going to evolve. You know,
the new model's going to come out, great. The new
[14:34] widget, great. You should definitely be focused on
tools, but if you're a CIO or an IT administrator,
[14:39] you can have tool sprawl everywhere. And how
can you make sure that you're doing this with
[14:43] security in mind? How can you make sure you can
support these tools? You can't buy, you know,
[14:47] one little thing for one little solution and then
you have to maintain this for the duration. Um,
[14:54] so the tool piece is actually super important and
I think when we come and talk with our customers,
[14:59] we focus more on a platform capability. How can
you build to scale? How can you do this stuff
[15:05] without having bespoke solutions that live all
over your environment? How can you recognize that
[15:11] you need to accomplish something far greater
than one person's goal and one person's need
[15:16] for something that they saw on YouTube that they
want? Um, so really the utilization of tools is
[15:21] table stakes. How can you think about using tools
to eliminate bottlenecks in process? It starts to
[15:27] move up that value shift where you need to think
more about enterprise workflows, line of business,
[15:34] real problems that they have. Get out of real tool
orientation and map the tools to big problems. And
[15:41] then you get into what is the agentic enterprise.
Everybody's trying to think about this. You know,
[15:46] my marketing team, my HR team, my finance
team, my manufacturing operations. I work
[15:51] in semiconductors. They're designing chips of the
future. How can you know semi A beat semi B in the
[15:58] race towards the next generation chip? How can I
think about aggregation of massive data sets using
[16:05] LLMs and really advanced ways, training models,
have very complex agentic workflows that can do
[16:10] R&D faster? That's the real agentic enterprise.
I should have completely automated creative with
[16:16] hyperpersonalization for marketing. I should have
agents that can do people operations. I should
[16:21] have finance agents that can give insights
into supply chain and demand planning and
[16:25] forecasting and all of these things that live
in data, but it's really hard to unlock with
[16:30] the context available and with the right level of
intelligence. So we see that individual using tool
[16:36] great going to give you a certain boost you got
to think about how you're going to really unlock
[16:40] value for an entire organization and that's a
harder thing to solve for. Okay. Um we also see
[16:48] you know in that sort of utilization of tools
to breath of fully agentic enterprise the KPIs
[16:55] change over time when you're delivering a tool
someone's saying hey what are you doing okay I
[17:00] saved 20% of my overall time in doing a marketing
campaign as you move up that value paradigm you
[17:07] start to get operational KPIs in the breath phase
of real scale agentic where you're really making
[17:13] valuable business decisions at more speed. You're
understanding what's around the corner. You're
[17:19] aggregating huge amounts of data and you're
building agents that let you make decisions as
[17:23] leaders faster. Helps you be more competitive
and helps you remove operational burden and
[17:28] bottlenecks from the entire process. So there is
that sort of flow that we're going on. Everybody's
[17:34] in tool phase right now. I want this tool. I
want that tool. I need this tool. We're going
[17:39] to get into phase of let's really think about our
organization and decouple processes so that we
[17:45] can do things that have really big marginal gains
and then we're going to get into that phase three
[17:50] where it's we have agents running everywhere.
How do we operate these things? They interact
[17:54] with humans. They interact with our systems and
applications. They interact with our data and
[17:58] they do it in a very fluid way. Um that's sort
of where everybody's going and again super hard
[18:03] to get there. Okay, speaking of hard, um I'm sure
it for folks out there, does everybody have the
[18:10] sheet? Nobody has the sheet. Everybody that I talk
to has the sheet. I have a spreadsheet of 100 use
[18:17] cases. I've g I've gone out, I've met with all
of my teams and I have the sheet. And I'm like,
[18:24] great. You know, you made it to the sheet. That's
a hard part. um understanding like what really
[18:29] you can do for the business and and you end up
with this sheet of 100 use cases, 150 use cases.
[18:35] Some are hard and then you go to, you know, the
box high complexity, low low level of effort,
[18:43] low level, low uh low value, high complexity,
and then you say, great, we're going to take the
[18:47] Gartner magic quadrant box and we're going to do
low level of effort, high impact. And that's where
[18:52] we're going to start. Everybody's going through
that process. You're not alone. It's really hard.
[18:57] You'll get to a point where you can start to scale
these things. You can do it faster. I I saw on the
[19:02] presentation board, you know, 99% of AI PC's fail.
Yours won't. I disagree. Yours will. It's going
[19:11] to fail. There's going to be constant failure.
There's going to be constant struggle. There's
[19:15] going to be constant starts and stops and restarts
and figure it out again and repivot. Fail fast.
[19:21] And I really think that even though PC's fail,
it's necessary to go through that process. You're
[19:30] never going to get to the assembly line by just
waving the magic wand. You're never going to get
[19:35] to be Mr. Olympia and Arnold Schwarzenegger on
stage by going to the gym for a week. You have to
[19:41] build a muscle. You have to build a mental muscle.
You have to build a cultural muscle. You have to
[19:45] build a data foundation. You have to build a
muscle. So, I encourage partners that I work
[19:50] with to actually embrace that. Fail fast. Go for
it. Like, you're going to fail. PC's are going to
[19:55] flop. Things are going to change. A new model's
going to come out and they're probably going
[19:59] to do what you spent 6 months doing. Like, it is
what's going to happen. But if you don't fail fast
[20:05] and you don't build that muscle, you won't have
the the muscle built when ultimately everything
[20:10] comes together and you can be like, "Hey, you know
what? We had the sheet over the past 6 months. We
[20:16] solved half the sheet and you know what? There's
going to be a whole new sheet right after. Like
[20:21] it's it's an absolute evolution. What we're doing
right now in this sort of transformational journey
[20:26] that we're all on is a an un not well-lit
path. Nobody's done it before. You know,
[20:32] this is something that's very new. So having
that mentality of, hey, you know what? We're
[20:36] just going to be resilient and we're going to do
what we can and we're going to fail fast and we're
[20:39] going to run and go get it. You're building those
muscles to when the technology comes and you get
[20:44] the platform and you have the data, you've
gone through the process and you can start
[20:48] really scaling some of these solutions to meet
the mark. So the first part is hard. Everybody's
[20:53] got the sheet. The middle part is hard because
you got to build a muscle around this stuff and
[20:57] you have to figure out how to do it. And then
the end is really hard because you have to run
[21:00] scale. If this goes the way we think it's going
to go, there's going to be hundreds, thousands
[21:06] of agents that operate. How do you as technology
leaders or administrators or CIOS and CTO's manage
[21:13] that? And that's going to be a really hard part
that's that's really down the not well-lit path
[21:19] is we're going to have systems and we're going
to have tools and we're going to have agents
[21:22] and they're going to be running everywhere. And
then we're going to have a Salesforce agentforce
[21:26] agent that's running over here and we're going
to have a Service Now agent and we're going to
[21:29] have a data agent and data bricks or BigQuery and
then we're going to have this agent and like it's
[21:33] going to get overwhelming. So the last part about
how you can really provide enterprisegrade quality
[21:38] and scale is for you know technology leaders to
figure out too and it's really challenging. Okay.
[21:45] Um does any has anybody gotten the call from
the CFO yet? All right. So I I don't believe
[21:54] the non- hands up like these costs are exploding,
right? Like we're we're burning more tokens than
[22:00] our 2025 data center budget. What's going on
down there? Are are we just in the token maxing
[22:05] mentality? What did you guys build that actually
was worth $10 million in LLM tokens? You have to
[22:14] think about hitting the bottom line. You have
to focus on the P&L. I think the the pendulum
[22:20] is starting to swing back. You know, there was
unbridled speed orientation. We don't care what
[22:25] it costs. Just build fast. Your whole developer
team, 10,000 people, 100 people, whomever, build
[22:31] as fast as you can. Don't worry about the cost.
token maxing mentality, that pendulum is swinging
[22:37] back because it's not durable. You can't do it
forever. And if you don't see direct impact on
[22:42] the on the top line with revenue growth or impact
on the bottom line with more efficiency, then
[22:47] it's it's a it's a false start. So, I encourage
everybody when you're making the decisions about
[22:52] where to invest a in AI and you're looking at the
sheet and you have the box that says high value,
[22:57] low level of effort or whatever, map it to the
top line and map it to the bottom line because
[23:01] that conversation's going to come down from the
executive level and you're going to have to have
[23:05] a story to say this is the value that we drove
with the investment that we had and that's super
[23:10] opaque. So, I I empathize with folks that need
to have that conversation, but I can guarantee
[23:15] that it's it's here. If it's not here now, then
it's coming because we need to have durability
[23:18] and we need to have tangible ROI associated
to the investments. Okay, so in summary,
[23:26] build the culture. You know, you need to have your
workforce ready to take hold and to uh to be on
[23:31] the wave. You need to be able to fail fast. You
need to encourage that. It comes from the top up,
[23:36] but you don't want to stifle innovation and
creativity and and real uh transformative impact
[23:40] from the bottom up. Get your data ready. You need
to have that in a place where you can harness
[23:45] it. so you can realize the value when it comes
and then focus on durable value, focus on ROI,
[23:51] topline, bottom line, make the decisions
that matter. All right, thanks everybody.
[23:59] Um, so in a number of talks in the last three
days, I've heard people say that do not target the
[24:06] small use cases, which is an typical anti-attern,
right? We don't uh we don't incrementally build
[24:13] with AI but target the large scale use cases.
Sure. Which are higher cost and higher value.
[24:19] What's your perspective on that? Um I think you
need to keep the large use cases at the center
[24:24] of the table and make sure that that's the main
event and that you're not, you know, eating too
[24:30] much bread before the meal, I guess, for lack of
a better better analogy. um those large use cases
[24:37] have to be identified, but solving big problems is
hard. And if you're really going to try to go and
[24:45] um tackle the gorilla and make make and and think
about this in a major way, you still need to have
[24:51] the capabilities to and the muscle built. You
know, I I can want to go and and fight Mike Tyson,
[24:56] but it's like that's a loss leader. You know, I'm
not going to be able to do that. So, as much as
[25:00] you want to think about those big problems and
make sure you're defining them and keeping in
[25:06] the center of the frame, there's a ton of value
in building a muscle around smaller things. And
[25:12] you know, you build an agent to do certain
functions, you build multi- aent workflows,
[25:16] you build data connectivity in a secure way.
You can do that without exposing yourself to
[25:20] big risk. Because I think there's also, you know,
a downside to failing on a large use case and not
[25:26] having the value to show and the muscle that you
built in solving, you know, not easy problems,
[25:32] but less scary problems. Um, so I think a
little bit of both like don't lose sight of
[25:37] the large use cases. That's what's going to be
the transformational value, but give yourself
[25:41] some grace and build a muscle so that you can
solve that and and not fail. Oh uh I asked how in
[25:48] general we can get the data AI ready and I think
um in general that's a hard one to solve because
[25:57] in general everybody is different you know um
every customer that I work with has systems that
[26:06] are 25 years old like hey we're running on an SAP
stack that was from the '90s how do we get data
[26:13] ready do I need to modernize the application to
make it AI ready? Do I need to move the data in
[26:17] a central data warehouse to make it AI ready? And
it's a really hard one because there's no general
[26:23] answer. You know, in some situations, you need
to take the data and put it in a place that it's
[26:29] usable. In some situations, AI now can actually
reach into your data. You know, we have MCP that
[26:35] can integrate with APIs. We have intelligence and
all these things. Sure. How are you handling KPI
[26:42] conversations with senior leaders? cuz I think
what I'm seeing is they're seeing projections
[26:46] for AI solutions that are a little absurd in my
opinion. So, I'm trying to bring them back down to
[26:52] realistic expectations, but how are you handling
those conversations at the moment? Um, bring them
[26:58] back to reality, you know, like you read the
brochure on the back of the of the airplane
[27:04] seat that said there was going to be, you know, a
unicorn that would fly in and solve your problems.
[27:09] You got to think about it more tactically. I think
that's the biggest problem with leaders is is
[27:14] meeting the the expectations of like the board or
the expectations of a CEO on how fast they want to
[27:19] go or how much they like we should be doing 10x.
I've talked to CEOs and they're like I want the
[27:24] 10x spark. I'm like well yeah you know I do too.
Um but the reality is it's it's the steam engine.
[27:32] I mean the steam engine to the distributed power
like it takes time and you know having grounding
[27:40] people on you know the capabilities are here
there's no question and this is going to do some
[27:48] unbelievable stuff for humanity. It's absolutely
incredible. They're talking about solving disease.
[27:53] They're going to have abundance of crops cuz they
can understand the weather. Like it is here but
[28:00] it's it's not tomorrow. So, I think reinforcing
that message of we're doing it. We're leaned in,
[28:07] but there's not going to be I'm not going
to show up to you tomorrow with an answer
[28:12] to all of the problems. Like, we have to build
it. And I think that's the that's the biggest
[28:18] challenge is there's a perception. I think one
of those slide was like the expectation is this,
[28:23] the reality is this. like you got to kind of set
the right expectations but also give the necessary
[28:31] um visibility into the plan you know.
Thank you. That's super helpful. Yeah.
