Ford Cuts Vehicle Aerodynamic Testing to 10 Seconds Using AI Models
Ai4
Integrating AI across vehicle lifecycles allows Ford to compress 15-hour wind tunnel simulations to 10 seconds while matching traditional physics results within a 2.3% margin of error.
## Background
Published on YouTube by the Ai4 channel on August 24, 2025, this presentation features Franziska Bell, Chief Data, AI, and Analytics Officer at Ford Motor Company. According to the description, Bell outlines how Ford strategically embeds AI as a digital partner for its workforce across the entire product lifecycle, from initial creative concept sketches to final manufacturing quality checks.
## TL;DR
Franziska Bell details Ford Motor Company's strategy of pairing workforce talent with AI across vehicle design, engineering, manufacturing, and customer service. Generative AI accelerates design by translating sketches and parametric prompts into 3D models, while proprietary AI simulation models cut aerodynamic physics runtimes from 15 hours down to 10 seconds within a 2.3% error margin. AI also reviews over 120,000 internal engineering specification documents, automates customer support queries with de-identified feedback loops, and powers computer vision quality inspections across more than 800 workstations globally. Throughout all initiatives, human-in-the-loop governance and customer data privacy ensure high quality, responsible deployment, and amplified employee productivity.
## Atmosphere & style
The presentation carries a polished, authoritative, and pragmatic corporate keynote tone. Bell speaks with steady executive confidence, structuring the session around clear, real-world engineering workflows rather than abstract hype. Watching the talk feels like an inside look at large-scale industrial modernization, where rigorous technical benchmarks and concrete cycle-time reductions underscore an optimistic vision of human-AI collaboration.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) Ford integrates AI as an augmenting digital partner rather than a replacement, establishing human-machine teams across the entire product lifecycle.
- [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) Responsible AI deployment at Ford requires strict data privacy, explicit customer consent, and mandatory human-in-the-loop validation to safeguard against hallucinations.
- [04:12](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=252s) Generative AI bridges the gap between design and engineering by instantly converting concept sketches into hundreds of 2D renders and 3D models.
- [04:12](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=252s) Designers transition into creative directors by leveraging text prompts and parametric controls to quickly explore and iterate on vehicle component concepts.
- [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) Proprietary AI physics models compress complex aerodynamic drag simulations from 15 hours down to approximately 10 seconds.
- [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) AI aerodynamic simulations maintain tight alignment with traditional physics-based models, achieving an average error precision within 2.3%.
- [11:34](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=694s) An AI reading assistant analyzes over 120,000 internal engineering requirement documents to eliminate ambiguities while domain experts retain final authority.
- [12:57](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=777s) A unified conversational AI customer support agent resolves complex queries across disparate manuals by detecting intent and executing calculation tools.
- [12:57](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=777s) De-identified customer interaction summaries create continuous feedback loops that alert quality control and content teams to emerging vehicle issues.
- [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) Computer vision models deployed across more than 800 workstations in North American plants have conducted over 200 million automated visual inspections.
- [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) AI-powered 'Code Reviews as a Service' across 600 software engineers achieved a 3x improvement in review cycle times.
## Detailed walkthrough
Here is a detailed, faithful walkthrough based on the transcript provided:
[00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) AI Strategy, Data Privacy, and Human-in-the-Loop Philosophy
Franziska Bell, Chief Data, AI, and Analytics Officer at Ford Motor Company, opens her presentation by explaining how Ford is strategically integrating artificial intelligence across its entire product lifecycle as a digital partner to its workforce. Rather than viewing AI as an isolated technology, Ford views it as a core capability designed to form collaborative "human-machine teams" that elevate collective human potential. Grounded in Ford’s North Star vision—building a better world where every person is free to move and pursue their dreams—all AI initiatives are built upon robust foundational platforms, infrastructure, and data products.
Data privacy and trustworthy AI are central pillars of this strategy. Customer data is strictly gated and integrated into AI applications only with explicit consent. Furthermore, Ford adheres strongly to a "human-in-the-loop" approach to mitigate AI risks such as hallucinations and inaccuracies. Human oversight ensures essential judgment, validation, and adherence to Ford’s high quality standards. By establishing tight feedback loops, Ford continuously refines its AI systems to remain effective and responsible, freeing employees from manual tasks so they can focus on high-value strategic thinking, creative problem-solving, and customer relationships.
- Customer data inclusion in AI requires explicit consent.
- Human oversight acts as a safeguard against AI hallucinations and errors.
- Strategic objective: Freeing up designers, engineers, and support teams to innovate faster with less friction.
---
[00:412] Accelerating Design with Generative AI and 3D Modeling
The walkthrough of the product lifecycle begins with the design phase. Traditionally, vehicle design started with a napkin or paper sketch that was manually sculpted into a time-consuming 3D clay model. With AI, Ford designers can now transform a single initial concept sketch into hundreds of high-fidelity 2D renderings in seconds—a task that previously took hours. Crucially, AI now generates 3D models directly from 2D images, creating an immediate and seamless bridge between design and engineering.
Designers utilize text-based prompts and parametric tools across exterior, interior, and component design (such as wheels). Bell demonstrates a custom tool where designers do not even need an input sketch. By entering a prompt—such as "a professional photo of a modern wheel for an off-road focused pickup truck"—and adjusting specific parameters (wheel diameter, spoke count, spoke width, spoke angle, lug nut count), the system generates multiple real wheel concepts almost instantly. This elevates designers from manual modelers to creative directors who guide AI partners to explore possibilities previously too time-consuming to consider.
```mermaid
flowchart LR
accTitle: AI Design Workflow
A["Text Prompt & Parameters"] --> B["AI Generative Tool"]
C["2D Concept Sketch"] --> B
B --> D["2D High-Fidelity Renderings"]
D --> E["3D Model Generation"]
E --> F["Engineering Integration"]
```
- Parameter controls include spoke count, spoke angle, spoke width, wheel diameter, and lug nut count.
- Shift in role: Designers transition from manual modelers to creative directors.
---
[06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) High-Speed Engineering Physics Simulations and Accuracy Calibration
Once a 3D model is created, it moves into engineering and physical testing. Traditional physics-based simulations, such as finite element analysis (FEA) for aerodynamic drag or structural stress analysis, pose severe computational bottlenecks. High-fidelity aerodynamic models require a granular mesh of approximately 14 million cells, demanding memory that exceeds a single GPU's capacity and requiring large GPU clusters. Consequently, a single traditional aerodynamic drag simulation takes around 15 hours to complete.
To overcome this, Ford developed proprietary AI models that run virtual wind tunnel simulations in just 10 seconds—representing a dramatic acceleration compared to the traditional 15-hour runtime. To address accuracy and precision, Ford's engineering teams specifically focused on eliminating errors in critical directional deviation quadrants (where AI predictions directionally oppose traditional physics outputs). On average, Ford's AI simulation outputs align within 2.3% of traditional physics-based results, allowing designers and engineers to iterate fluidly together.
```mermaid
flowchart TD
accTitle: Physics Simulation Error Alignment Quadrants
A["Simulation Data Evaluation"] --> B{"Quadrant Alignment"}
B -- "Directionally Aligned" --> C["Top-Right & Bottom-Left: Acceptable"]
B -- "Substantial Deviation (Red)" --> D["Top-Left & Bottom-Right: Target for Optimization"]
D --> E["AI Model Fine-Tuning"]
E --> F["2.3% Average Precision Alignment"]
```
- Traditional aerodynamic mesh scale: ~14 million cells per vehicle model.
- Time reduction: Traditional 15-hour simulation reduced to 10 seconds with AI.
- Accuracy: AI simulation predictions remain within ~2.3% of traditional physics-based simulation outputs.
---
[11:34](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=694s) AI Assistance in Engineering Documentation and Requirements
Beyond physical simulations, AI supports engineering documentation and compliance. Ford manages approximately 120,000 internal engineering requirement documents. Maintaining absolute clarity across these documents is vital for downstream vehicle build quality and safety.
To solve this, Ford developed an AI-based reading assistant that reviews all 120,000 specification documents to identify ambiguities and highlight opportunities for enhanced clarity. The AI surfaces these suggestions directly to domain-expert engineers, who evaluate the outputs, make final decisions, and refine the specifications, leading to measurable improvements in engineering productivity and documentation quality.
- Scope: Over 120,000 engineering specification documents analyzed.
- Process: AI surfaces clarity improvements; domain-expert engineers maintain final decision authority.
---
[12:57](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=777s) Customer Support Knowledge Agent and Feedback Loops
When vehicles reach end-users, customer information has historically been fragmented across service manuals, towing guides, warranty papers, and service articles. To eliminate the burden on customers of searching across multiple disparate portals, Ford built an intelligent conversational AI agent to act as a unified front door.
When a customer submits a query, the AI agent assesses user intent, formulates a query plan, searches internal knowledge bases, and calls special tools (e.g., towing calculators) to compute exact answers before delivering a single, synthesized, referenced response. Crucially, customer interactions undergo an automated pipeline where personally identifiable information (PII) is removed for privacy. The conversations are then summarized and categorized to surface emerging issues to quality control teams and identify knowledge gaps for content creators.
```mermaid
flowchart LR
accTitle: Customer Support AI Pipeline
A["Customer Query"] --> B["Intent & Search Planning"]
B --> C["Knowledge Base & Tool Query"]
C --> D["Synthesized Reference Response"]
A --> E["De-identification / PII Removal"]
E --> F["AI Summarization & Categorization"]
F --> G["Quality Control & Knowledge Teams"]
```
- Functional scope: Unified access across user manuals, warranty documents, service articles, and towing guides.
- Capabilities: Intent detection, dynamic tool calling/calculations, source citation.
- Strategic value: Privacy-preserving (de-identified) feedback loops for immediate quality issue identification.
---
[15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) Manufacturing Quality Inspection, Code Reviews, and Closing Vision
Bell concludes by highlighting additional scaled human-AI implementations across Ford:
- **Manufacturing Defects:** Computer vision algorithms have been deployed across more than 800 workstations and dozens of manufacturing plants globally in North America, completing over 200 million automated visual inspections to detect defects.
- **Software Engineering:** Ford launched "Code Reviews as a Service," supercharging human reviewers with AI. Initial results across 600 software engineers demonstrated a 3x improvement in cycle times.
In closing, Bell reiterates that across the entire journey—from napkin sketch to real-world customer support—AI at Ford operates not as a replacement, but as an augmenting partner. By raising team capabilities, reducing simulation wait times from 15 hours to seconds, and turning designers into creative directors, Ford accelerates its overall product lifecycle and its capacity to deliver innovative mobility solutions at enterprise scale.
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) building a better world where every person is free to move and pursue their dreams — Franziska Bell
> [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) human-machine teams — Franziska Bell
> [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) human-in-the-loop — Franziska Bell
> [06:52](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=412s) a professional photo of a modern wheel for an off-road focused pickup truck — Franziska Bell
> [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) Code Reviews as a Service — Franziska Bell
## Mentioned
Franziska Bell, Ford Motor Company, Ai4, North America
## Video description
Franziska Bell, Chief Data, AI and Analytics Officer, Ford Motor Company
Abstract
In this talk, I will share how AI is helping accelerate the product lifecycle at Ford. At Ford, we are strategically embedding AI as a powerful digital partner for our workforce. I will guide you thorugh some parts of our product lifecycle to show you this human-AI partnership in action, from the first creative sketch to the final quality check.
Franziska Bell, Chief Data, AI, and Analytics Officer at Ford Motor Company, outlines how the automaker integrates artificial intelligence across its entire vehicle lifecycle as an augmenting partner to human teams rather than an autonomous replacement. Supported by human-in-the-loop oversight and strict data privacy, the strategy aims to eliminate engineering bottlenecks and accelerate product development.
Human-in-the-Loop Governance and Data Privacy (0:04)
Ford treats artificial intelligence as a core capability built on foundational data platforms and digital infrastructure rather than an isolated technology. To mitigate risks such as hallucinations and model inaccuracies, the company enforces a strict human-in-the-loop philosophy, requiring human judgment and validation before outputs are finalized. Customer data integration into AI workflows is gated strictly by explicit user consent.
According to Bell, establishing tight feedback loops ensures that AI systems adhere to company quality standards while freeing designers, engineers, and support staff from manual tasks. The primary organizational goal is to redirect employee effort toward higher-value problem solving, strategic thinking, and customer relationships.
Generative Design and Instant 3D Prototyping
In the vehicle styling phase, generative AI is replacing manual sketching and clay sculpting with automated digital iteration. Ford designers can convert a single 2D concept sketch into hundreds of high-fidelity renderings within seconds, as well as generate 3D models directly from 2D imagery to streamline handoffs to engineering teams.
Bell highlighted specialized parametric tools that allow designers to synthesize functional component concepts without preliminary drawings. By inputting text prompts and adjusting explicit constraints—such as wheel diameter, spoke count, spoke width, spoke angle, and lug nut count—designers transition from manual modelers into creative directors who evaluate multiple generated variants rapidly.
High-Speed Physics Simulations and Requirements Engineering (6:56)
Traditional engineering physics simulations present severe computational bottlenecks; for example, high-fidelity aerodynamic drag models require granular meshes of roughly 14 million cells, demanding large GPU clusters and up to 15 hours per run. Ford developed proprietary AI surrogate models that complete virtual wind tunnel simulations in 10 seconds. Engineering teams calibrated these models to eliminate opposing directional errors, achieving outputs that align within an average of 2.3% of traditional physics-based results.
Beyond computational physics, Ford applies AI to internal documentation. To maintain build quality and safety standards across approximately 120,000 internal engineering specification documents, an AI reading assistant reviews texts for ambiguities and surfaces suggested clarifications. Domain-expert engineers evaluate every recommendation and retain sole authority over final document changes.
Operational Automation Across Manufacturing, Code, and Support (12:57)
At later lifecycle stages, Ford deploys specialized AI systems to assist production, software delivery, and customer care. A conversational customer support agent acts as a unified portal across user manuals, towing guides, and warranty documents, executing tool-based calculations and supplying source citations. Customer interactions are automatically stripped of personally identifiable information (PII), summarized, and routed to quality control teams to flag emerging product defects.
In manufacturing operations, computer vision systems across more than 800 workstations in North American plants have conducted over 200 million automated defect inspections. For software engineering, Ford introduced an AI-assisted code review service across 600 developers, which Bell reports improved code review cycle times by 3x.
Ford's enterprise deployment demonstrates that substantial operational gains—such as cutting aerodynamic simulation times from 15 hours to 10 seconds—depend heavily on keeping human domain experts at the center of the workflow. The program emphasizes that AI acts as an accelerator, relying on ongoing expert validation to correct model drift, eliminate errors, and maintain vehicle safety standards.
精炼阅读摘要
Ford Motor Company（福特汽车）首席数据、人工智能与分析官 Franziska Bell 阐述了将人工智能作为企业级数字化协同伙伴的战略，通过在整车全生命周期中构建“人机团队”并辅以严格的人工监督与隐私保护机制，来加速创新并优化运作效率。
人机协同架构与数据隐私底线 (0:04)
Franziska Bell 表示，Ford 并非将 AI 视为孤立技术，而是将其定位为贯穿产品生命周期的核心协作能力，旨在将员工从重复的体力劳动中解放出来，专注于高价值战略规划、创新及客户关系。这一战略依托底层数据基础设施，将 AI 定义为辅助人类的协同伙伴，而非替代力量。
在数据安全与风险治理方面，Ford 严格推行“人在回路”（Human-in-the-Loop）机制，利用人工监督作为防范 AI 模型幻觉与输出偏差的防线。同时，系统在涉及客户数据时设立了严格的权限关卡，只有在获得客户明确许可授权后，相关数据才会被接入 AI 应用程序中进行处理。
生成式设计与三维建模重塑
在传统汽车设计流程中，概念往往从草图起步并耗费大量时间手工雕刻三维油泥模型。引入生成式 AI 后，设计师能够将一张概念草图在数秒内扩展为数百幅高保真二维渲染图，并进一步直接由二维图像生成三维模型，打通了设计与工程团队之间的连接鸿沟。
Bell 演示了针对汽车外饰、内饰及零部件定制的参数化工具。设计师无需绘制草图，只需输入提示词（如越野皮卡的现代轮毂样式）并调节轮毂直径、辐条数量、辐条宽度、辐条角度及螺母孔数等参数，系统便能瞬间生成多套概念方案。这促使设计师的角色由传统的手工作图人员转型为指导 AI 探索方案的创意总监。
工程高保真仿真加速与规范审核 (6:56)
传统空气动力学风阻模拟等有限元分析（FEA）依赖约 1400 万个网格单元，计算量超出单张 GPU 承载上限，往往需要耗费约 15 小时的集群运行时间。Ford 研发的专用 AI 仿真模型将虚拟风洞测试时间缩短至 10 秒。工程团队重点消除了预测方向与传统物理规律相悖的误差象限，使 AI 的平均输出结果与传统物理模拟的偏差控制在约 2.3% 之内，实现了设计与工程的高频流体迭代。
在工程合规与管理方面，Ford 针对内部约 12 万份工程规范文档部署了 AI 辅助阅读工具。该系统专门扫描文本中模糊不清的表述并提出清晰度优化建议，由具备专业知识的工程师进行最终裁定与修订，从而降低下游车辆试制中的潜在质量与安全风险。
智能售后支持与闭环质量反馈 (12:57)
为了改变以往客户信息分散在保养手册、拖拽指南、保修单据与维修文档中的局面，Ford 构建了集成化的对话式 AI 统一入口。当用户提问时，该代理会自动评估用户意图、规划检索路径、查询内部知识库并调用专用工具（如拖拽负荷计算器）给出附带参考来源的精确答复。
在数据回流链路上，系统通过自动化管道滤除所有个人身份信息（PII）。随后，脱敏后的对话记录会被 AI 自动分类与提炼，向质量控制团队实时报告新出现的潜在缺陷，同时协助内容团队识别当前知识库的覆盖盲区，形成持续优化的服务闭环。
工厂质检规模化与软件研发赋能 (15:48)
在规模化制造端，Ford 已在北美数十家整车与零部件制造厂的 800 多个工位上部署了计算机视觉算法，累计完成逾 2 亿次自动化视觉缺陷检测，提升了生产线质量把控的稳定性。
在软件开发领域，Ford 推出了“代码审查即服务”（Code Reviews as a Service）平台。Bell 指出，在首批试点的 600 名软件工程师团队中，AI 辅助审查使得开发周期的运转速度提升至原来的 3 倍，展示了人机协同在研发交付层面的提效成果。
Franziska Bell 总结认为，从概念设计、物理仿真、软件工程到售后维护，AI 在 Ford 的核心价值在于赋能专业人员并压缩生产周期。不过，企业级规模化部署仍高度依赖严格的数据脱敏许可、持续的模型对齐校准以及领域专家在关键决策节点上的人工裁定与最终把关。
Detailed notes — zh
## Background
该视频由 Ai4 频道于 2025 年 8 月 24 日发布在 YouTube 平台上。内容为福特汽车公司首席数据、AI 与分析官 Franziska Bell 的主题演讲，重点阐述了福特如何将 AI 战略性地作为数字伙伴嵌入汽车产品全生命周期，展示从最初创意草图到最终制造质检的人机协同落地实践。
## TL;DR
演讲系统介绍了福特汽车在企业级规模构建“人机协同团队”的落地成果与核心战略。在设计端，生成式 AI 实现了从 2D 草图到数百种高保真渲染图及 3D 资产的秒级生成；在工程端，自研物理 AI 代理模型将长达 15 小时的整车气动 CFD 仿真大幅压缩至 10 秒，且平均误差控制在 2.3% 以内。此外，福特还深度部署了工程标准审查助手、全渠道客服智能体，并在工厂车间实现超 2 亿次计算机视觉质检及代码审查提效，全面赋能员工聚焦高价值创新。
## Atmosphere & style
演讲整体洋溢着专业、务实且兼具前瞻视野的工业领袖气质。主讲人通过详实严谨的工程数据对比与具体的产线案例，将复杂的企业数字化转型解构得极为清晰透彻。整场分享节奏干练平稳，摒弃了虚泛的概念炒作，让观众在严谨沉稳的叙事中深刻感受到前沿 AI 深度赋能百年传统车企所带来的产业变革力量。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) 福特将 AI 定位为员工的“数字伙伴”（digital partner），致力于在全企业规模构建人机协同团队（Human-AI Teams），加速产品全生命周期创新。
- [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) 确立严格的数据隐私与显式授权原则，并坚持 Human-in-the-Loop（人类在环）机制，将人类专业监督作为防范 AI 幻觉和确保工程质量的核心底线。
- [04:13](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=253s) 在概念设计阶段，生成式 AI 可在数秒内将 2D 手绘草图转化为数百种高保真渲染图，并快速生成 3D 几何模型，无缝衔接造型与下游工程研发。
- [04:13](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=253s) 通过免草图的参数化 Prompt 工具，设计师能够直接指定尺寸、轮辐及孔位等几何参数生成方案，角色升级为探索创意的“设计总监”。
- [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) 自研物理 AI 代理模型，将原本需要大规模 GPU 集群运算 15 小时的整车虚拟风洞 CFD 仿真压缩至 10 秒，实现计算速度的量级突破。
- [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) AI 气动代理模型将平均预测误差控制在 2.3% 以内，并重点消除几何增量趋势（Deltas）的象限误差，支持工程团队在会议中实时调整设计。
- [11:34](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=694s) 利用 AI 助手遍历并审核约 12 万份内部工程规范文档，自动识别歧义并提出改进建议，由领域专家把关以大幅提升工程标准质量。
- [11:34](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=694s) 构建统一客户服务 AI Agent 整合手册与指南并动态调用计算工具，结合去标识化和摘要管道反哺知识库并捕捉潜在车辆质量问题。
- [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) 计算机视觉缺陷检测算法已落地全球数十家工厂的 800 多个工位，累计完成超过 2 亿次零部件与整车质检。
- [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) 在 600 名软件工程师中试点 AI 辅助代码评审（Code Review as a Service），将代码审查交付周期缩短至原先的三分之一。
- [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) AI 深度赋能让设计师摆脱繁琐建模、让工程师免于漫长仿真排队，释放全员精力专注于高价值创新与战略突破。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) 福特的人机协作愿景与 AI 核心战略
福特汽车公司首席数据、AI 与分析官 Franziska Bell 阐述了 AI 如何作为强大的“数字伙伴”（digital partner）战略性地嵌入福特的产品全生命周期。福特不仅仅是 AI 浪潮的旁观者，更在积极塑造其产业影响，目标是在全组织内构建人机协同团队（Human-AI Teams），全面释放集体的创新潜能。
为支撑这一愿景，福特持续重金投入底层基础设施平台与数据产品开发。福特的战略北极星是“建设一个更美好的世界，让每个人都能自由出行并追求梦想”，所有 AI 项目均严格对齐该目标。
```mermaid
flowchart LR
accTitle: Ford AI Strategy Framework
subgraph Foundation["底层支撑与原则"]
A["数据基座与基础设施"]
B["严格数据隐私 & 显式授权"]
C["Human-in-the-Loop 机制"]
end
subgraph Core["核心赋能"]
D["AI 数字伙伴"]
end
subgraph Outcome["业务赋能价值"]
E["加速全生命周期创新"]
F["提升员工创造力与效率"]
end
Foundation --> Core --> Outcome
```
- **数据隐私原则**：客户数据仅在获得明确且显式许可（explicit consent）的前提下才会输入 AI 系统。
- **人机协同（Human-in-the-Loop）**：福特坚持人类监督是防范 AI 幻觉（AI hallucinations）与确保高质量输出的核心底线，通过构建闭环反馈机制持续迭代模型。
---
### [04:13](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=253s) 设计阶段：从餐巾纸草图到秒级 3D 建模
在传统汽车设计流程中，设计师通常从餐巾纸上的草图起步，逐步打磨并手工制作出全尺寸实体油泥模型（clay model）。该流程虽然成熟经典，但极为耗时。福特引入生成式 AI 后，设计师可在数秒内将单张二维草图转化为数百种高保真 2D 渲染图，而以往生成单张方案就需要数小时。
更为关键的突破在于三维生成能力：AI 可以直接从 2D 图像快速构建出 3D 几何模型，无缝打通了造型设计与后续工程研发环节。
```mermaid
flowchart TD
accTitle: AI-Driven Design Workflow
subgraph Input["设计输入阶段"]
A1["手绘概念草图"] --> B["2D 高保真渲染图 (秒级生成数百种)"]
A2["文本 Prompt + 几何参数"] --> B
end
subgraph Modeling["3D 资产生成"]
B --> C["AI 快速构建 3D 模型"]
end
subgraph EngineeringBridge["下游工程对接"]
C --> D["CAE 仿真与工程分析接口"]
end
```
在轮毂、外观及内饰设计中，福特开发了免草图的参数化提示词工具。设计师只需输入诸如“越野皮卡现代轮毂的专业摄影图”，并设定轮毂直径、轮辐数量、轮辐宽度/角度以及螺栓孔数（lug nuts），点击生成后即可获得多种可用设计方案。这使得设计师的角色从单纯的“手工建模师”升格为指挥 AI 进行多维探索的“创意总监”。
---
### [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) 工程与仿真：AI 驱动的秒级空气动力学预测
当设计进入物理工程验证阶段时，传统有限元分析（FEA）与高精度计算流体力学（CFD）仿真成为了研发周期的主要瓶颈。传统仿真需要将整车划分为约 1400 万个微细计算网格（14 million cells），单次整车气动阻力计算需要消耗超出单块 GPU 显存的算力，必须依托大型 GPU 集群连续运行约 15 个小时。
福特自主研发了专用物理 AI 代理模型，将原本需要 15 小时的虚拟风洞仿真压缩到了仅仅 10 秒钟，实现了计算速度的量级飞跃。
| 评估维度 | 传统物理仿真 (CFD/FEA) | 福特自研 AI 代理模型 |
|:--- |:--- |:--- |
| **单次计算耗时** | 约 15 小时 | 约 10 秒钟 |
| **算力与显存要求** | 需大规模多 GPU 集群运算 | 本地/轻量级 GPU 即可运行 |
| **相对物理仿真误差** | 基准（Ground Truth） | 平均误差控制在 2.3% 以内 |
| **核心误差控制指标** | 绝对数值精度 | 几何增量趋势（Deltas）一致性 |
在模型精度验证中，福特尤其注重几何构型变化量（deltas）的方向一致性。工程团队重点消除了预测值与真实物理值趋势相悖的象限误差（即坐标图中出现反向判断的高危区间），保证了在 2.3% 极低平均误差下，工程团队与设计师能够在设计会议中实时调整并验证车身气动特性。
---
### [11:34](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=694s) 工程规范审核与全渠道客户服务智能体
除了物理流场仿真，福特在工程文档管理和客户服务环节也深度部署了 AI 助手：
- **12 万份工程标准智能审核**：福特内部沉淀了约 120,000 份工程需求文档。AI 助手能够遍历全部文档，识别歧义并提出清晰度改进建议，最终由资深领域专家工程师把关决策，显著提升了工程规范的质量与生产力。
- **统一客户服务智能体（AI Agent）**：针对以往用户手册、拖拽指南（towing guides）、维修指南与质保文件分散在各个独立系统的痛点，福特构建了智能前端入口。该智能体能够精准识别用户意图，调用底层工具链（如拖拽能力计算工具）动态计算并生成精准答复。
```mermaid
flowchart TD
accTitle: Customer Service Feedback Loop
UserQuery["用户多渠道提问"] --> Agent["统一 AI Agent 意图识别 & 检索计算"]
Agent --> Answer["精准结构化响应与参考出处"]
Answer --> DeID["去标识化管道 (隐私安全脱敏)"]
DeID --> Summary["AI 智能归类与摘要管道"]
Summary --> HumanTeam["人工团队补充知识库盲区"]
Summary --> Quality["质量团队快速捕捉潜在车辆缺陷"]
```
---
### [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) 制造质检、软件研发及全链路赋能总结
AI 数字伙伴的协同模式正在福特制造与软件研发等多个环节规模化落地：
- **计算机视觉智能质检**：缺陷检测视觉算法已部署至北美及全球数十家工厂的 800 多个工位，累计完成超过 2 亿次零部件与整车质检（200 million inspections）。
- **代码审查即服务（Code Review as a Service）**：在 600 名软件工程师中试点 AI 辅助代码评审，将代码审查的交付周期缩短到了原先的三分之一（3x cycle time improvement）。
Franziska Bell 总结指出，从初期的设计手绘草图到终端用户的用车疑问，AI 正在彻底重塑福特的人才效能：设计师不再受困于繁复的手工建模，工程师不必在长达 15 小时的仿真排队中被动等待，全体员工得以专注于高价值的战略构想与创新突破，共同加速推进移动出行的未来。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) 建设一个更美好的世界，让每个人都能自由出行并追求梦想 — Franziska Bell
> [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) 客户数据仅在获得明确且显式许可的前提下才会输入 AI 系统 — Franziska Bell
> [00:04](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=4s) 人类监督是防范 AI 幻觉与确保高质量输出的核心底线 — Franziska Bell
> [04:13](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=253s) 设计师可在数秒内将单张二维草图转化为数百种高保真 2D 渲染图，而以往生成单张方案就需要数小时 — Franziska Bell
> [04:13](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=253s) 设计师的角色从单纯的‘手工建模师’升格为指挥 AI 进行多维探索的‘创意总监’ — Franziska Bell
> [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) 将原本需要 15 小时的虚拟风洞仿真压缩到了仅仅 10 秒钟，实现了计算速度的量级飞跃 — Franziska Bell
> [06:56](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=416s) 重点消除了预测值与真实物理值趋势相悖的象限误差，保证了在 2.3% 极低平均误差下，工程团队与设计师能够在设计会议中实时调整并验证车身气动特性 — Franziska Bell
> [11:34](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=694s) AI 助手能够遍历全部文档，识别歧义并提出清晰度改进建议，最终由资深领域专家工程师把关决策，显著提升了工程规范的质量与生产力 — Franziska Bell
> [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) 在 600 名软件工程师中试点 AI 辅助代码评审，将代码审查的交付周期缩短到了原先的三分之一 — Franziska Bell
> [15:48](https://www.youtube.com/watch?v=tKE7j1p6bEY&t=948s) 设计师不再受困于繁复的手工建模，工程师不必在长达 15 小时的仿真排队中被动等待，全体员工得以专注于高价值的战略构想与创新突破 — Franziska Bell
## Mentioned
Franziska Bell, 福特汽车公司 (Ford Motor Company), Ai4, CFD (计算流体力学), FEA (有限元分析), CAE 仿真, 物理 AI 代理模型, 计算机视觉智能质检系统, 代码审查即服务 (Code Review as a Service)
## Video description
Franziska Bell, Chief Data, AI and Analytics Officer, Ford Motor Company
Abstract
In this talk, I will share how AI is helping accelerate the product lifecycle at Ford. At Ford, we are strategically embedding AI as a powerful digital partner for our workforce. I will guide you thorugh some parts of our product lifecycle to show you this human-AI partnership in action, from the first creative sketch to the final quality check.
[Source](https://www.youtube.com/watch?v=tKE7j1p6bEY)
[0:04] Franziska Bell: building a better world where every person is free to move and pursue their dreams
[0:04] Franziska Bell: human-machine teams
[0:04] Franziska Bell: human-in-the-loop
[6:52] Franziska Bell: a professional photo of a modern wheel for an off-road focused pickup truck
[15:48] Franziska Bell: Code Reviews as a Service
Original transcript
[0:04] Thank you to the organizers for having
[0:05] me here today. I'm very excited to be
[0:08] sharing with you how we are leveraging
[0:10] artificial intelligence at Ford to
[0:13] accelerate the product life cycle. At
[0:16] Ford, we are strategically embedding
[0:18] artificial intelligence as a powerful
[0:21] digital partner to our workforce. Um,
[0:25] and uh, today I'm going to guide you
[0:28] through different parts of our product
[0:30] life cycle and show you this human AI
[0:33] partnership in action. From the first
[0:36] creative sketch to the final quality
[0:39] check and you'll see firsthand how this
[0:42] human AI collaboration is helping make
[0:45] our teams faster, more insightful, um,
[0:48] as well as more effective as we build
[0:50] the future of mobility.
[0:53] Now the landscape of artificial
[0:55] intelligence is very rapidly evolving
[0:58] and at Ford we are not just observing
[1:01] we're actively shaping its impact. We
[1:04] envision a future where humans and AI
[1:07] work collaboratively together and form
[1:10] human machine teams thereby rising our
[1:13] collective potential and driving
[1:15] innovation across the entire
[1:18] organization.
[1:19] Now to realize this vision, we continue
[1:22] to heavily invest in the corresponding
[1:24] infrastructure platforms and data
[1:27] products that underpin and power these
[1:30] AI applications.
[1:32] Um, our northstar at Ford is to build a
[1:36] better world where every person is free
[1:38] to move and to pursue their dreams. And
[1:42] everything we do, including our AI
[1:44] applications, is aligned with this
[1:47] purpose.
[1:49] And one of the key principles at Ford
[1:52] that AI is heavily leaning on here is
[1:55] data privacy. We take data privacy at
[1:58] Ford very seriously. For example, any
[2:01] data that we get from our customers is
[2:04] only included in AI systems and
[2:06] applications if we have very clear uh
[2:10] and explicit consent. Our commitment to
[2:13] trustworthy artificial intelligence also
[2:16] means acknowledging its limitations. And
[2:19] this is why we're so huge proponents of
[2:22] the human in the loop approach. Human
[2:26] oversight provides the essential um
[2:29] guidance, validation, and judgment that
[2:32] is really incredibly important in order
[2:35] to make these systems work to ensure
[2:37] that we safeguard against inaccuracies
[2:40] from artificial intelligence, for
[2:42] example, AI hallucinations, and ensure
[2:45] that the output meets Ford's high
[2:48] standards. By building robust feedback
[2:52] loops, we empower our teams to
[2:54] continuously refine these EI systems uh
[2:58] and to ensure that they're operating
[3:00] effectively as well as responsibly.
[3:04] Now, at uh Ford, artificial intelligence
[3:07] is not just a silo technology. It is
[3:10] very much a core capability that we're
[3:13] deeply integrating across every single
[3:16] part of our enterprise. Our philosophy
[3:20] here is that EI's largest value is
[3:22] unlocked if we deploy it as a strategic
[3:26] digital partner to our talented
[3:28] workforce.
[3:30] [snorts]
[3:31] To illustrate this transformation that
[3:33] we're undergoing at Ford, I'm going to
[3:35] show you today how we're building human
[3:38] AI teams at scale. This collaboration
[3:42] helps our engineers, designers, and
[3:45] support experts to be freed up and to
[3:48] leverage their invaluable skills and
[3:50] capability to do what they do best,
[3:53] strategic thinking, creative problem
[3:56] solving, and building relationships with
[3:58] our customers. Now the result of this
[4:01] human AI team collaboration is that we
[4:04] can innovate faster, reduce friction um
[4:08] as well as ultimately deliver better
[4:11] products. So let's look at some concrete
[4:13] examples. Uh the first step in our um
[4:17] product life cycle is design where
[4:19] brilliant ideas take shape.
[4:22] Now, historically, design started off
[4:25] with a creative sketch, sometimes on a
[4:27] napkin, which is then transformed into a
[4:30] tangible 3D clay model. This is a very
[4:33] much proven, but also very timeconsuming
[4:36] process. Now, thanks to artificial
[4:39] intelligence, we're radically
[4:41] accelerating this creative cycle. Our
[4:44] designers uh can now transform a single
[4:47] sketch into hundreds of variations of
[4:51] highfidelity 2D renderings and doing
[4:54] that within seconds. A task this
[4:56] previously would have taken hours to
[4:59] complete. But the real gamecher is in
[5:02] the third dimension as we make 3D
[5:04] products. Thanks to artificial
[5:06] intelligence, we can now build 3D models
[5:10] directly from these 2D uh images. And
[5:14] this really allows us now to form an
[5:16] immediate and seamless bridge to our
[5:19] engineering processes.
[5:24] Now whether it is the exterior, the
[5:26] interior, or intricate wheel designs,
[5:29] our teams now have access to textbased
[5:32] prompts and parameter tools that allow
[5:35] them explore numerous uh possibilities
[5:38] and iterate on them extremely rapidly.
[5:42] And here for example is a tool that
[5:44] we've built that doesn't even require a
[5:46] sketch as an input. The user just can
[5:49] start with various different parameters
[5:52] and a textbased prompt as you can see
[5:54] here. For example, a professional photo
[5:57] of a modern wheel for an off-road focus
[6:00] pickup truck. And then the user can
[6:03] specify additional parameters. for
[6:06] example, the wheel diameter, the number
[6:08] of spokes, the width of the spokes, the
[6:11] angle of the spokes, and the number of
[6:14] lug nuts. And once the user has
[6:17] completed specifying the parameters, all
[6:20] they need to do is click generate. And
[6:22] within a short amount of time, various
[6:25] real designs emerge.
[6:28] But this is not just about making things
[6:30] faster. This is about completely
[6:32] rethinking the creative process and
[6:35] elevating our designers from manual
[6:38] modelers to creative directors that now
[6:41] guide their artificial intelligence
[6:43] partners to iterate and generate new
[6:47] possibilities and options that
[6:49] previously would have been far too time
[6:51] consuming to even consider.
[6:56] Now once a design is realized in three
[6:59] dimensions we are now shifting our focus
[7:01] to the rigorous demands of engineering
[7:03] and testing. This is where innovation
[7:06] meets physical reality. Now a primary
[7:10] challenge on this front has always been
[7:12] how timeconsuming physics-based
[7:14] simulations have been. For example, if
[7:17] you want to do a stress analysis on a
[7:20] single wheel design, this is a very
[7:22] complex and timeconsuming bottleneck
[7:26] using traditional methods.
[7:28] In the bottom right hand corner, you can
[7:30] see here an example of a stress analysis
[7:33] uh of a wheel. Historically, we have
[7:36] used both physical um simulations as
[7:40] well as then eventually built a physical
[7:43] prototype for testing. [snorts] And
[7:46] these physics-based simulation often use
[7:49] finite element analysis.
[7:51] Now here we have started to leverage
[7:54] artificial intelligence to dramatically
[7:57] increase uh the pace at which we can do
[8:00] engineering and testing. This results in
[8:03] a new collaborative approach where both
[8:06] engineers and designers aided by
[8:08] artificial intelligence now come
[8:10] together and complete uh the development
[8:14] cycle with unprecedented speed.
[8:18] Now double clicking here the primary
[8:20] challenge of these physics-based uh
[8:22] traditional simulations is their immense
[8:25] computational scale. In order to get
[8:28] high fidelity results, we need to model
[8:31] a vehicle in a very fine granular mesh
[8:34] of around 14 million cells. And such a
[8:38] massive model and simulation can provide
[8:41] significant hurdles as it comes to the
[8:44] computation. For example, um this $14
[8:48] million 14 million cell model uh
[8:52] requires more memory than can be found
[8:55] in a single GPU, requiring us to run
[8:59] these simulations in very large clusters
[9:02] of powerful GPUs.
[9:06] And given how computationally expensive
[9:08] these computations are, running a single
[9:12] aerodynamic drag calculation on a
[9:14] vehicle could take around 15 hours to
[9:17] complete. But this is where artificial
[9:20] intelligence can be a pivotal force for
[9:22] innovation. We have developed
[9:25] proprietary artificial intelligence
[9:27] models at Ford that provide these
[9:31] complex computations now at
[9:34] unprecedented speed and precision.
[9:37] Now speaking about precision, another
[9:40] challenge that we're facing is the
[9:42] precision that's required uh for these
[9:44] simulations. Of course, we also insist
[9:47] on very high accuracy in addition to
[9:50] high precision. But as you'll see
[9:52] momentarily on this slide u getting the
[9:55] difference um in uh or deltas between
[9:58] geometries is much more important. So
[10:02] let's look at this chart here. On the x
[10:05] axis you can see the actuals from the
[10:07] physicsbased simulation and on the yax
[10:10] axis the predicted outputs from the AI
[10:13] model. errors that fall in the bottom
[10:16] right uh top right hand side or bottom
[10:19] left hand side quadrant are less of an
[10:21] issue um as the actuals and predicted
[10:24] values are directionally aligned.
[10:27] However, if the results fall into the
[10:30] top leftand side or bottom right hand
[10:33] side quadrants as indicated here in red,
[10:36] this indicates that the AI models
[10:38] deviate substantially from the
[10:40] physics-based approach, therefore posing
[10:43] a significant challenge. As a result,
[10:46] our teams have focused predominantly on
[10:48] minimizing the errors in these two right
[10:52] hand two uh red quadrants.
[10:55] And thanks to the fantastic work of our
[10:58] teams, we can now run virtual wind
[11:01] tunnus simulations in mere seconds, 10
[11:04] seconds to be precise, instead of 15
[11:08] hours with traditional meth methods. Now
[11:11] moreover as you can see on the right
[11:13] hand side the output of the AI based
[11:16] models align very well with the physics
[11:20] based uh simulations and the AI results
[11:24] are within around 2.3%
[11:27] of the physics-based outputs on average.
[11:34] Now, the principle of augmentation and
[11:37] human AI teams doesn't stop with
[11:39] physical simulations. This is just one
[11:42] example. And here's another one in the
[11:44] engineering and testing stages.
[11:47] We have around 120,000 documents at Ford
[11:51] that specify requirements for the
[11:54] engineering world. Now, of course, this
[11:56] means that clarity is paramount. To help
[11:59] with this, we have developed an AI based
[12:03] assistant uh that helps our engineers.
[12:06] It's uh reviews all of these 120,000
[12:09] documents and I identifies um any
[12:12] opportunities for increasing clarif
[12:15] clarity. It then surfaces these
[12:18] opportunities to our engineers. Um this
[12:21] has greatly improved both quality as
[12:24] well as productivity of our works uh of
[12:26] our engineers output.
[12:29] Now as always with human in the loop
[12:31] systems having a deep domain expert in
[12:34] the loop is absolutely critical. So in
[12:37] this case our engineers with their deep
[12:39] domain knowledge are the ones who are
[12:41] reviewing uh and interpreting the EI's
[12:44] output. They apply their expertise and
[12:47] they make the final strategic decisions
[12:50] and choices to ensure worldclass
[12:52] products are are built.
[12:57] Now moving on in our uh life cycle
[13:00] journey. Now we get to our end users.
[13:03] Our commitment to our customers extend
[13:06] extends throughout their usership and
[13:08] ownership journey. But historically
[13:11] finding information could have been a
[13:14] challenge. Answers were often siloed
[13:17] across various different sites as well
[13:19] as knowledge bases. Um and this meant
[13:22] that is a very fragmented experience.
[13:25] anything from user manuals uh to towing
[13:28] guides or for example service articles
[13:31] as well as warranty documents. This put
[13:34] the burden on our customers to stitch
[13:37] this information together, navigate a
[13:40] complex web of sources in order to get a
[13:42] complete answer.
[13:46] To solve this, we have built an AI agent
[13:49] that is now a single and intelligent
[13:52] front door uh to our customers.
[13:55] Now, when a customer comes and asks a
[13:57] question, this AI agent does the heavy
[14:00] lift and complex work behind the scenes.
[14:03] It first identifies the user's intent.
[14:06] Um, it then builds a plan to query the
[14:10] most relevant sources and it then
[14:12] executes this search across our
[14:15] knowledge base.
[14:17] This now allows us to have a much more
[14:20] seamless and simple experience for our
[14:23] customers, allowing to them to get
[14:25] answers very quickly and very
[14:28] accurately. And importantly here is that
[14:31] this AI agent also has access to
[14:34] relevant tools. for example, towing
[14:36] guides. And so it can go beyond just
[14:39] retrieving text to actually dynamically
[14:42] calculating responses for the user and
[14:44] then synthesizing that into a single and
[14:47] referenced response.
[14:50] But the true strategic value here
[14:52] actually lies in the feedback loop that
[14:55] this generates.
[14:58] Any customer engagement that we have, we
[15:00] then run through an automatic
[15:02] intelligence pipeline. We deidentify any
[15:06] personal information for privacy and
[15:08] security and then have an AI based
[15:11] summarization capability to immediately
[15:14] categorize
[15:15] issues as well as questions. We then
[15:19] surface common questions and issues to
[15:21] our human teams who can then investigate
[15:25] and fill in any gaps in our knowledge
[15:27] bases to continuously refine and improve
[15:30] the answers our customers are getting.
[15:33] But this data based approach is also
[15:36] invaluable feedback mechanism for our
[15:38] quality teams who can identify any
[15:41] emerging vehicle issues much faster than
[15:45] before.
[15:47] [snorts]
[15:48] Now if we continue to look across our
[15:51] product life cycle, there are many many
[15:54] more human AI teams that you would find
[15:57] that have significant value at Ford. Let
[16:01] me give you two further examples. One is
[16:03] in our manufacturing space. Here we have
[16:06] rolled out computer vision algorithms in
[16:09] in order to spot defects.
[16:12] Um this has now been rolled out at scale
[16:14] to over 800 workspaces um in North
[16:17] America as well as dozens of plants
[16:20] across the globe and have completed over
[16:23] 200 million inspections. And in software
[16:27] engineering, we recently rolled out code
[16:29] reviews as a service. This is a
[16:31] capability where we are supercharging
[16:34] human code reviewers with artificial
[16:37] intelligence and initial results across
[16:39] 600 software engineers showed that cycle
[16:43] times improved by a whopping threefold.
[16:47] Now there's many more examples like this
[16:50] across Ford. But I think one thing is
[16:52] clear as we have been journeying from
[16:54] the first creative sketch to the final
[16:58] question of our customer. A pattern has
[17:01] emerged at Ford. Artificial intelligence
[17:04] is augmenting our workforce.
[17:08] We have seen designers um who have been
[17:11] elevated from manual modelers to
[17:14] creative directors leveraging their
[17:16] artificial intelligence partners to make
[17:20] possibilities happen that otherwise
[17:22] would be unreachable.
[17:24] We've also seen our engineers move from
[17:27] passive observers waiting for 15-hour
[17:30] simulations to finish to be able to
[17:33] optimize and iterate and solve problems
[17:36] within mere seconds.
[17:40] But this is not just about making
[17:42] processes faster. It is also about
[17:45] making them smarter and more creative.
[17:49] And tying this back to our Northstar
[17:51] vision, by building these powerful human
[17:54] AI teams, we're not just accelerating
[17:57] the product life cycle. We're
[17:59] accelerating our ability to innovate at
[18:02] scale, to solve harder problems, and to
[18:06] build a better world where every person
[18:09] uh is free to move and to pursue their
[18:11] dreams. Thank you very much. [applause]
