---
格式版本: 2
标题: "The agentic era for mechanical design"
原文链接: "https://www.engineering.com/the-agentic-era-for-mechanical-design/"
发布日期: "2026-08-12"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:scrape:strict_html_metadata"
发布时间证据: "article:published_time: 2026-08-12T17:34:25+00:00"
发布时间校准原因: "规则确认唯一严格发布时间，来源 scrape:strict_html_metadata"
发布时间校准置信度: "high"
发布时间候选数量: 8
发布时间严格候选数量: 1
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-13T18:27:42+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-13T18:24:31+08:00"
入库时间: "2026-08-13T10:27:42.726Z"
来源平台: "Engineering.com 搜索"
搜索渠道: "source_template"
搜索词: "https://www.engineering.com/?s=AI"
匹配关键词:
  - "AI"
相关厂家:
  - "NVIDIA"
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 8
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "正文为Leo AI机械设计Agent介绍，仅提及Nvidia作为客户，与超节点/AI Rack/机柜级AI基础设施完全无关，无相关技术或商业信息。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-13T18:27:49+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 2
AI技术细节: 0
AI商业部署信号: 1
AI完整性: 0
AI摘要: "Leo AI发布Design Agent与Inspection Agent，CEO称其已通过“机械图灵测试”。该设计代理能根据需求生成DFM合规、可编辑的CAD装配体，并在过程中提问、计算和审批；"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:39:29.824Z"
采集批次: "2026年8月13日18点24分27秒"
采集批次ID: "20260813-182427-1564e572"
去重键: "https://www.engineering.com/the-agentic-era-for-mechanical-design"
---

Leo AI CEO Maor Farid thinks his design agent has passed the “mechanical Turing test.”

You’re reading Engineering Paper, and here’s the latest design and simulation software news.

I recently spoke with Maor Farid, co-founder and CEO of Leo AI, a mechanical engineering copilot that I’ve [covered before](https://www.engineering.com/leo-ai-can-now-generate-full-cad-assemblies/). Maor was eager to show off Leo’s latest capabilities, which, in his description, have finally delivered on “what sounded like bullshit and craziness” when the company launched in 2023.

That got my attention. Bullshit and craziness just happen to be my specialties.

“The threshold that is going to start the agentic era for mechanical design,” Farid told me, “is when a human can communicate a set of requirements, and then an AI agent not only designs it but asks leading questions, shows reasoning and decision making, and is able to retrieve and assemble parts in a way that’s aligned with industry standards.”

Leo, Farid believes, has now crossed that threshold.

He showed me a [video demo](https://www.youtube.com/watch?v=cG8h-t8D0zs) of the updated Leo, which started with a sophisticated prompt: “Design a coaxial contra-rotating propeller drive for a marine environment.”

In the video, Leo creates a spec document and asks the user to pick a type of transmission configuration: bevel gear differential drive, epicyclic (planetary) gearbox, or other. Leo then runs through a series of calculations, updates the spec document, and asks for various approvals before creating a 3D assembly using parts it creates or pulls from the user’s inventory.

The result, in Farid’s words, is “a full-blown DFM-compliant manufacturing-ready CAD assembly” that passes what Farid has dubbed the mechanical Turing test: “you won’t be able to distinguish that it’s an AI that did it.”

![](https://www.engineering.com/wp-content/uploads/2026/08/img_leo_3dassembly.png)

Screenshot from the Leo demo video showing Leo Design Agent. (Image: Leo AI.)

That assembly can then be transferred to the user’s CAD program of choice. “Everything is in B-Rep, everything is editable, so it’s not a mesh,” Farid said. Leo’s website lists integrations for Autodesk (the video demo showed Inventor), Siemens (it does not specify which software), PTC Creo, Onshape, Windchill, Solidworks, and Catia.

That whole shebang is called Leo Design Agent. The video then went on to demo another agent, Leo Inspection Agent, which “runs through your entire CAD assembly and its drawings to catch errors before they reach manufacturing: DFM issues, misalignment with industry standards, internal guidelines and project requirements, opportunities for part reuse,” per the video’s narrator.

![](https://www.engineering.com/wp-content/uploads/2026/08/img_leo_inspector.png)

Screenshot from the Leo demo video showing Leo Inspection Agent. (Image: Leo AI.)

This is all intriguing stuff. A cynic might be put off by the fast-moving, pre-recorded demo, and the fact that these capabilities are far beyond those of [other AI engineering tools](https://www.engineering.com/backflips-back-is-the-mesh-to-cad-ai-real-this-time/) of which said cynic has been [similarly cynical](https://www.engineering.com/autodesks-neural-cad-looks-good-on-paper/).

On the flip side, at least one reader [has told me](https://www.engineering.com/is-autodesks-neural-cad-worth-getting-excited-about/) that Leo is genuinely helpful, and Farid says it’s being used by heavyweight customers including HP, Nvidia, and the U.S. military.

“All of those customers, they’re testifying that they’re doing stuff in minutes that used to take months,” Farid told me. He also pointed out that Leo placed #1 on G2’s [Best New Software Products 2026](https://www.g2.com/best-software-companies/rookies) rankings and #12 on its [Best AI Software Products 2026](https://www.g2.com/best-software-companies/top-ai), for whatever that’s worth.

So has a true threshold in mechanical engineering really been crossed? I’m not in any hurry to pay back investors, so I prefer to wait and see before making such claims. I’d love to hear your thoughts either way, which you can mail electronically to [\[email protected\]](https://www.engineering.com/cdn-cgi/l/email-protection#e88589848a89a8899a9a879f8e8491c68b8785).

## Quick hits

- Simulation developer VCollab has introduced [Rapid Results Viewer](https://vcollab.com/news/vcollab-delivers-the-engineering-intelligence-layer-for-industrial-digital-twins-using-nvidia-technology/), or RRR, which it describes as “the engineering intelligence layer for simulation, helping engineers understand results, identify risks, capture knowledge, and make confident decisions.” RRR is built on Nvidia’s agentic AI infrastructure and is solver-neutral, according to VCollab.
- Remcom has updated its [XFdtd electromagnetic simulation software](https://www.remcom.com/press-releases/remcom-announces-updated-version-of-xfdtd-with-expanded-circuit-solver-and-project-organization-tools) with new component types, more material support, support for component tagging and more.

## One last link

Brian Buntz, Editor-in-Chief of R&D World, explores AI perspectives in [Where AI pioneers Hinton, Li and Ng converge and diverge regarding AI cybersecurity, jobs and education](https://www.rdworldonline.com/where-ai-pioneers-hinton-li-and-ng-converge-and-diverge-regarding-ai-cybersecurity-jobs-and-education/).

Got news, tips, comments, or complaints? Reach me at [\[email protected\]](https://www.engineering.com/cdn-cgi/l/email-protection#c7aaa6aba5a687a6b5b5a8b0a1abbee9a4a8aa).