AI Agents in 3D Design Automation

AI design tools are moving beyond visuals. Autodesk’s Mike Haley has set out how neural CAD foundation models could change design workflows by reasoning in 3D, physics and editable CAD geometry. The important distinction is not just faster generation. It is whether AI can produce usable design data, not a flat image or rough concept model. For AEC and manufacturing, that could shift AI from visual ideation into earlier-stage geometry, constraint handling and performance insight. https://s.cad.onl/neural #CAD #AEC #AI #DigitalDesign

Agentcad Enables Text-to-CAD Generation via MCP and CLI Support 📌 Revolutionize your engineering workflow with agentcad, a new open-source tool that empowers AI agents to transform plain text descriptions into precise 3D CAD models. By integrating via MCP and CLI, this framework enables autonomous agents to design, render, and validate complex geometries for manufacturing. It bridges the gap between AI coding and physical engineering, allowing agents to self-correct topological errors through a high-fidelity feedback loop. 🔗 Read more: https://lnkd.in/draZmEWu #Agentcad #Modelcontextprotocol #Texttocad #Geometricmodeling #Commandlineinterface

I built an AI-assisted parametric 3D modeling platform from scratch and it's live. As an MSc Mathematics student specializing in differential geometry, I kept asking: why can't you just *say* what you want a 3D model to be? "Make it 3 units wide." "Add a cable routing hole." "Round the edges." So I built MAKER AI a browser-based parametric CAD tool where natural language drives real B-rep geometry via an AI tool use loop: → OpenCascade.js (WASM) running fully client-side → n8n webhook as AI proxy layer → DFM self-verification: AI detects illegal geometry, auto-corrects, explains → Every NL instruction stored as a spec history commit → One-tap "Send to Farm" for 3D print dispatch 🔗 Live demo: https://lnkd.in/eciKckzc 💻 GitHub: https://lnkd.in/eXrN2Ze3 Built as part of GNI LABS / FOFUS | 3Dprinting, 3Dscanning and 3Ddesigning an AI + physical manufacturing platform for the maker economy. #MathematicsInTech #3DModeling #AIEngineering #ComputationalGeometry #WebAssembly #BuildInPublic

🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 AI 3D tools are moving from image-like concepts toward editable, production-aware geometry. 🧩 Orca is interesting because it frames AI design around real CAD models, manufacturable output, and assembly-aware details. For exhibitions and pop-ups, that shift matters. A booth idea cannot stop at a pretty visual. It has to become dimensions, joints, material logic, and fabrication review. The stronger workflow is faster handoff from concept to buildable decision.

Anyone else experimenting with AI in their CAD workflows lately? From generative design to automated 2D drafting tools, the speed of iteration is getting wild. Are you using AI as a design co-pilot yet, or are you still holding out for traditional modeling? 🛠️🚀 #MechanicalDesign #CADSoftware #FutureOfWork #ProductDesign

Last week, I had an energizing deep dive with Michael Rabinovich into the reality of AI‑driven CAD. Michael’s new CADGenBench (created with Thomas Wolf and with Elie Cuevas of Mecado) is the first open benchmark that asks a simple, sobering question: can frontier language models turn technical drawings into production‑ready parts? The answer: not yet. The benchmark includes 81 tasks drawn from real engineering projects. Models must both generate complete 3D STEP files from drawings and make precise edits to existing parts. A strict validity gate checks whether the geometry is watertight and meshable; only then do models get scored on shape accuracy, topology (how many holes or voids you have), and whether the part would even bolt together on the right fixture. Even the top models (GPT‑5.5 Pro and Claude Opus 4.7) scored around 0.38 out of 1.0 and produced invalid solids 10–12 % of the time. Other models fared worse. Generation tasks were especially challenging, while editing tasks were only marginally better. The takeaway: today’s LLMs approximate the outline of a part but miss critical features and mating surfaces. This is not a “maybe”—the data is clear. For anyone in manufacturing or design, this benchmark is a wake‑up call. AI is encroaching on the CAD world, but we’re still miles from replacing skilled engineers. We’re going to need specialised geometry reasoning, constraint solvers and domain‑specific models to close the gap. Kudos to Michael for pushing the field forward and for letting me peek behind the curtain! What do you think—is your industry ready to trust generative CAD? Here's the benchmark: https://lnkd.in/eV6yB6dS DM me for a more detailed analysis and breakdown! #BetterCallFino #DesignIntelligence #AIAcrossTheProductLifecycle #HuggingFace

The very first AI Agent for 3D is here! I am honored to have participated in the project's testing. The Meshy AI Agent empowers users with enhanced style-consistency capabilities, enabling a complete workflow that allows for the one-click generation of multiple sets of images and 3D models via conversational prompts—followed by one-click downloading or 3D printing. Link to try: https://www.meshy.ai

Carl Bass Calls It "The Spellchecker for Manufacturing". He compared it to spellcheck: quietly but significantly raising the floor for every engineer who uses it. Physical AI is not a chatbot on a CAD system. It is AI that reasons about real-world constraints such as stress, vibration, CNC tool paths, draft angles, and material behavior and generates geometry that is optimized for performance and ready for the shop floor. When he saw what InfinitForm's Physical AI platform does, bringing performance requirements, manufacturing constraints, material properties, and machine capabilities into one intelligent system that generates production-ready geometry, his reaction was immediate: "To the extent that you can embed standard knowledge... you get to a basic level that's way above." Watch Carl's full assessment: https://hubs.la/Q04k0j9z0 #PhysicalAI #Engineering #Manufacturing #CNC #CAD

We are excited to share the third demo in our SOLIDWORKS AI series! In this video, CEO, Manish Kumar shows how our virtual assistant, Leo, helps engineers quickly analyze error messages and pinpoint the root cause in complex models. See how AI is making troubleshooting faster and smarter for design teams. Watch the demo to see Leo in action! https://lnkd.in/eiQseT6c

We are excited to share the third demo in our SOLIDWORKS AI series! In this video, CEO, Manish Kumar shows how our virtual assistant, Leo, helps engineers quickly analyze error messages and pinpoint the root cause in complex models. See how AI is making troubleshooting faster and smarter for design teams. Watch the demo to see Leo in action! http://go.3ds.com/nN9N