Agentic CAD and Chip-Design AI: How Engineering Teams Can Apply These Tools

For organizations that need to put AI into concrete technical workflows, the useful question is not whether a model is impressive, but where it enters the design process, what human review it requires, and what kind of output it produces. A new cluster of releases centered on CADFit and agentic electronic design automation (EDA) tools shows several practical patterns: conversion from scan data to editable models, agent-assisted planning before schematic work, agent-assisted execution during chip design, automated board checks, and domain-specific model portfolios that act through existing engineering software.

CADFit is the most directly hands-on example. Engineering Paper has previously tracked this mesh-to-CAD direction, which is comparable to Backflip AI and Autodesk Fusion’s AutoTimeline. Here, however, the project is open-source and can be used freely for non-commercial work. It is led by MIT PhD student Ghadi Nehme. The early research was funded by Altair, now part of Siemens, and Nehme and Siemens have applied for a patent on the CADFit method.

Where CADFit Fits in an Engineering Workflow

In practical terms, CADFit is intended to bridge imported mesh data and parametric CAD editing. Nehme’s paper frames the goal as a “hybrid optimization-based CAD reconstruction framework that recovers complex, editable CAD construction sequences from meshes by incrementally fitting and validating parametric operations using geometric feedback.” The value proposition is not a static rendering; it is a sequence of operations that an engineer can inspect, revise, and reuse.

“[CADFit] basically starts with the mesh and then extracts some key sketches, similar to how engineers would extract them,” Nehme said. From those sketches, the method creates solids and then consolidates them with unions and cuts. It compares the result with the source mesh and repeats the loop to fill gaps. In a video narrated by Nehme, the workflow is shown step by step. A simple machine-learning model helps decide which sketches matter. According to Nehme, that model is not required for the method to work, but it speeds the process by a factor of three.

What CADFit Can Do Today

The current operation set is best understood as a coverage boundary. Nehme explains: “Currently, the operations that CADFit supports are extrusions, revolutions, fillets, and chamfers, and then also union and cut and all the sketch operations.” For very complex geometry, the limits are clearer: “But if a shape is very complex, like loft-based or also sweep-based shapes… the math will struggle.” He is developing a new CADFit version for those shape families.

The generated CAD models are written in CadQuery, an open-source Python library. Nehme also created Studio CAD, a viewer he describes as supporting “human friendly editing of the shapes.” There are not yet plug-ins for mainstream CAD applications, but a Siemens NX integration is in progress. On the roadmap, Nehme says: “The future iteration of this work will potentially be able to be really robust on a very huge set of shapes,” while noting: “But this one is already pushing the boundary by a huge amount… This method is able to really extract the key components of the geometry much better than what’s been seen before.”

A Step-by-Step Way to Pilot CADFit

1. Select a non-commercial benchmark: choose a mesh with clear extruded or revolved features rather than a heavily organic or swept model.
2. Run the basic reconstruction flow: let CADFit extract sketches, build solids, apply unions and cuts, and compare the output with the original mesh.
3. Check operation coverage: confirm the model depends only on supported extrusions, revolutions, fillets, chamfers, sketch operations, unions, and cuts.
4. Use the CadQuery output as a review artifact: treat the Python-generated construction sequence as editable engineering data, not merely a display model.
5. Inspect geometry through Studio CAD before importing elsewhere: use the viewer for human-friendly editing and validation of shape features.
6. Plan integration carefully: since no plug-ins exist yet, document how results will move into existing tools; Siemens NX work may change that path later.

Agentic EDA: Four Application Positions

Tool	Application position	What it changes	What still needs human control
Celus Design Platform	Pre-schematic planning in electronics design	Celus launched a beta agentic workflow described by the company as “helps hardware engineers plan the architecture and component work at the start of an electronics design, before schematic capture, and carries out that plan once the engineer approves it,” the company says.	Approval before execution; architecture choices remain an engineering decision.
ChipAgents	Digital and analog chip-design execution	ChipAgents says it is expanding its agentic engineering platform across both domains. The company states: “ChipAgents’ full-stack approach enables engineers to define objectives, constraints and critical judgment calls while AI agents carry more of the execution forward, reasoning over design context, operating engineering tools, evaluating results and determining the next action toward closure,” the company says.	Objectives, constraints, and critical judgment calls stay with engineers.
Circuit Mind AVE	PCB schematic review	Circuit Mind introduced AVE, the Agentic Verification Engine. The company calls it an “AI application that checks Printed Circuit Board (PCB) schematics against component datasheets, application notes, and each organization’s own design review checklists, catching errors that slip past even careful manual review before they turn into costly board respins.”	Organization-specific checklists and final acceptance still require review.
Cadence Maestro	Chip and system design tool operation	Cadence released Maestro, a portfolio of domain-specific AI models for agentic chip and system design. Cadence CEO Anirudh Devgan said: “Combining our Maestro portfolio of domain-specialized models with leading frontier and open-weight models enables our AI Super Agents to drive Cadence EDA and SDA tools with exceptional skill and efficiency to transform chip and system design,” Cadence says.	Tool access, governance, and design sign-off remain organizational concerns.

A recurring pattern across these announcements is not a single “AI button,” but a more distributed handoff: humans set intent, constraints, and acceptance criteria; agents operate software tools or verification processes within those boundaries.

Common Misconceptions

• “Open-source means plug-and-play.” CADFit’s open-source status lowers trial friction for non-commercial use, but it does not yet provide plug-ins for common CAD software; CadQuery output and Studio CAD inspection still need workflow design.
• “Agentic equals fully autonomous.” The Celus, ChipAgents, Circuit Mind, and Cadence announcements all point to a model in which engineers or design rules still define objectives, constraints, approvals, or checklists.
• “If the output is editable, it is complete.” CADFit supports key parametric operations today, but loft-based or sweep-based complexity can still stress the method.
• “Faster extraction equals immediate production use.” The simple ML sketch-identification step speeds the process by a factor of three, but that benefit may be most meaningful in pilot testing before it justifies production integration.

Action Advice

• For product-design teams: start with a small CADFit pilot on mesh-to-parametric cases that use supported operations, and capture failure modes before expecting a plug-in path.
• For electronics and semiconductor teams: map each agentic tool to the exact place in the flow—Celus before schematic capture, ChipAgents during digital/analog execution, Circuit Mind AVE during schematic verification, and Cadence Maestro when operating EDA/SDA tools.
• For governance and risk owners: define approval points in advance. The announcements repeatedly separate human judgment—objectives, constraints, approvals, design-review checklists—from machine execution.
• For infrastructure owners: verify IP status and commercial-use terms. CADFit is open-source for non-commercial use, while the method is patent-filed by Nehme and Siemens.

Related Context

If a team wants a broader framing for why engineering AI can feel different from other enterprise AI, a guest article by Luminary CEO Pete Schlampp on Engineering.com distinguishes physical AI from physics AI and discusses the potential of AI that understands physics.