What "AI for CAD" actually means
The phrase covers a lot of marketing noise, so here is the honest version. A large language model cannot replace a mechanical or design engineer, and you should be suspicious of anyone who implies it can. CAD geometry is exact, tolerances matter, and a hallucinated dimension is a scrapped part or a failed assembly. What AI is genuinely good at is the repetitive, rule-bound, high-volume work that wraps around the actual modelling, and the retrieval and first-draft work that an engineer then checks. That is what a Bedstone CAD agent does. It is software that uses an LLM as a reasoning layer, bolted to your CAD tool's API and your file formats, doing the parts of the job that do not need an engineer's judgement so your engineers spend their time on the parts that do.
Where AI fits in a CAD workflow
The high-value, low-risk applications, roughly in order of how quickly they return time:
- Drawing and documentation automation. Generate 2D drawings, dimensions, annotations, and title blocks from a 3D model to your drawing standard, for an engineer to review rather than produce by hand.
- BOM generation and reconciliation. Extract and structure bills of materials from assemblies, reconcile against ERP or purchasing, and flag mismatches.
- Design-rule and DFM checking. Check models against your company design rules, drafting standards, tolerance and GD&T conventions, and design-for-manufacturability constraints, and surface violations before they reach the floor.
- CAD file intelligence. Parse STEP, IGES, DWG, and native files to extract features, dimensions, and metadata, classify parts, and make a parts library searchable in plain language.
- Engineering knowledge retrieval. An agent over your past designs, specifications, standards, and project files that answers "have we built a part like this" and surfaces the precedent, instead of an engineer trawling the PDM system.
- Quote and estimation from CAD. Extract manufacturable features from a model to support cost estimation and quoting, connecting CAD to the commercial side (see AI agents and automation).
- Generative and parametric assistance. Draft parametric variants or first-pass geometry from constraints and requirements, including text-to-CAD approaches, as a starting point an engineer refines, never as an unchecked output.
- Revision and change management. Summarise what changed between revisions, draft engineering change documentation, and route approvals.
What we build
We build bespoke. There is no off-the-shelf Bedstone CAD product, because every engineering team's tools, standards, and workflow differ. A typical engagement starts with one high-value task from the list above, builds the agent against your actual CAD stack, validates it on your real designs, and then extends to the next task once it has earned its place. The agent runs with a human checkpoint wherever a design decision carries risk, ships with an eval set so it does not silently drift, and is yours at the end, code, prompts, and integration included.
Tools and formats we integrate with
We build against the documented APIs, SDKs, and file formats of the tools you already run.
- CAD platforms. SolidWorks, Autodesk Fusion 360, AutoCAD, Inventor, Onshape, PTC Creo, Siemens NX, and Rhino/Grasshopper, via their APIs and SDKs.
- Neutral and native formats. STEP, IGES, DWG, DXF, Parasolid, STL, and 3MF, parsed via CAD kernels such as OpenCASCADE where direct geometry access is needed.
- Generative and text-to-CAD. Emerging programmatic CAD interfaces, including code-based modelling and text-to-CAD APIs, where they fit the task.
- The reasoning and retrieval layer. Model-agnostic LLMs (OpenAI, Anthropic, Google, or open-weight models in your own VPC for sensitive IP), with retrieval over your design history, standards, and specifications.
- The systems around CAD. PDM and PLM, ERP and purchasing, and document stores, so the agent connects design to the rest of the business.
Where AI does not belong in CAD
We are explicit about this because over-promising on engineering AI is how trust gets destroyed. We do not build agents that produce final, safety-critical geometry without engineering review. We do not replace the engineer's judgement on tolerances, materials, or load cases. We do not pretend a text prompt produces a manufacturable part with no human in the loop. Where a task genuinely needs an engineer, the agent assists and the engineer decides. If a workflow you have in mind sits on the wrong side of that line, we will tell you on the first call rather than sell you a build that should not exist.
How an engagement runs
- Scope the workflow. A discovery sprint maps your CAD workflow, the tools and formats in use, and the repetitive tasks worth automating. You get a recommended architecture and a fixed-price build quote.
- Build against your stack. We build the agent against your CAD tool's API and your file formats, with human checkpoints designed in.
- Validate on real designs. We run it against your real parts, drawings, and standards, and tune until it holds up on the work your team actually does.
- Hand over and operate. Working automation in your environment, with a runbook and full ownership. Extend it to the next CAD task as a build or a retainer.
What it costs
Most teams start with a discovery sprint from $7,500 to scope the workflow and produce a fixed-price build quote, or an Automate Your Role engagement from $12,000 to take a single high-value CAD task off an engineer's plate. Full custom AI agent builds start at $40,000 and scale with the number of systems and the complexity of the workflow. The pricing page covers the full breakdown.
R&D Tax Incentive
Integrating AI with CAD workflows involves genuine technical experimentation, and that work commonly qualifies for the Australian R&D Tax Incentive. We document the experimentation contemporaneously so your accountant or RDTI consultant can use it. We do not provide tax advice. See our R&D Tax Incentive guide for the detail.
Common questions
Can AI actually do CAD?
Not the way a marketing demo implies. CAD geometry is precise and engineering mistakes are expensive, so an AI agent does not replace an engineer's judgement. What it does well is the repetitive layer around the modelling: drawings and BOMs, design-rule and standards checking, file parsing and extraction, first-pass parametric variants for a human to refine, and retrieval across past designs. We build those, integrated with your CAD tools, human-in-the-loop where it matters.
Which CAD tools do you integrate with?
The documented APIs and SDKs of the major platforms: SolidWorks, Fusion 360, AutoCAD, Inventor, Onshape, Creo, and Rhino/Grasshopper, plus neutral formats like STEP, IGES, DWG, and Parasolid via kernels such as OpenCASCADE. The approach is scoped in discovery against your actual stack.
Do you have a CAD AI product I can buy?
No. We build bespoke, because every engineering team's workflow, standards, and tooling differ. We scope your specific workflow in a discovery sprint and build the agent against it.
What does a CAD AI engagement cost?
A discovery sprint from $7,500, an Automate Your Role engagement from $12,000 for a single task, or a full custom build from $40,000. See the pricing page.