17.5% vs 87%
AI agents finished 17.5% of CAD tasks. Human experts finished 87%.
Human experts87%
Best AI agent17.5%
arXiv, Sept 2026 · Benchmark · Preprint
What the source measured
In the CADWorld benchmark, the best of seven AI agents completed 17.5% of 200 FreeCAD tasks versus 87.0% for human experts. Every model scored 0% on CAM, FEM and TechDraw tasks. The benchmark measured design tasks, not review.
What it means for Indian businesses
AI agents are not ready to design parts on their own.
The benchmark tested agents doing design work end to end, and they mostly failed. It did not test checking work, so what follows is our view, not the paper's.
We use AI on the parts of CAD work that are closer to reading than drawing: checking a drawing against your tolerances and drafting rules, comparing revisions, and pulling dimensions into a quote. An engineer stays in the loop on every output.
How we use AI for design review