What the AI actually does
Modern AI reads the shape of a part from an image and proposes a full drawing around it: front, top and side views, a title block, and dimension lines placed where a drafter would put them. The work that used to take an afternoon of clicking happens in seconds.
This part of the problem is genuinely solved. Recent research on generating editable CAD from a single image moves fast, and the quality of auto-placed views and annotations is already good enough to save real time.
The one thing a photo cannot tell you
A single photograph cannot reveal the absolute size of an object. A 40 mm bracket photographed up close and a 400 mm bracket photographed from far away can produce the same pixels. That is the physics of a camera, not a flaw the next model will fix.
The workflow that actually works
The honest division of labor is simple. The AI proposes the geometry and the drawing. You put a caliper on one feature, enter that measurement, and confirm the tolerances that carry your intent. That single reference fixes the scale of the whole drawing and turns hours of drafting into minutes of checking.
For the full picture of how that pipeline works, see what AI can actually do from a photo, and for the bar a finished sheet has to clear, our manufacturing-ready checklist.