Wedge

Start with structural steel fabricators finishing heavy weldments like I-beams and base plates. This niche suffers severe labor shortages and features simpler, rectilinear geometries that are easily mapped by early vision models. Expand outward by targeting precision sheet metal fabrication, followed by complex multi-axis aerospace or medical device deburring.

Timing

Recent advancements in 3D point-cloud processing and spatial AI models allow software to map part geometry and weld seam variations in seconds. Combined with the deployment of cheaper optical sensors in robotic cells, closed-loop AI path generation executes accurately without human intervention.

Why This ICP

High-mix, low-volume contract manufacturers face acute shortages of skilled grinders and cannot amortize the cost of traditional manual robot programming over small production runs. They require immediate, zero-touch setup to make robotic finishing profitable on the shop floor.

Size Of Prize

There are roughly 15,000 robotic-capable metal fabrication shops in the US. Capturing an average annual spend of $20,000 per shop for AI path-generation software yields a total addressable market of $300M.

Gap Narrative

Metal fabricators spend hours manually programming robotic grinders or relying on high-turnover manual labor to clean welds and deburr parts. Current CAM software requires tedious, part-specific path definition that makes robotic grinding uneconomical for high-mix, low-volume production. AI translates 3D scans and CAD models directly into optimized grinding toolpaths instantly, removing the programming bottleneck entirely.

Defensibility

The system builds a compounding data moat by recording the delta between generated toolpaths and actual operator corrections or surface finish quality inspections. As the model ingests specific data on abrasive wear rates across different alloys and robot kinematics, its first-pass yield becomes a permanent operational dependency with high switching costs.

Why This Thesis

An Agentic workflow acts as an invisible CAM programmer, ingesting a CAD file and scan data to directly output robot-native instructions. This approach bypasses the need for operators to learn complex new interfaces, mapping directly to the shop floor's demand for hands-off automation.