Forge Robotics Is Teaching Welding Robots to See, and American Manufacturing Needs It Yesterday

The Macro: Nobody Wants to Be a Welder Anymore

There is a labor crisis in American manufacturing that does not get nearly enough attention. The American Welding Society projects 600,000 unfilled welding positions by 2030. That is not a rounding error. That is a structural failure in the workforce pipeline that threatens everything from shipbuilding to infrastructure repair to the entire reshoring narrative that politicians on both sides love to talk about.

The problem is demographic. The average age of a skilled welder in the US is climbing every year. Young people are not entering the trades at replacement rates. Welding is physically demanding, the training pipeline is slow, and the work is legitimately dangerous. Meanwhile demand is accelerating. Defense spending is up. Infrastructure bills are flowing. Semiconductor fabs need steel structures. Data centers need metal fabrication. Everybody needs welders and nobody can find them.

Industrial robots have been welding for decades, of course. But traditional robotic welding requires extensive programming for every new part geometry. A human operator has to teach the robot exactly where to move, how fast to travel, what angle to hold. When parts come in with slight variations, which they always do, the robot does not adapt. It follows its program and produces defects. Industry-wide first-pass yield on robotic welding sits around 90%. That means one in ten welds needs rework by a human. The very human you could not hire in the first place.

This is the gap that vision-based AI is built to fill. Not replacing welders, but making robots smart enough to handle the variability that currently requires human judgment.

The Micro: Two Irish Engineers and a Very Good Camera System

Forge Robotics was founded by Eoin Cobbe and Robert Cormican (CTO), who came through Y Combinator’s Fall 2025 batch. They are building what they call intelligent automation for metal fabrication, and the core innovation is a vision system that lets welding robots see their work in real time.

Here is how it works. Before welding, the system scans the workpiece using computer vision and AI feature detection to build a real-time 3D map of the part. This eliminates the need for manual robot programming because the robot can see exactly what it is working with, identify weld joints, detect misalignments, and generate its own toolpath on the fly. During welding, a constant feedback loop monitors the process and adjusts parameters to prevent defects. After welding, the system inspects its own work.

The results they are claiming are significant. First-pass yield improvement from 90% to 99%. That is a massive reduction in scrap and rework. They also claim 90% less setup time and sub-millimeter precision for detecting misalignments. If those numbers hold in production environments, this is a genuinely transformative product for any metal fabrication shop.

The competitive landscape is not empty. Path Robotics, also in autonomous welding, raised significant venture capital and built a vision-guided system with similar ambitions. Wandelbots offers no-code robot programming. Ready Robotics does universal robot orchestration. Veo Robotics works on human-robot collaboration with 3D sensing. But most of these companies are tackling the programming problem, not the perception problem. Forge is specifically focused on giving robots the ability to see and adapt to real-world variation, which is the harder and more valuable problem to solve.

Robert Cormican is listed as CEO on the website, reachable directly. The team is operating out of San Francisco. The site is clean, the messaging is tight, and they are clearly targeting manufacturing operations that already have robots but are frustrated by the programming bottleneck.

What I like about this approach is the pragmatism. They are not trying to build a humanoid robot or reinvent the factory from scratch. They are adding intelligence to robots that already exist in thousands of fabrication shops. That is a faster path to deployment and revenue than building custom hardware from the ground up.

The Verdict

Forge Robotics is positioned at the intersection of two powerful trends: the manufacturing labor shortage and the maturation of real-time computer vision. The labor problem is not going away. If anything, it is accelerating. And the technology to solve it is finally good enough to work in the messy, variable, high-stakes environment of metal fabrication.

The risk is the classic robotics startup risk: hardware-adjacent businesses are capital intensive and sales cycles in manufacturing are brutally long. Getting a fabrication shop to trust a new vision system on their production line takes months of proof-of-concept work, integration, and hand-holding. Path Robotics learned this lesson at scale. Forge will need to as well.

I would also watch the 99% yield claim carefully. Lab conditions and production floor conditions are very different animals. Heat distortion, fixture variation, material inconsistency. Real-world welding is messy in ways that are hard to simulate. If the 99% number holds across diverse part geometries and production volumes, this company will have no trouble finding customers. Every percentage point of yield improvement translates directly to money saved.

At thirty days, I want to see pilot deployments with real fabrication shops. At sixty days, I want yield data from production runs, not demos. At ninety days, the question is whether the sales cycle is manageable for a small team or whether enterprise manufacturing procurement is going to grind them down before they can scale. The product thesis is sound, the market need is urgent, and the founding team has the right technical DNA. Now they need to survive the sales cycle.