For decades, programming a factory has looked the same. Engineers write thousands of lines of code to tell a machine exactly what to do, exactly when to do it, and exactly how to respond when something changes. If a product shifts even slightly on a conveyor belt, the whole system can fall apart. It’s rigid, it’s slow, and it hasn’t kept up with the pace that modern manufacturers need to move.
AIR, short for Automated Industrial Robotics, founded by Irish entrepreneur Darragh de Stonndún, is doing something about that.
Instead of relying on traditional hard-coded programming, AIR is training its vision systems and robots the same way you’d teach a person. Show the system what a good part looks like. Show it what a bad one looks like. Let it learn the difference on its own. Just examples, data, and systems that get smarter the more they see. Despite sounding simple, in practice, it’s a fundamental shift in how automation works.
Traditional manufacturing code is brittle. A vision system that’s been programmed to inspect a medical device, for example, needs specific lighting conditions, specific angles, and specific tolerances written into its logic. Change the packaging, change the lighting, or introduce a new product variant and an engineer has to go back in and rewrite portions of that code. It’s expensive, it’s time-consuming, and it creates bottlenecks that slow production down.
Now multiply that across an entire factory. Hundreds of machines. Thousands of product variations. Every change means someone has to sit down, open the code, figure out what needs to move, test it, validate it, and push it live. That cycle can take weeks. In industries like life sciences or food and beverage, where regulatory requirements add another layer of complexity, it can take even longer.
AIR’s approach borrows from the same principles behind how companies like Tesla train their self-driving systems. Instead of writing rules for every possible scenario, you feed the system thousands of examples and let it build its own understanding. A vision system trained this way doesn’t need to be told that a scratch on a surface is a defect. It’s seen enough good parts and enough bad parts to make that call on its own. And when a new variant comes through, you don’t rewrite the logic. You show it new examples and let it adapt.
Apply that to a factory floor and you start to unlock something that traditional automation has never been able to deliver: flexibility at scale.
That’s the piece most manufacturers are missing right now. The pressure to produce more variants, faster, with shorter runs and tighter turnaround times has been building for years. Consumer demand has shifted. Personalisation is no longer a nice-to-have. But the automation infrastructure most companies are running on was designed for a world where you made one thing, the same way, millions of times. That world is disappearing.
Hard-coded systems can’t keep up with that shift. They weren’t built for adaptability. Every time the product changes, the code has to change. Every time the code changes, the line stops. Every time the line stops, money walks out the door.
AI-trained systems flip that equation. The line doesn’t stop because the system already knows how to handle variation. It learned from the variation. It was trained on it.
AIR is deploying these systems in production environments today, across real product lines, in real facilities. Not as pilot programs or proof-of-concept demos. As the actual infrastructure running the floor.
The manufacturing industry has been programming machines the same way since the 1980s. The factories that figure out how to teach them instead of code them are the ones that will still be competitive in ten years. The rest will still be waiting for an engineer to rewrite a script.
