Nomagic deploys AI brain in warehouse robots, halves human intervention in live operations

Nomagic Puts AI Brain Into Live Warehouse Robotics Operations

Nomagic, a Warsaw-based warehouse robotics firm, has deployed a vision-language-action model into live customer operations, achieving what most robot labs are still demonstrating in controlled settings1

1

. The company reports that its VLA model has roughly halved the rate at which its robots stall and require human intervention, marking a significant step forward in autonomous warehouse operations2

2

. The firm maintains European headquarters in Warsaw and a U.S. base in Sandy Springs, Georgia.

Earlier this year, Nomagic established an AI research lab and hired Markus Wulfmeier, a former Google DeepMind researcher and core member of the Gemini Robotics team, as chief scientist1

1

. A vision-language-action model combines the ability to see objects, read plain-language instructions, and then act—capabilities that many labs are pursuing for embodied AI applications.

Mastery Before Generality: A Different Approach to AI-Driven Robotics

Nomagic's strategy diverges sharply from the prevailing industry approach. While most companies race to build general-purpose robot brains that can drop into any machine, Nomagic runs the logic backwards1

1

. The company aims for models that nail one task out of the box, then builds towards a general system. "Most of our community is racing to build the most general robot brain," Wulfmeier told Fortune. "We're betting that the harder part is actual mastery and that it has to be earned in real deployments first"1

1

.

This reasoning rests on the long tail problem that has slowed self-driving cars. Training in simulation or by remote control gets a model to roughly 80 per cent accuracy, but in warehouse robotics, that threshold is insufficient1

1

. If a robot needs a human once an hour, the economics collapse. The company acknowledges that general-purpose AI models perform significantly below human-level accuracy on each task right out of the box2

2

.

Wrapping VLA Models in Safety Harnesses for Live Production Environment

Nomagic admits its VLA model is not perfect. "Our VLAs aren't at 99.9% success on their own yet," the company stated, adding that no rival's deployed VLAs are either1

1

. To bridge this gap, the company wraps the model in older "classical" software that catches errors and enforces safety. "The bar in the physical world is high: 99.9% isn't a marketing number, it's the cost of being allowed in the building," said co-founder and chief executive Kacper Nowicki. "So we built a harness that clears it from day one, while the AI inside keeps getting better"1

1

.

The company's edge comes from data generated by its deployed fleet, which makes millions of successful picks each month. Two million picks come from fashion platform Zalando alone1

1

. Nomagic trains its VLAs on this live stream of operational data, not on simulations, allowing the AI brain to learn from real-world edge cases.

First Deployment Shows Tangible Results at Brack.Alltron

The first VLA deployment runs at Brack.Alltron, Switzerland's second-largest e-commerce platform1

1

. Founder Roland Brack confirmed the shift was real: "Today, we are seeing robots that truly understand their environment. This intelligence allows us to run autonomous shifts through nights and Sundays"1

1

. By targeting the most common edge cases where robots get stuck and call for human assistance, Nomagic has achieved measurable improvements in live customer operations2

2

.

The deploy-first stance marks a real contrast with labs chasing general-purpose humanoids and raises the stakes in the race to automate Europe's warehouses. Co-founder Tristan d'Orgeval emphasized this approach: "We didn't build a lab and then go hunting for a problem. The order matters. It's what separates a demo from a business"1

1

. As the company continues to reduce robot stalls and improve operational efficiency, the industry will be watching whether mastery-first proves more viable than the generality-first approach dominating Silicon Valley's embodied AI investments.