The Factory Deployment Gap: Bright Machines, Roboto AI & Kinisi Robotics | MACHINA 2026

In this talk

1. 0:45

   Why pilots stall before deployment

   The panel points to operational ownership, uptime, and evidence of value as barriers between a demo and factory use. Sviatoslav says the split between product designers, contract manufacturers, and technology suppliers slows adoption, while Yves says teams lack both application-specific training data and operational evidence to justify scaling.

2. 10:53

   Choose practical tasks and account for the full system

   Brennan says Kinisi screens customer requests and focuses on repetitive work, such as moving totes and assembly-line tasks, that can be reused across customers. He describes combining learned models with conventional controls to handle edge cases, and says a recent pilot ran reliably for an eight-hour shift at about half human speed, with robot costs also about half those of a person. The panel stresses that integration, data, maintenance, and service are central parts of the solution.

3. 9:19

   Build a production data and support loop

   Yves describes a data paradox: robot fleets generate huge volumes of multimodal data, yet companies still lack enough well-curated examples to train dependable models. He says useful infrastructure should collect and index data, make it searchable, and help teams find failures and route them to the right engineers. The panel also calls for forward-deployed reliability engineers and stronger DevOps practices at customer sites.

4. 22:42

   Design for manufacturing, safety, and economics

   Brennan warns that hardware costs can erase margins, giving the example of a $50 component adding $500,000 in annual cost at 10,000 units; he argues companies may need to earn through software or services. Vincent says a manufacturable robot must be reproducible a thousand times by someone outside its design team, and that industrialization needs dedicated planning. The panel adds that large customers expect safety standards and certified components, which early prototypes may overlook.

5. 29:00

   Set realistic expectations for automation

   Sviatoslav distinguishes established, task-focused automation from more experimental general-purpose robotics, noting that production use can already reach hundreds of thousands of units in specific applications. Another panelist cautions that data quality and provenance matter more than simply collecting more. Brennan observes that companies sometimes pursue humanoids while basic AMR deployments could address unmet automation needs.