The Next AI Disruption Might Happen on the Factory Floor, Not Your Computer Screen.
Explore how AI can revolutionize manufacturing by enhancing inspection processes and supporting operational efficiency.
Article
The operations director of a growing British manufacturing business is reviewing a problem that has become increasingly expensive. The company produces specialist moulded components for industrial customers, operating several production lines with employees responsible for inspecting finished parts before they are packed and dispatched.
Most products meet the required standards, but some contain surface defects, dimensional irregularities or other issues that need further investigation. Experienced inspectors recognise many of these problems quickly, although the work is repetitive and the appearance of acceptable products can vary between production batches. Management wants to improve consistency without slowing production or compromising quality.
A technology supplier has proposed an AI-supported inspection system. Cameras would capture images of components as they move through the production process, while a trained model would identify visual characteristics associated with potential defects. Depending on the result, the system could alert an employee or direct a component towards an established inspection station.
The demonstration is impressive, and management can see how the technology might reduce some repetitive visual work. The proposal also introduces costs, technical dependencies and operational responsibilities that don't exist in the current process.
The operations director is interested but cautious. The factory already uses computer-controlled machinery, sensors and conventional automation, so introducing AI isn't automatically a step forward. The question is whether the proposed system can identify defects more consistently, support employees and operate reliably under real production conditions.
It also needs to avoid rejecting too many acceptable components or allowing faulty products to reach customers. A technically sophisticated model would offer little comfort if it created production delays or introduced new quality problems.
This is a hypothetical manufacturer, but I think its situation illustrates an important development in the AI conversation. Most public attention still concentrates on software: language models, coding assistants, document analysis and agents that interact with business applications. Those technologies are significant, and many organisations have practical reasons to investigate them.
But AI is also developing capabilities involving perception, robotics and interactions with physical environments. Those capabilities could influence businesses whose most important work happens around equipment, materials and production processes rather than computer screens.
I was thinking about this while reading Boston Consulting Group's October 2026 report, *The AI Maturity Matrix 2026*. BCG argues that current AI exposure is concentrated heavily in knowledge-intensive activities, where models can work with language, code and other digital information. It also anticipates that the range of exposed activities could expand as physical AI becomes more capable.
Countries with established strengths in manufacturing, robotics and industrial engineering may therefore possess advantages that differ from those associated with developing foundation models. For some businesses, the next useful AI application may not involve generating an answer. It may involve understanding what's happening in a physical process and helping people act on that information.
## Factories have used automation for decades. What does AI actually add?
Industrial automation isn't a new development created by the recent advances in artificial intelligence. Manufacturers have used programmable machinery, industrial robots, sensors and computer-controlled systems for many years. These technologies already perform repetitive movements, monitor operating conditions and coordinate production activities with considerable reliability.
In many environments, conventional automation is effective precisely because the tasks are predictable and the required responses can be defined clearly. Introducing a more complex AI component isn't automatically an improvement.
Consider a production line that handles components with highly consistent dimensions and appearance. A conventional inspection system may be able to check whether a part falls within an established tolerance or whether a particular feature is present. If those checks perform reliably, the business may have little reason to replace them with a machine-learning model.
The alternative could increase development and maintenance costs without producing a meaningful improvement. The commercially sensible choice may be to retain the existing approach.
Our hypothetical manufacturer faces a more variable problem. Some components have slight differences in surface finish, colour or texture that are acceptable within the product specification. Other variations indicate defects that should trigger further inspection or rejection. Employees make these distinctions using experience, knowledge of the product and an understanding of what matters to the customer.