AI & the future of manufacturing

AI inspection links every camera image to the product variant and the process conditions, so the same scratch can be a defect on one product and acceptable on another. The system learns from examples of good and faulty items, and can also flag deviations it has never seen before. Across thousands of products, AI spots patterns: a flaw that keeps recurring with the same mould, at the same time of day or whenever the same batch of raw material is used. Linked to process data, that helps trace the step where a defect starts. Ahead lies a shift from finding defects to preventing them: the line recognises that the risk of deviations is rising and adjusts before more products are rejected. The inspector judges the borderline cases.

AI compares wear patterns between machines running under different loads, and can tell the signs of a wearing bearing from the effect of a heavier product variant. The fitter gets an alert with a possible cause, sometimes with an estimate of the remaining service life, and alerts are ranked by their impact on production. Using planning and stock data, the system shows which maintenance can be carried out during the next stoppage, with the required part ready. The next step is a machine that arranges its own maintenance: an AI agent orders the part and books the fitter, and in time a robot carries out simple replacements itself.

Where fixed planning rules reach their limits, AI draws on patterns in historical lead times, product variants and past disruptions. It works through thousands of possible sequences for several lines at once and produces a new proposal within minutes, with as little changeover time and delay as possible, including the knock-on effect on the next orders. In food production, the order of allergens matters too: products without nuts first, then those with nuts, so the line needs cleaning less often in between. The planner reviews the proposal, weighs what the system cannot see and stays in touch with sales and customers. Further ahead, the planning system factors in supplier stock and customer demand by itself and keeps updating the schedule.

AI process control means a line or process keeps adapting to what is happening, rather than running on fixed settings and rules. From sensor data, camera images and quality measurements, AI learns how raw materials, machine settings and the end result are connected. Those connections often only show up across thousands of products or batches. While production runs, the system adjusts speed, temperature, pressure or dosing, on an injection moulding machine as readily as in a battery plant or a food process. That can make a process more stable, with less scrap and less energy per product. The operator sets the limits, approves larger interventions and steps in when something deviates. Each time the model is retrained, its predictions are checked for improvement.

An industrial copilot is a language model connected to a factory's own documents: manuals, fault logs, work instructions and maintenance reports. Querying those documents is already routine. The next step is a copilot that can see what the fitter sees. Through a camera or a pair of light AR glasses it follows the work, recognises the machine and the part, and guides the repair step by step with spoken instructions: which screw first, where the sensor sits, what was done last time. Assistants like this are now turning up in trials. Even at three in the morning, a new employee or a multilingual crew then gets guidance drawn from manuals and the fault history, matched to what the camera sees.

In generative design an engineer starts with requirements rather than a drawing: the load the part must carry, its maximum weight, the space it must fit into and how it will be made. AI then works through hundreds or thousands of shapes that meet those requirements. The results often look organic, like a bone or a branch, because AI places material mainly along the load paths. The parts come out lighter, which cuts fuel use or the battery capacity needed. The engineer's work shifts from drawing to setting the right requirements. Later, AI also designs how the part is made, and the engineer describes the requirements in plain language.

An AI agent uses a language model to plan a series of steps itself and carries them out across different systems, within limits people set in advance. Take a fault on a packaging line: the agent reads the machine's alerts, searches the log for similar cases, raises a work order, checks whether the part is in stock and proposes a time for the fitter. Manufacturers are testing such agents in clearly bounded processes, such as maintenance, buying spare parts and quality reporting. The next step is agents from the factory, the supplier and the maintenance contractor coordinating with each other, so a fault is fixed before anyone picks up the phone.

Many factories still burn natural gas for steam, drying ovens or melting. Heat pumps upgrade waste heat, electric boilers supply steam and electric furnaces heat directly. For very high temperatures, hydrogen is also being trialled as a fuel. Power prices and the emissions of the electricity mix change over the course of the day. AI helps match production to those swings: an energy-hungry step such as drying or cooling moves to hours with plenty of solar and wind power, and stored heat covers the hours in between. Further ahead, the factory becomes a flexible user of the grid, with AI deciding when to draw less power or use stored heat.

Cobots work alongside people, often without a fence: depending on the application, the robot limits its force, or sensors make it slow down or stop when someone comes close. That puts automation within reach of companies with small batches and frequent changeovers, where a large robot cell does not pay off. AI extends what cobots can do. With cameras they recognise parts that are not lined up neatly, and some learn a new task when an employee guides the arm through the movement a few times. The cobot takes over the lifting, fastening or packing; the employee sets it up, checks the work and improves the process.

A change to one line ripples through buffers, energy use and lead times elsewhere in the factory. A digital twin of the factory connects live machine data with process models, so engineers can work out those knock-on effects before they change anything: an extra robot, a different hall layout, a new product on an existing line. AI runs those scenarios for different order books and utilisation levels, and shows where bottlenecks appear. Manufacturers plan new plants and lines entirely on screen first, so mistakes surface there rather than during construction. The next step: digital twins of factories and suppliers linked into a twin of the whole supply chain that coordinates and corrects itself.

Measuring, repeating, calculating, monitoring and predicting: in an AI-first factory, robots and AI do that work so people do not have to. People concentrate on what takes judgement and experience: assessing a deviation that appears nowhere in the data, coming up with an improvement, helping a customer with a tricky question. Where skilled workers are scarce, that division of labour can ease staffing pressures. Every new process is a chance to redraw that division.

As machines do more on their own, the work of the people around them changes. The operator who used to run one machine now watches several automated lines from a screen and steps in where the system flags a deviation. Towards 2040, roles will emerge that barely exist today. A fleet manager deploys a team of humanoid robots across the shifts. A trainer teaches robot models new skills, such as handling a different product or gripping a part in a new way. A supervisor of AI agents sets which decisions the agents in maintenance and purchasing may take on their own. And a materials designer works with AI and an autonomous lab on the raw materials of tomorrow. Work on the floor shifts from operating to directing.

In the factory of the future, people, robots and AI solve a problem together. A fitter, a cobot doing the heavy lifting and a copilot that knows the manuals and the fault history form one team. The robot does the repetitive and heavy work, the AI searches, calculates and predicts, and the person judges, chooses and works with colleagues and customers. The interesting moment comes when the copilot's advice does not match what the fitter sees and hears. Then experience counts; recorded corrections later help improve the knowledge base and the model. In the 2030s a humanoid robot could be the fourth team member, with an AI agent dividing the work within the crew for each shift.

The factory of the future takes shape when leaders give direction and give their people a say. Leaders in industry take the initiative on AI, robots and new technology, and show how the factory could work in five or ten years and what that means for the crew. They start small and in plain sight, with one line or one process. In a factory where machines and AI agents do much of the work, leaders set the goals and limits for those systems, and help their people through work that changes every year.

An experienced fitter can hear a bearing wearing out and knows which deviation is still normal for this machine. For AI, that knowledge is something to learn from, once employees link their experience to measurements and decisions: which sound goes with which vibration, which deviation matters. That takes more than technology. It takes a willingness to share knowledge and work differently, including from people who wonder what their role will be. Further ahead, cameras, AR glasses and voice recordings capture the work of experienced craftspeople as they do it. An AI apprentice learns from it, new colleagues can watch it back, and that practical knowledge stays in the factory when the craftsperson retires.

What employees need to know about AI and new technology changes every year. Colleges and training providers need more time to build that into their programmes. Organisations therefore set up their own future academy, where they prepare their people for the future themselves. Existing staff learn to work with AI, robots and AI agents. New colleagues start there with the knowledge tomorrow's work demands. Experienced craftspeople pass on their know-how, so it does not disappear when they leave.

Innovation rarely starts inside a single organisation. A new technique emerges in an ecosystem of partners, start-ups, scientists, customers and governments. Organisations in the same region share test facilities, skilled people and sometimes data, for instance to train an AI model together, often in programmes that governments help make possible. Front runners share their approach in international networks, so others do not have to start from scratch. Training a model together without sharing raw data is already possible. Done at scale, it lets even competitors learn from each other's breakdowns and mistakes.