Around the world, governments are racing to build AI capacity through investments in compute infrastructure, talent programmes, and sovereign AI initiatives. This June, the UK government used London Tech Week to reset its AI strategy with a £1.1 billion AI Hardware Plan, committing £400 million to compute capacity and £20 million to an Early Careers Jobs Alliance to protect entry-level roles and help young people adapt to AI-driven work.
The chips and hardware investment is the right call. Compute capacity and skills are genuine constraints, and the Early Careers Jobs Alliance shows a government paying attention to something most companies are still ignoring.
But compute isn’t where the productivity problem lives – deployment is. Capability has moved fast, but getting AI working inside real operations has not kept pace.
AI is stuck in the office
The economy doesn’t run from a desk. Most AI adoption in this country is concentrated in general-purpose productivity tools and administrative automation. Research from YouGov shows that the most popular uses for AI in the UK workplace include summarising information (60%), research (58%), and editing and checking text (56%). Its study shows that the most popular tools are, by a long way, Microsoft Copilot (58%) and ChatGPT (48%).
Yet research from Boston Consulting Group counters that the vast majority of operational work – up to 70-80% – happens in the field, not behind a screen. It is carried out by maintenance technicians, service engineers, and operators keeping the plant running, but most AI investment hasn’t caught up with that reality. Manufacturing, energy, logistics and infrastructure – the sectors carrying most of the productivity problem – are being handed tools designed for someone else’s day. If we want AI to actually move the needle on productivity, it has to follow people out onto the floor and into the field, not just sit in their inbox.
Focusing AI where it helps the most
UK productivity sits 10% lower than the G7 average, according to the Office for National Statistics (ONS). Manufacturing carries much of that burden, while construction’s low productivity is being exacerbated by a shrinking workforce.
Capital intensity, systemic complexity and ageing estates are real constraints on these sectors, but they’re not what’s kept AI out of the operation. The harder deployment isn’t a physics problem, it’s an unglamorous one. Making AI useful in an industrial setting means grounding it in the system of record, wrapping it in deterministic guards so it can’t guess its way into harm, and curating the historical data it depends on. Most industrial AI use cases still need large volumes of accurate, well-maintained data, and organisations that haven’t invested in collecting and connecting it stay stuck in pilots. That’s the gap between experimentation and enterprise scale, and no amount of compute closes it.
If the Government wants the UK to succeed here, the focus should be on how businesses apply AI beyond office productivity, inside core operational systems. Three areas stand out:
Maintenance and asset management
A thread of intelligence is taking shape across industrial operations: real-time data feeds from assets, predictive algorithms running under the bonnet, and intelligent systems connected across otherwise intricate business processes. Too many businesses have got into the habit of replacing assets rather than maintaining them. Predictive maintenance changes that calculation. By continuously analysing telemetry, thermal imaging and acoustic vibration data, and giving technicians direct digital access to an asset’s history, blueprints, manuals and past repairs right at the machine, AI can catch failures before they occur. One of our customers, a major electronics manufacturer, put this into practice and cut maintenance turnaround time by 40%, error rates by 25%, and costs by 15%.
Supply chain forecasting
The traditional approach relies on historical lead-time averages to predict what happens next. AI adapts in real time to variables such as port congestion, border friction, and energy market fluctuations. For example, a sudden disruption to a key shipping lane when ordering windows can be adjusted automatically to bypass bottlenecks before they disrupt production. It is worth being precise about what does the work here: the forecasting and optimisation engines are established machine learning and AI, not language models. What is new is that agents can reason over the top of those engines – calling them, not replacing them.
Field service
Specialised Industrial AI systems can support crews in the field by, for instance, translating photos, videos, and voice memos from the site into structured, enterprise-grade work orders. AI tools can also deliver step-by-step diagnostic guidance to support engineers in the field with easy-to-understand walkthroughs. The technician stays in charge, using AI to inform each decision. That is a design choice, not a limitation. Where information is incomplete, a well-built industrial system asks a human or defers to recommending rather than executing. In safety-critical work that is not friction – it is the price of operating, and it is what makes everything above it usable.
The promise of AI in operational sectors
A growing challenge across industrial economies is the rapid retirement of experienced engineers, technicians, and frontline workers. Businesses risk losing decades of operational expertise unless they capture, codify, and transfer that knowledge to the next generation.
AI can help bridge this gap. Rather than seeing years of valuable institutional knowledge vanish, businesses can use AI to capture how veteran engineers troubleshoot complex problems, translating that know-how into instant, step-by-step guidance for frontline workers. AI effectively becomes the memory of the business.
The limits matter, though. Codifying procedure is much easier than transferring judgment. A veteran engineer’s value is not only the fix; it is knowing which of six plausible explanations to check first built through years of hands-on experience that AI now absorbs. If younger workers only ever receive the answer, they never build the muscle that made the veteran worth listening to.
This is the most underrated risk in AI adoption, and it is a design problem for employers rather than an inevitability. It is also where the £20 million Early Careers Jobs Alliance could do real work, if it funds a redesign of how expertise is built rather than another round of AI awareness training.
Building the right foundations
The Government is taking AI seriously, and the infrastructure investment is good news. It needs to be matched by the same seriousness about deployment.
Three things would do more for that than another headline compute number: explicit accountability when an agent takes a real-world action, so responsibility sits with the organisation that deploys it, never the software; a UK baseline for secure, interoperable, tool-connected agents, so industry can scale without importing risk; and funding aimed at the entry-level pathways where operational expertise is actually built, not just AI literacy training.
Bob De Caux
Bob De Caux is Chief AI Officer at IFS. An experienced technology executive, Bob combines deep technical expertise (PhD in AI and complex systems simulation) with an in-depth knowledge of how to use those techniques to build products and create customer solutions across a number of sectors, including financial services, insurance and healthcare.
