Feb 13, 2026
Most companies are using AI wrong
4 MINUTES READ
Most articles, papers, and discussions agree: new technologies such as Artificial Intelligence, Digital Twins, and Extended Reality can help optimize, simplify, and accelerate processes. The White Paper “Proof over Promise” (https://www.weforum.org/publications/proof-over-promise-insights-on-real-world-ai-adoption-from-2025-minds-organizations/) from the World Economic Forum describes a long list of companies using Artificial Intelligence to transform and improve their processes. Most of these companies are in the IT sector, but the list also spans energy, health, retail, finance, and many more. The White Paper presents impressive figures such as “50% less operator effort”, “$140.6 million reduction in R&D costs”, or “projects delivered 18% faster”. The areas covered by AI include supply-chain AI agents, AI-guided control centres, and AI-assisted design.
At first glance, the overview of listed companies from various countries and sectors reads as an overwhelming flood of success stories — and rightly so. It demonstrates that AI can be deployed across many industries to make processes leaner and faster. However, this White Paper does not have the space to explore individual AI integrations in detail; it is intended as a high-level overview. Yet this same trend extends well beyond the WEF White Paper and runs through much of the broader reporting on AI: the benefits are praised, but the specifics of how they were achieved are rarely mentioned.
The paper “Digital Twins, Extended Reality, and Artificial Intelligence in Manufacturing Reconfiguration: A Systematic Literature Review” (https://www.mdpi.com/2071-1050/17/5/2318) examines the specific case of AI in manufacturing — particularly in combination with other technologies to map the entire process. In their literature review, the authors identify similar issues under the heading “Research Gap”. While the analysed articles and papers outline the benefits and potential of AI (in combination with Digital Twins and Extended Reality), they provide no frameworks or other integrations showing exactly how these technologies are deployed. Even papers like this one, which focus on individual industries, only touch on specific integrations at a surface level. Furthermore, reports pay insufficient attention to the synergies between AI and other technologies such as Digital Twins or Extended Reality. AI is seen as a bridge between humans and technology, but what that bridge looks like — and what else contributes to it — is often unclear.
And yet these are precisely the points where AI implementations often fail, or simply don’t pay off in the long run. AI must be integrated into the existing process in a way that creates a seamless transition. With so many missing details in published reports, it is difficult to assess how AI can be effectively deployed within one’s own organisation. Simply standing up a generic AI agent that operates separately from running systems is a neat gimmick — often impressive at first glance — but not viable in the long term. Such an agent is fed data manually, is not specialised, and therefore produces formally correct but often contextually inappropriate answers. The manual integration of new data means the agent requires frequent updates, leading to additional effort and cost. To counteract this overhead, the AI agent should be embedded in the process and the existing pipeline, and be able to source its own data autonomously.
Data preparation, provisioning, and regular updates are therefore at the core of any AI application. Only with a clean, accurate foundation can outputs be reliable and usable. The design for AI integration should therefore be oriented around the entire process. AI agents are integrated into the complete decision architecture and deployed as components of the whole process — otherwise AI remains merely an interesting gimmick that gets shelved after a few months and fails to deliver long-term value. AI is a means to an end: a tool to support decisions and optimise processes, and it should be deployed as a specialised instrument — not as a magic solution for everything, even if the results often look like a miracle.
News & Insights