Exciting as it is, the breakneck pace of AI deployment is creating confusion for organisations, in part because the rapid evolution causes so-called ‘analysis paralysis’. At the other end of the scale, though, there is another issue: some large organisations are ploughing ahead at different speeds. Chris Donnelly, design principal at user experience (UX) house Each&Other, believes AI’s potential benefits are now well understood. As are the risks. What is less clear, however, is how AI programmes should be planned. “Some organisations find it difficult to get going. You look at the contact centres, or change management, and you ask: ‘Does the data flow contain PII [personally identifying information]? Can we build a meaningful product? Can we do it in time?’ “By the time they're ready, the central group function has already built it,” he said. This is a classic issue for UX designers, who can bring a concrete and testing-validated methodology to process and service design and deployment. But UX itself is changing in the face of changing customer expectations, the desire to reduce factotum tasks, and, of course, new modes of interaction. AI adoption gap This is not a trivial issue. In spite of the long acres of purple prose about our AI-augmented future (much of it written by AIs) – and their eschatological funhouse mirror image, for that matter – the reality is that AI’s impact has been uneven. A recent report, The GenAI Divide: State of AI in Business 2025, published in August by the Massachusetts Institute of Technology Networked AI Agents in Decentralized Architecture (NANDA) initiative, found that some 95 per cent of generative AI pilots are failing. Is this a reason to slam on the brakes? Perhaps. On the other hand, not implementing AI is already leading to ‘shadow AI’, which is to say the unauthorised use of AI tools. Consultancy Auxilion recently published its own survey in which more than a quarter of respondents said their use of AI would be considered unethical. People might be trying to automate processes that they should really be trying to delete Worryingly, of this 27 per cent, 45 per cent fessed-up to using AI without authorisation. Donnelly said some deep thought was required. “One of the big problems is people are trying to exploit AI to solve problems. We’re not in that phase yet; you want to explore the possibility. Instead of saying ‘let’s automate customer service,’ they should work to understand the processes. “People might be trying to automate processes that they should really be trying to delete.” Is UX the answer? It can be. But UX designers themselves have some thinking – and testing – to do. “In the pre-language model world, UX involved designing what the interface looked like. The key thing that makes it different from [traditional] graphic design was the testing,” Donnelly said. “It [this process] works really well for deterministic software, but in the world of probabilistic software, it’s very different. The next step for UX as a craft is going to be less and less about the interface – that will still be important – but it will be about how the models interact, how precise they are, and how often they hallucinate. Writing evaluations is essential,” he said. Forward-thinking UX designers are now tooling up to tackle this challenge by, as the old saying goes, ‘eating their own dog food’. With UX processes re-evaluated for AI, agencies can help to deliver reliable, test-validated AI solutions, he said. Donnelly said Each&Other is addressing the gap directly through a structured workshop called LaunchPath AI (with partner Phased AI), helping clients move from confusion to foundation. “What the UX world is starting to work out is what the methodology for it [AI UX] is. The people currently doing evaluation on [AI] models are engineers; they’re not design people yet.” “You’re able to do user testing at a much bigger scale, but the phases are not going to change. “LLM technology presents a totally new paradigm but the principles of the design process will not change. We still need to understand the human factors and then work to shape the technology around them.”