Sold as a cure for everything, AI is too often a solution bought before the problem is found. Firms are now paying hard to undo the mistake.
We’ll fix it with AI. Get AI to do it. AI can do that. One day, those phrases may be true, but right now they look a great deal like wishful thinking. Ask Ford, which acted on precisely that logic but which has spent much of 2026 more or less reversing it. By the end of June the firm had reemployed hundreds of veteran engineers to catch quality faults that its automated inspection systems had missed, CNBC reported in July. And Ford is not alone. The Commonwealth Bank of Australia reversed AI-related job cuts after finding the technology could not do the whole job. Forrester’s 2026 Future of Work report put the share of employers who regret their AI layoffs at 55 per cent. In May, Boston Consulting Group (BCG) released the findings of a poll (covering 625 chief executives and directors) which found that 35 per cent of bosses said their boards “overestimated” what AI could replace, and that some 60 per cent said boards were “too impatient” to wait. Get AI to do it, in such circumstances, looks a lot more like wishful thinking than an actual strategy.
And that’s the point, in essence. AI is not, in and of itself, a strategy. Brian Stafford, chief executive of Diligent, an AI governance, risk, and compliance firm, even gave the phenomenon a name June, when a survey his firm helped run ranked AI as directors’ top capital-spending priority but, at the same time, their most-overlooked area of oversight. Stafford called it “a real disconnect”, which is a polite understatement. Boards are approving AI spending, he said, without the discipline to govern it (or, he might have added, a precise idea of what they might be doing with it).
And it might get even worse. Anushree Verma, a senior analyst at Gartner, has forecast that more than 40 per cent of agentic-AI projects could be scrapped by the end of 2027, undone by a mix of cost, unclear value, and weak controls. Gartner believes that only about 130 of the thousands of firms selling ‘agentic’ tools offer anything of the sort, a phenomenon it calls ‘agent washing’. Most projects are, in Verma’s words, “driven by hype and often misapplied”, not least that many use cases might never need an agent at all.
Nor is the dash for AI restricted to the private sector. Kathrin Frauscher, deputy director of the Open Contracting Partnership, has found the same pattern in government. Her group’s Buying AI report described agencies buying off-the-shelf tools they had no in-house means to vet, in pursuit of savings they had not defined. One official at America’s General Services Administration warned buyers against becoming the AltaVista or Ask Jeeves of AI. (Indeed, if all of this sounds highly familiar to anyone who lived and worked through the dotcom bubble, that’s not a coincidence).
Meanwhile, the AI hype machine continues to tout miracle cures for just about any organisation on Earth. The gap between rhetoric and reality, however, is in many cases becoming wider and wider. It might be argued that is perhaps how it should be, given that genuine reinvention has never been about technology (whether it’s deploying spinning jennies or AI) alone. Real reinvention requires a lot more, such as vision, courage, coherent systems, and the patient, painstaking work of changing how organisations actually operate. No algorithm can automate that away, at least not yet. Until then, most firms do not need more AI, but instead clearer strategies, better data, redesigned processes, and cultures that are ready, willing, and able to embrace change. They need to stop chasing the latest technological fad and start doing the hard yards of fundamental business reinvention. AI may help with that eventually. But first, organisations must stop treating its deployment as an end in itself and recognise it for what it truly is: a tool whose value depends entirely on the skill and wisdom with which it is wielded. The revolution, for now, remains more rhetorical than real.
Photo: Dreamstime.