Why AI Changes Jobs Slower — and More Unevenly — Than the Headlines

A demo and a deployment are not the same thing. A model that can draft a contract, answer a support ticket, or write a function in a controlled test is impressive — but whether that capability is actually running inside a company, on real work, at scale, is a separate question. Looking at 2026, the distance between those two things still seems surprisingly wide.

The pilot that never reaches production

One of the more sobering findings of 2025 came from MIT's Project NANDA. Its preliminary report, The GenAI Divide, said that about 95% of organizations were getting zero return from their generative-AI initiatives, with only about 5% of integrated pilots extracting millions in value. The report counts a tool as successful only if users or executives describe a marked and sustained productivity or profit-and-loss impact, and it rests on a review of more than 300 public AI initiatives, interviews with 52 organizations and a survey of 153 senior leaders, so it is best read as an early signal rather than a census. Its authors attribute most failures to brittle workflows, a lack of contextual learning and a poor fit with day-to-day operations, not to the quality of the models.

Government data shows a more modest picture. The U.S. Census Bureau's Business Trends and Outlook Survey asks firms whether they use AI in any business function (the wording was widened in November 2025 from use in producing goods or services). It found use staying between 17% and 20% of U.S. businesses from December 2025 to May 2026, and higher among larger firms: about 37% of those with 250 or more employees. That is real, but it is a long way from the impression that every company is now running on AI.

When companies move too fast, some move back

The gap is not only about companies being slow. Sometimes the correction runs the other way. In early 2024 the payments company Klarna said its AI assistant was doing the work of about 700 full-time customer service agents. In May 2025 its chief executive, Sebastian Siemiatkowski, told Bloomberg that cost had been too predominant a factor in how the company organised customer service and that the result was lower quality, and Klarna was reported to be recruiting human agents again, while continuing to use AI. The lesson is not that AI does not work, since it clearly handles large volumes of routine contact, but that the last stretch, the exceptions and the moments that need a person, is harder to hand over than a demo suggests.

The gap between what a model can do and what a workplace has actually changed is not a delay before the 'real' disruption. For most people, that gap is where the disruption is being negotiated right now.

What the gap means for you

It is tempting to read a fast-moving capability story and assume your own job changes on the same timeline. In practice, deployment is mediated by budgets, data quality, regulation, managers, and plain organisational inertia — which is why two people with the same title, at different employers, can be on very different clocks.

We would suggest watching deployment rather than demos. The signal that matters for your own work is not the launch of a new model; it is the moment a tool is quietly wired into a real workflow near you — the first draft that now arrives pre-written, the queue that is triaged before you see it. That is the change worth preparing for, and the deployment gap is, for now, the time you have to prepare. Our scores try to track that slower reality rather than the demo cycle.