A Sombra survey of 2,000 professionals found that 87 percent say they personally get value from AI tools — but only 55 percent say their organisation does. The gap between individual benefit and company-level impact is where AI projects go to die.
The survey, which Sombra is calling the sAIbotage report, found that 70 percent of companies face active internal pushback on AI initiatives, and one in three projects are already being delayed by it. The reasons are not subtle: 48 percent of respondents say they don't trust AI outputs, and 47 percent are concerned about what automation means for their role.
The layoff context gives the numbers additional weight. Meta announced 8,000 job cuts this week, joining Microsoft, Amazon, and Oracle in a quarter that saw roughly 80,000 technology sector layoffs, with many companies attributing the reductions to AI-driven productivity. The Sombra data suggests that many of the same companies doing the cutting are simultaneously struggling to get AI out of pilot.
A lot of companies are trying to adapt their processes to AI, but there are several obstacles: data isn't always ready, employees resist the change, and ROI is still hard to measure. That's why so many projects can't move past the testing phase and make a real difference in the business.
The scaling gap in the data is particularly striking. Individual contributors see the upside — they use AI to move faster, draft better, analyse more. But the productivity doesn't aggregate in the way that would show up in company-level output or justify restructuring decisions. Brui's prescription is communication rather than technology: "The companies that move faster are usually the ones that explain where AI fits into daily workflows, what decisions it is meant to improve, and how risks will be managed."
For enterprise IT leaders, the implication is a familiar one: adoption is a change management problem before it's a technical one. The tools exist. The organisational scaffolding — policy clarity, training, transparent governance — often doesn't.
