Did AI adoption level off in 2024 because of regulation, or because of problems inside companies and products?

This explores whether regulation and other sector-specific obstacles explain why AI adoption seemed to stall in 2024. The short answer is that the collection has no study that measures a 2024 plateau, so it can't confirm or rule out regulation as the cause. What it does have points somewhere less obvious: the slowdown may come less from outside rules blocking adoption and more from differences inside firms and from gaps in what AI products can actually do.

Start with regulation itself. The corpus describes rules that lag behind the technology rather than hold it back. EU, US and UK lawmaking moves on a scale of years, while new models arrive every few months, so the law keeps trying to catch up with systems that have already shipped Can regulation keep pace with AI's rapid evolution?. Industry's own proposals for slowing down get a skeptical reading too. Karpf argues that pacing plans and embedded evaluators, modeled on banking supervisors, have little force unless a government can impose penalties Can industry self-regulation slow AI without government enforcement?. If regulation is mostly slow and has few ways to enforce anything, it's hard to see it as the main brake on adoption, at least by the corpus's account.

The more revealing evidence is about uneven adoption. Firms with more exposure to AI replaced online freelance workers with AI tools faster and more cheaply than less-exposed firms. The pattern looks like returns to scale in a company's own AI capability, not a technology spreading evenly through the economy Do firms substitute labor for AI at different rates?. That changes the question. An average adoption rate that looks flat can hide two groups moving apart: firms that have built the know-how keep accelerating, and the rest wait. The barrier may be inside the company rather than in the regulatory environment.

A second clue is that being able to finish a task is not the same as being ready to deploy. In tests of phone-based agents, completing the task, protecting the user's privacy and reusing saved preferences turned out to be separate abilities, and no model was best at all three Do phone agents succeed at all three critical tasks equally?. This is where a domain-specific barrier really does appear, though not in the form of regulation. In a regulated or privacy-sensitive setting, a model that ranks well on task success can still fail the requirement that decides whether a company can use it. Leaderboards built only on task success would hide that gap. Over long tasks, success depended mostly on whether a model kept iterating on feedback rather than giving up early, and most models quit too soon What predicts success in ultra-long-horizon agent tasks?. Shortfalls in reliability like these could slow real deployments regardless of what the law says.

To sum up, the corpus suggests that regulation alone is the weakest explanation for a plateau. Uneven skill inside firms and the gap between benchmark scores and readiness for real deployment fit the evidence better. If you want direct data on adoption rates in 2024 by sector, this collection doesn't have it yet.