The full findings are worth your time: anthropic.com/features/81k-interviews
Here’s what stood out for anyone evaluating AI, Gen AI, or Agentic AI for their business.
What should comfort you
The #1 use case people reported? Professional excellence. Nearly 1 in 5 users said AI helps them shed routine work so they can focus on higher-value thinking. That is exactly the ROI story enterprise buyers are trying to make. And 81% of respondents said AI had already delivered a meaningful step toward their vision. This is not hype. It is reported experience.
The productivity gains are also more personal than expected. A software engineer in Japan described leaving work on time to pick up his daughter from daycare for the first time. A healthcare worker said the cognitive load of documentation had lifted, giving them more patience and presence with patients and families. Your employees are people. When AI gives them time back, the impact goes deeper than headcount math.
What might surprise you
The #1 concern was not job loss. It was unreliability. 27% of respondents flagged hallucinations, inaccurate outputs, and the verification burden as their top worry. One researcher described getting caught in what they called a slow hallucination, internally consistent, confident, and subtly wrong in ways that compounded over time. Nearly half of all lawyers reported hitting AI unreliability firsthand. One respondent put it plainly: an AI that sounds certain but is often wrong does not free your attention. It creates a permanent fact-check tax.
For enterprise deployments, this is the adoption blocker hiding in plain sight. And it points to something the study does not say explicitly but every data professional will recognize: unreliable AI output is usually a data problem upstream. Garbage in, garbage out has not changed. It has just gotten faster and more convincing.
The infrastructure behind trustworthy AI
If people are going to trust AI enough to act on it, the foundation has to hold. That means three things working together.
Data quality is the starting point. AI does not manufacture insight. It surfaces patterns from what you feed it. If your underlying data is inconsistent, stale, or poorly defined, the AI will reflect that back with confidence. Cleaning and standardizing data before it reaches an AI layer is not optional prep work. It is the difference between a tool that accelerates decisions and one that quietly introduces error at scale.
Data governance is what makes AI enterprise-ready. Knowing which data sources are authoritative, who can access what, and how outputs are audited is not a compliance exercise. It is how you give your teams permission to trust what the AI tells them. Without governance, every output carries an asterisk. With it, AI becomes something people can actually build workflows around.
Guardrails at the application layer close the loop. Even with clean data and strong governance, agentic AI systems need defined boundaries. What actions they can take autonomously, where human review is required, and how errors get flagged and corrected. The more capable the AI, the more important it is to design these constraints deliberately rather than reactively.
What the study also gets right about adoption
The second surprise in the data: the people seeing the strongest gains from AI are independent workers and entrepreneurs, not institutional employees. Nearly half of entrepreneurs reported real economic empowerment, compared to 14% of traditional employees. That gap should prompt a hard question for every enterprise leader: are we actually removing the friction that would let our people use AI the way an independent worker and entrepreneurs would, or are we deploying tools and calling it transformation?
The organizations that close that gap tend to have one thing in common. They treated the data foundation as part of the AI investment, not a separate workstream that would get to eventually.
The bigger picture
The study found that hope and fear do not divide people into separate camps. They coexist in the same person. Your employees will feel this tension too. The organizations that win with AI will not be the ones who picked the right vendor. They will be the ones who designed for trust, removed the right friction, and built the infrastructure that lets people act on what AI tells them without second-guessing every output.
If you want to talk about where your data foundation stands and what it would take to get your AI deployment to that level of trust, let’s talk.
Vertex Tower, Plot no-
C-33, 4th Floor, Phase 2, Industrial Area, Sector 62, Noida, Uttar Pradesh, 201309