AI is surrounded by a strange mix of excitement and concern. On one hand, it is already part of everyday work, helping with writing, analysis, and planning. On the other hand, people still talk about hallucinations, data privacy risks, and what it might mean for their jobs.
To better understand what people are actually concerned about, the professional services platform Productive surveyed 256 agency and professional services roles to examine how they use and perceive AI agents.
The results show a clear tension. Adoption is not the issue. Most people are already using AI in some form. But at the same time, many still have real concerns about how it works and where it can go wrong. Those concerns vary, but some come up more often than others, pointing to shared patterns in how people think about AI today.
1. AI Can Be Wrong, and You Still Have to Check It
Accuracy and reliability were the most common concerns in the research. People pointed to hallucinations, failure to account for the full context, and outputs that still need to be checked before anyone can trust them. With AI already in widespread use for a while, it’s likely experience that makes people more cautious.
Even when AI is useful, people still feel they have to review everything closely, which limits how much work they are willing to hand off. In other words, the concern is not just that AI can be helpful and imperfect at the same time, but that the need to verify its output doesn’t seem to go away.
2. What Happens to Your Data Still Feels Unclear
Data privacy and security remain major concerns, especially in work involving client information, internal documents, or sensitive financial data. People want to know what happens to their data once AI is involved, including which model is processing it and who can access it. That may sound technical, but it shapes whether the tool feels safe to use at all.
That concern shows up outside the workplace, too. A 2025 Consumer Trust Survey commissioned by Relyance AI found that 82% of respondents see AI data loss-of-control as a serious personal threat, while 81% suspect companies are already using their personal data for undisclosed AI training.
There is also concern about over-permissioning, where AI tools surface or access information they should not. For example, the research quotes a respondent saying that an over-permissioned agent might accidentally share a confidential salary spreadsheet because someone asked it to “summarize recent attachments.”
3. AI Taking Action Without Approval Feels Risky
Unintended or irreversible actions are a key concern, especially when AI moves beyond generating output and starts doing things on our behalf. Interestingly, this concern is more present in some roles than others. In Productive’s research, individual contributors were especially sensitive to AI taking action without explicit approval, likely because they feel more exposed when a system acts on their behalf.
The risk isn’t limited just to incorrect output. It’s the possibility that a message is sent, a change is made, or a decision is triggered before anyone gets the chance to intervene. Once an action has been taken, the damage may be harder to contain or reverse.
4. Relying on AI Too Much Could Weaken Human Skills
Erosion of human skills is one of the more familiar AI concerns, even if it was not the loudest one in the research. People have been asking some version of the same question since generative AI entered everyday work: if AI keeps taking over more thinking, writing, and problem-solving, do people get worse at doing those things themselves?
Several respondents raised that same concern. In roles where judgment develops through practice, offloading too much cognitive work could weaken the habits people need to think problems through, express ideas clearly, and solve tasks independently.
5. AI Can Flatten Voice and Make Output Sound Generic
AI-generated content can start to feel the same, even when the output is technically correct. Many people have already noticed the familiar patterns: similar tone, similar phrasing, similar structure.
Over time, those patterns make work feel less distinctive and blur the individual voice. That becomes a bigger issue in creative and client-facing work, where sounding generic is not just a style problem. It can make the output feel less thoughtful, less personal, and easier to dismiss.
6. Job Security Questions Have Not Gone Away
Job security is still on people’s minds, even if it isn’t the first concern they raise. More than panic, there is a sense of uncertainty; questions about how roles might change, which responsibilities could lose value, and what these shifts could mean for long-term stability.
Although this concern is less vocal than others, it’s no less real. It lingers in the background of adoption, shaping how people think about where AI fits and how far they want it to go.
7. The Ethical and Environmental Impact of AI Raises Real Questions
Beyond day-to-day use, broader concerns continue to surface, particularly among senior roles. People are asking about the carbon footprint of large-scale AI, the methods behind its training, and the trade-offs that shape the outputs they receive.
There is also a more practical tension underneath it all: whether automating work that feels repetitive or low-value actually creates value, or just replaces effort without improving outcomes. That makes this less about immediate risk and more about whether the direction itself feels justified.
What These Concerns Say About AI Adoption
People are already using AI. That part is not in question. But using it and feeling comfortable with it are not the same thing.
What comes through in these concerns is not resistance, but uncertainty. AI is useful, and often impressive, but also unpredictable, hard to fully understand, and sometimes difficult to control. That mix is what makes people cautious.
For now, that tension remains unresolved. AI is becoming part of everyday work faster than people can decide what they are comfortable handing over to it. The technology is moving into the workflow, but the expectations, boundaries, and trust around it are still catching up.
