Skip to main content

The AI trust gap: We use AI everyday - so why won’t we let it act?

75% of people use AI daily, but 61% don't want it acting alone. Explore the AI confidence gap and how to build trust in autonomous systems.

Connect and support people

In this article: 

  • The trust problem everyone’s talking about
  • AI is already an everyday habit
  • The confidence gap, explained
  • Where comfort breaks down
  • Not everyone sees the gap the same way
  • How to close the confidence gap
  • What this means for the future of work

The trust problem everyone's talking about

Why IT leaders need a new approach to AI

For most people, there’s not a day that goes by that isn’t now punctuated by some kind of interaction with AI. In and outside of work, we use it constantly. It’s integrated into the rhythms of daily life. 

Despite this, we often hesitate to let AI act on its own. The reasons for this are various and overlapping: but the main ones center around security, compliance, and privacy; as well as the consequences of AI for work. 

Add in an at-times dire discourse surrounding frontier AI, and it’s little surprise that there’s an AI confidence gap. On an individual level, this gap means that workers and teams are missing out on many benefits of AI. This then translates into a missed opportunity for companies to get real value and ROI.

The task now is to figure out how to navigate the AI confidence gap and build trust. And the stakes are high. As TeamViewer CEO Oliver Steil put it: 

"The next stage of workplace AI will be defined by trust. Organizations will not succeed by giving AI unlimited authority. They will succeed by building systems that can act safely, transparently, and accountably, while helping people and businesses perform at a higher level."

AI is already an everyday habit

To understand this situation a bit more, we spoke to 4,200 IT managers and employees worldwide to create the report, Path to the Autonomous Digital Workplace. 

When it came to AI adoption, the report’s findings confirmed what we're already seeing in daily life. Adoption is already very high, and experience with AI is largely positive: 

  • 75% of respondents use AI at least once a day.
  • 97% identify at least one workplace benefit (only 3% report none).
  • 64% describe their experience with AI as positive. 

So, this is hardly an adoption problem: most of us are using AI in some shape or form every day. The question is: are we getting real value from AI? 

Our research implies that adoption, while buoyant, is largely confined to more limited use cases, and by that, we mean ones that demand a high level of human oversight. In these cases, the payoff of using AI is minimal – which makes sense as managers worldwide struggle to demonstrate the ROI of AI initiatives. 

And here is where the confidence gap shows up. Because real value starts only happens when we move from using AI, to using it well, which, at least in productivity terms, means letting it act with less human oversight. 

So, what's the takeaway? AI has crossed from novelty into daily routine – which makes the hesitancy that follows even more striking. 

The confidence gap, explained

Logically speaking, you’d think that more frequent usage would help to build trust, but that doesn’t seem to apply to AI. So, while adoption is high, our research shows that we still have a problem with granting it authority or autonomy. 

And the numbers here are quite conclusive. More than three in five (61%) prefer AI to take no independent action. 43% want AI to provide recommendations as they make the final decision, while 18% prefer to make decisions without AI recommendations at all. 

The remaining 39% are open to AI taking action – but most still want human approval. 

Respondents also reported an AI verification burden, with more than half (56%) often or always verify AI outputs before relying on them, spending an average of two hours per week checking AI-generated work. 

So, while we might be increasingly heavy users of AI, we're still very cautious about how we're using it. We want to be consulted before it acts, and when it does, we don't really trust the output and spend considerable time checking it. 

You might think about it this way: we're micromanaging AI. And this is a tendency that applies to IT managers and workers across the world. 

This micromanaging stems from the AI confidence gap, which means that we're failing to get the real benefits from AI. And, as already stated, this is also very understandable. Many people fear the consequences of AI on IT as a whole; some others, the very real implications for cybersecurity and privacy. 

There are numerous reasons to be cautious around AI. The problem is that this caution is also arising in situations where it's feasible to grant AI more autonomy. And that's the part that needs to be addressed. 

First, let's look at where confidence in AI breaks down.

Where comfort breaks down

When it comes to AI, people are generally more willing to delegate routine, contained activities, where the potential consequences are limited and easy to reverse. But we get uncomfortable when AI is given access to sensitive information, asked to exercise greater judgment, or allowed to make decisions with wider consequences. 

Our report's findings reflect this tension. While 57% of respondents are at least moderately comfortable with AI handling sensitive or confidential information, 22% are not comfortable with it at all. In other words, the same task that feels manageable to some is a 'hard no' for others.

But autonomy doesn't necessarily mean losing visibility or control. In fact, 40% want AI to detect and fix issues early while keeping them informed. And 70% are comfortable with AI acting autonomously, provided they can step in when needed. 

That distinction matters. People are not necessarily rejecting autonomous AI outright. They're asking for autonomy to operate within clear limits. They want to know what the system is doing, where it has access, and whether they can intervene if something goes wrong. 

So, the confidence gap is not simply about whether people trust AI. It depends on the task, the level of risk, and how much control remains in human hands. 

Not everyone sees the gap the same way

Quote from Teamviewer DEX perspective survey 2025

There's another dimension to the confidence gap: the people leading AI adoption do not necessarily experience it in the same way as the people most affected by it.

Managers report more benefits from AI and describe their experiences more positively than non-management employees. IT decision makers are also more frequent users and more willing to let AI act. In fact, 99% of IT leaders say they would be comfortable delegating at least one task to AI. 

There are good reasons for this confidence. IT leaders tend to have a clearer understanding of how systems work, where guardrails can be applied, and what oversight is available. They're also more likely to see the wider organizational benefits of automation. 

But employees may experience the risks more personally. If an autonomous system makes a mistake, it may affect their work, their data, or a decision that concerns them directly. And they may have less visibility into why that decision was made or how it can be challenged. 

This creates a potential disconnect. Those responsible for introducing autonomous systems may focus on efficiency and scale, while employees are more concerned with control, accountability, and what happens when something goes wrong. 

Closing the confidence gap therefore requires more than technical safeguards. It means involving the people affected and giving them visibility into how autonomy will work in practice.

How to close the confidence gap

Quote from Teamviewer DEX perspective survey 2025

If the confidence gap comes down to control, risk, and visibility, then simply giving AI more autonomy isn't the answer. Trust has to be built. 

IT is a natural place to start. The shift is already underway from reactive IT, where teams respond to problems after they happen, towards proactive and predictive systems that can spot issues earlier. The next step is autonomous IT, where systems can resolve certain problems themselves. 

But autonomy doesn't have to mean operating invisibly. People want to know what changed and why, and to be able to step in when necessary. This is why IT is leaving reactive firefighting behind and moving towards proactive, autonomous operations. 

As TeamViewer Chief Product and Technology Officer Mei Dent puts it: "The next step is designing systems that know when to act, when to wait, and when to bring people into the decision." 

That requires clear boundaries around what AI can do. Our research points to five pillars for getting this right: security, decision quality, control, reversibility, and accountability. Together, they create a framework where greater autonomy doesn't require people to simply surrender control. 

And crucially, trust can grow over time. As we've explored in this report, every successful fix can give people greater confidence in letting systems act more independently next time. 

Autonomy earns trust when people can see what changed – and undo it if needed.

What this means for the future of work

So, what do we take from all of this? For one thing, it's clear that there's a marked disconnect between the seismic shifts being predicted in today's AI discourse, with actual perceptions and expectations of AI within IT. 

For example, IT decision makers expect an average of 39% of digital workplace services to run autonomously by 2030. But that means 61% of digital workplace services will not be autonomous. 

Moreover, only 35% are mostly or completely ready to trust AI to act more autonomously; 32% are ready only in limited, low-risk situations; 31% are slightly ready or not ready at all. 

By 2030, 33% expect AI to assist while they continue doing most tasks; 28% expect to delegate more tasks and focus on higher-value work, and only 2% expect their role to be eliminated. 

All in all, this points to a future that, in the words of Mei Dent, is "human plus AI, not human versus AI."

"The path to autonomy is not about handing over control. It is about giving trusted systems the authority to act within clear boundaries, so people have more time and energy for higher-value work."

Rebecca O' Dwyer
Rebecca O’ Dwyer

Senior Content Marketing Manager at TeamViewer

Berlin-based Rebecca is the editor of the TeamViewer Blog. With extensive experience in long-form writing and editing, she loves nothing more than translating TeamViewer expertise into useful and engaging content.

Want the full picture?

Read the Path to the Autonomous Digital Workplace report for all the data and the five pillars of trusted autonomy.