Most organizations have already rolled out an AI tool. Copilot, Gemini, ChatGPT Enterprise, or something similar. Access is there. Licenses are paid. Training has been done.

And yet, usage stays low.

A recent paper by Wharton School researchers studied why people resist AI in the workplace. The paper focused on AI agents, but reading through the findings, it became clear: these are not agent-specific problems.

These are the same frictions showing up every time a team tries to adopt any AI tool. Chatbots. Copilots. Agents. The technology changes. The psychology doesn’t.

The researchers identified three core frictions. Think of them as three walls standing between your team and meaningful AI adoption.

Wall 1: “I don’t think it can do this.”

This is perceived competence. Not whether AI is actually capable, but whether the person believes it is.

Among the findings from University of Basel: people are less likely to use AI that sounds warm and friendly than AI that sounds competent. So if your internal AI tool opens with “Great choice! You’re going to love this!”, you may actually be reducing adoption.

This matters because first impressions shape ongoing usage. If someone’s early experience feels gimmicky or vague, they write the tool off before it gets a fair chance.

What the research suggests breaks this wall:

• Show outcomes, not features. “Processed 531 expense reports today” beats “powered by GPT-5.”
• Use precise numbers. Figures like 83.71% increase trust by 12% compared to rounded numbers.
• Label AI as “still learning.” People chose AI with that label 55% of the time, compared to 43% without it.

Wall 2: “I don’t trust it enough.”

Some might say trust is about capability. It is also about predictability. People need to feel that AI behaves consistently and that they can anticipate its limits.

According to Technical University Berlin, people trust AI more when its limitations are made visible. Not less. More. And they trust it when they see proof of real outcomes, not some benchmark the model just achieved.

What the research suggests breaks this wall:

• Be transparent about what the AI can and cannot do.
• Show results, not architecture. Proof beats explanation.
• Pair AI with a human expert. This alone reduces resistance by 16%.

Wall 3: “I’m not comfortable letting go.”

This is delegation of control. And it carries 26% of the weight in someone’s adoption decision.

Too little AI autonomy feels like extra work. Too much feels like losing agency. The sweet spot is moderate autonomy: “recommend and approve” beats “fully autonomous” almost every time. Full automation reduces ownership by 5.4% and engagement by 21.3%. That is a real cost to culture.

What the research suggests breaks this wall:

• Let people approve before AI acts.
• Start with low-stakes tasks where letting go feels safe.
• Build the muscle gradually. Control is earned, not removed.

Why This Matters Beyond AI Agents

The Wharton paper studied agents. But these three walls show up in every AI rollout. The tool does not matter. The psychology does.

When teams are stuck, the instinct is often to add more training or switch tools. But if the real blockers are competence perception, trust, and control, more training on features will not move the needle.

The better question is: how do we help people feel competent, safe, and in control with the tool? That leads to better onboarding, clearer guardrails, and safer first use cases. Not more prompts.

Ending thought:

The walls are real. But they are not permanent.

Most AI adoption challenges come down to three psychological frictions that have nothing to do with the technology itself. Once leaders and L&D teams recognize what is actually blocking progress, the path forward becomes clearer: show competence through real outcomes, build trust through transparency, and earn control by starting small.

If your organization is working through these frictions, Radiant Institute can help. Our AI enablement programs are designed around these exact dynamics, helping teams move past the psychology and into practical, sustained AI usage. Reach out to explore how we can support your next step.