The Wharton Blueprint for AI Agent Adoption
Published on April 21, 2026
AI agents are here, and they work. The main barriers to adoption are now psychological. Using an agent requires believing it can perform tasks correctly, trusting it with sensitive information, and relinquishing control. Created by Wharton Human-AI Research and Science Says, this Blueprint bridges that gap by drawing on behavioral science in human-AI interaction, guidance from Wharton faculty, and lessons from organizations deploying agents at scale.
The Blueprint identifies and helps overcome three key frictions to AI agent adoption.
Psychological Friction 1: Perceived Competence
Perceived competence is the user’s subjective belief in an AI agent’s ability to perform desired actions, rather than a measure of its technical capability. Elements of this friction include explaining AI agent decisions, communication style, and handling of mistakes.
Psychological Friction 2: Trust
Trust is the user’s willingness to rely on the AI agent despite uncertainty. It reflects a belief that the system will behave reliably, consistently, and in line with the user’s interests. Elements such as showing successful outcomes, being transparent about limitations, and using communication styles that reinforce reliability all influence trust.
Psychological Friction 3: Delegation of Control
Delegation is the user’s willingness to grant an AI agent the autonomy and control required to act on their behalf. Elements include the AI agent’s level of control, the user’s feeling of ownership, and the perceived efficiency of the agent’s actions.
Strategy: Create a Sense of Ownership
Creating a sense of psychological ownership (e.g., naming the agent) can increase adoption by up to 20%.
“We asked business leaders at companies with a combined workforce of over 700,000 employees what challenges they are seeing on the ground. Then, we used the latest scientific research and insights from Wharton’s faculty to understand how to overcome them.”
—Thomas McKinlay, Founder of Science Says
Strategy: Address Control Concerns
Control concerns (e.g., worry of not being able to audit the agent’s process) accounted for 26% of the weight of the decision to adopt AI.
“Technically speaking, using AI agents is already not particularly difficult. What is difficult is the series of uncomfortable decisions that the end user and organizations need to take. Do I give it access to sensitive files? Do I allow it to make payments on my behalf? Without these, AI agents struggle to unlock real value.”
—Stefano Puntoni, Professor of Marketing and Faculty Co-Director, Wharton Human-AI Research
Key Takeaways
Demonstrate Competence
- Avoid giving your AI agent an overly friendly or warm personality.
- Clearly show how the agent is helping the user co-create value together.
- Show detailed explanations of the agent’s process.
- Pair AI agents with a credible human professional and position the AI as supporting the human expert.
Build Trust
- Clearly define the agent’s limitations, and highlight where it fails.
- Show proof of successful outcomes carried out by the agent.
- Use the label “learning” or “improving” to describe your agent.
- Use precise numbers and metrics in all of the agent’s outputs, even if rounded figures seem easier.
Encourage Moderate Autonomy
- Design the agent for a moderate, “human-in-the-loop” mode that requires approvals.
- Clearly communicate the control the user has, showing where they can edit, pause, stop, or reverse the agent’s actions.
- Leverage workflow pressure. For example, highlight full inboxes and upcoming deadlines to encourage agent use.
- Design your agent to keep people attentive, especially in high-stakes outputs.
