Trust is becoming a constraint on enterprise AI adoption. According to Harvard Business Review, 59% of enterprise organizations use agentic AI, yet only 9% have turned it into autonomous workflows.
Employees remain wary of giving agents meaningful authority over sensitive work, limiting how much value the technology can deliver.
The financial stakes are significant. Only 27% of middle managers report seeing real ROI from AI deployments, while companies expect to invest an average of $202 million in AI over the next 12 months.
Closing that gap requires earning enough trust for employees to delegate meaningful work without giving up control over consequential decisions.
Why It Matters: Agents deliver limited value when employees restrict them to chatbot-like tasks because they are uncomfortable granting meaningful permissions. Research reviewed by the authors suggests that trust depends in part on whether employees understand an agent’s limitations and how much authority it has. The goal is calibrated trust, giving agents enough autonomy to handle useful work while keeping people in control when the consequences are significant.
- Be Clear About Where Agents Can Fail: Telling employees about known weaknesses can increase trust instead of damaging it. Across two experiments, disclosure made an AI appear up to 14.6% more transparent and helped people work with it up to 7.2% more effectively, even though the underlying AI was unchanged. Generic disclaimers about possible errors provide little guidance in day-to-day use. Organizations should identify the tasks where an agent may struggle and make clear when employees should verify its work.
- Make Competence Visible: Research cited by Harvard Business Review found that people were less willing to use AI perceived as friendly and warm than AI perceived as competent. Concentrix CEO Chris Caldwell reports a similar response from customers who become frustrated with overly polite technology that fails to get work done efficiently. Agents can establish competence by explaining what they did and why they made a particular choice, giving employees a clearer basis for deciding when to rely on their work.
- Connect Actions to the Employee’s Goals: People accepted AI recommendations 54% more often when the AI demonstrated that it understood their larger objective. A budgeting agent, for example, could explain how a recommendation relates to next year’s headcount plan, giving the employee context for why the recommendation matters. This approach helps agents connect individual actions to goals employees have already established.
- Keep the Agent in a Supporting Role: People initially perceived 13.8% lower privacy risk from an AI agent compared with a person doing the same task. That perception changed when they were reminded that the agent could make decisions on their behalf. Harvard Business Review recommends describing agents as tools that “assist” or “support,” especially when sensitive information is involved. Microsoft’s “Copilot” naming follows the same principle by presenting AI as technology that works under human direction.
- Give Agents Autonomy With Clear Limits: Agents need enough freedom to perform useful work without removing human control over consequential actions. Research cited in the article supports a moderate level of autonomy. ServiceNow’s “control tower,” for example, lets organizations set guardrails, monitor activity, and override decisions. Workato CTO Adam Seligman describes a similar model where an agent can draft an email without permission to send it or recommend an inventory move without executing it. Requiring confirmation before consequential actions allows agents to handle routine work independently while preserving human control when the cost of an error is higher.
Go Deeper -> To Adopt AI at Scale, Employees Need to Trust Agents – Harvard Business Review
