From Copilot to Colleague: How Employees Are Actually Reacting to AI Agents

Employee concern about job loss due to AI climbed from 28% to 40% year over year, even as most organizations have simultaneously accelerated AI rollouts and increased reassurance messaging. Employee "thriving" — a composite measure tracked across more than a decade of workforce research — has fallen to its lowest recorded level. The rollout is happening faster than the trust is being built.

The trust gap nobody's roadmap accounts for

Most AI adoption plans focus on capability: what the tool can do, how accurate it is, how much time it saves. Far fewer account for how it feels to work alongside it. A majority of employees say leaders underestimate AI's emotional impact on the workforce, and only a small fraction of HR leaders currently build that into their AI rollout planning at all.

"The AI is watching me" vs. "the AI actually helps me"

The same underlying technology produces two completely different employee experiences depending on how it's positioned. An AI system framed around monitoring — tracking activity, flagging performance signals, feeding a dashboard someone else reads — reads as surveillance, regardless of intent. An AI system framed around answering an employee's own question, on their own terms, reads as help. The capability can be identical; the trust outcome isn't.

What actually earns trust

A few patterns show up consistently in what makes employees comfortable working alongside an AI agent:

• Transparency about scope. Employees want to know what's automated and what's escalated to a person, not discover it by accident.
• A visible human backup. Knowing a person is one step away, even if rarely needed, changes how an AI interaction feels.
• Consistency. An assistant that gives reliable, sourced answers builds trust faster than one that's occasionally impressive but unpredictable.
• No hidden monitoring function. An assistant that quietly reports employee questions back to a manager for evaluation purposes will lose trust the moment that becomes known.

Designing for colleague, not monitor

The practical takeaway for anyone rolling out an employee-facing AI assistant: design decisions that seem purely technical — where escalation happens, what gets logged and why, how confident the assistant sounds when it isn't sure — are actually trust decisions. Get those right and adoption follows naturally, without needing to oversell the technology itself. See how these decisions show up in Eva's full feature set.

Sources

• Mercer's Global Talent Trends 2026 Report | Mercer
• As Organizations Race to Adopt AI in 2026, Mercer Says Empower Talent and Redesign Work | Mercer