Executive AI agent use does not prove deployment readiness

Executive AI-agent use sets an incomplete benchmark

The survey report found that leaders use nearly double the number of AI agents per week that workers use. Executive use also exceeds employee use in generative AI, which produces content, but the gap is wider for agentic AI, in which software agents can access information and carry out actions. Uneven adoption threatens enterprises’ ability to extend these tools across their organizations. Yet executive usage provides an incomplete benchmark because workers also report spending time making agent results usable.

This extra work changes what higher usage means for deployment decisions. Workers must choose tools, manage outputs and resolve conflicts, so more agent activity can create more operating demands. Differences in familiarity may contribute to lower worker trust, but workers’ direct experience also gives their caution a practical basis. Leaders need to understand that experience before treating increased usage as evidence that a deployment is ready to expand.

Workers carry the coordination cost

The operating demands begin with choosing and managing agents. In the survey, nearly half of workers using them said they spent about 10% of their working time switching between tools, checking which agent to use, managing outputs or dealing with conflicts. This estimate describes the time those workers spent; it is not a workforce-wide average. Even when an agent completes its assigned task successfully, people still need to coordinate the work before its results become useful.

Coordination becomes harder when agents cover the same ground. Among those surveyed, 38% said more than one-quarter of agent work was duplicated or overlapping. The first percentage describes respondents; the second describes the share of agent work they considered redundant. Overlapping tasks leave people to determine what has been completed and which results to use, so more agent activity can add little useful work.

Beyond overlap, workers encounter problems with the information and decisions behind an output. “Workers who work alongside agents day-to-day are more likely to notice when an agent acts on incomplete context, surfaces conflicting outputs, or escalates something that needed human judgment,” the survey report said. Incomplete context can make a clear-looking answer unsuitable for the task, while conflicting outputs require someone to choose which result should guide the work. An escalation involving human judgment creates another decision for the person receiving it.

These experiences help explain why the survey report links trust to use and adoption, and describes workers’ response this way: “Their caution is calibrated, not uninformed.” Workers’ hands-on checking shows the effort required to use agents. It does not establish an explicit rejection of AI, justify every hesitation or mean every executive assessment is misplaced. It gives leaders specific operating problems to investigate before asking workers to increase usage.

Trust shapes agent authority and oversight

Workers’ experience also affects the authority they are willing to give agents. Leaders are more likely than workers to trust agents both to access information and to complete actions autonomously, meaning without a person directing each action. That authority determines what information an agent can reach and what it can do on the organization’s behalf. Usage counts reveal little about whether those permissions are appropriate.

Permissions matter because both leaders and workers across industries continue to report concerns about security and agent overreach. Overreach occurs when an agent acts beyond its intended authority, so usefulness alone cannot settle which actions it should be allowed to take. Trust therefore needs to inform access and action permissions as well as adoption decisions. A team can find an agent useful while still requiring limits on its authority.

Those limits also depend on oversight, yet the survey report found that about one in four AI agents operates unmonitored in enterprise environments. Monitoring lets an organization observe how an agent uses its authority to act independently. The finding therefore makes worker concerns relevant to governance: leaders need to consider permissions and monitoring together. Expanding access or autonomy changes the operating requirements even if the number of agents stays the same.

A separate Workato report adds a measure of executive concern about possible failures. Workato has a commercial interest in enterprise AI-agent adoption, and nearly two-thirds of executives in its report said they expected an AI agent in their environment to cause a public security or compliance incident within the following year and a half. This measures an expectation about a future event. It warrants attention to incident risk, while the leader–worker trust findings address a different question: how much authority respondents are willing to grant agents.

Measure whether important work gets finished

Authority and oversight set conditions for deployment, but assessing its value requires looking at the work people and agents complete together. Bose argues that adding agents alone will not create organizational value: people and agents need a common plan, enough context to perform the task and enough clarity for workers to decide what to do with the output. Each condition addresses a problem workers encounter in practice. Worker feedback can therefore help leaders assess whether the deployment supports useful work.

A common plan connects individual agent assignments to human responsibilities and the team’s objective. This addresses the overlap workers report, because separate tasks can each succeed while contributing little to the same goal. Management needs to determine how those assignments fit together. The test then becomes whether the combined effort advances the work the team is responsible for completing.

A shared objective still needs enough context to guide execution and review. An agent can act on the information available to it while lacking information the job requires, and workers need enough context to judge whether its result is usable. Bose’s emphasis on a clear next action extends this requirement to the handoff: the output should give the receiving worker enough information to proceed. When conflicts or judgment calls remain, the human task they create belongs in the deployment assessment.

The next action matters because it connects an agent’s output to completion of the team’s work. “CIOs should move away from measuring agent activity and start measuring whether work actually gets finished, and what is the most important work for that team and the business,” Bose said. Leaders must therefore identify which work matters before interpreting completed agent tasks as success. A high task count has limited value when priority work remains unfinished.

Completion measures also need to account for the effort required to get there. Before authorizing wider use or greater autonomy, leaders should examine how much checking, coordination and judgment workers must contribute to make the existing deployment function. Addressing those demands may improve readiness without eliminating the executive–worker adoption gap. The decision to expand should include whether the organization can sustain that human workload as agent use grows.

Key executive takeaways

Assess readiness beyond executive usage: Leaders use nearly twice as many AI agents per week as workers. Deployment owners need to examine workers’ coordination demands and trust concerns before treating executive adoption as a benchmark for expansion.
Account for coordination costs: Nearly half of surveyed workers using agents reported spending about 10% of their working time choosing tools, managing outputs and resolving conflicts. Process owners need clear task assignments and handoffs to reduce overlap and make results usable.
Review permissions and monitoring together: About one in four enterprise AI agents operates unmonitored, while workers report lower trust than leaders in granting agents access and autonomy. Security and deployment owners need to review monitoring coverage alongside each expansion of agent authority.
Measure completed priority work: CIOs need measures that connect agent activity to important work finished and the human effort required to finish it. Expansion decisions should account for whether workers can sustain the checking, coordination and judgment that wider deployment requires.