Do Your Employees Know They’re Working Alongside AI Agents?
Awareness Gap: Many employees are unaware of AI agents operating in their workplaces, leading to distrust and confusion.
Transparency Importance: Governance frameworks assume human oversight, but lack of employee awareness risks accountability failures.
Trust Erosion: Many workers feel AI systems are unreliable, as 62% express concerns about AI agents’ trustworthiness.
Employee Involvement: Disclosure is crucial; employees must understand AI's role to effectively oversee and engage with systems.
Training Deficit: Only 38% of organizations provide training on AI tools, amplifying workers' feelings of obsolescence.
Most employees don't know what they're working alongside.
Agentic AI has moved well past the pilot stage and is now routing customer service cases, flagging performance outliers, scheduling interviews, and summarizing meetings before anyone has logged off.
In a growing number of organizations, it's drafting responses and making low-stakes decisions on behalf of workers who haven't been told what's happening.
Keep Reading—and Keep Leading Smarter
Create a free account to finish this piece and join a community of forward-thinking leaders unlocking tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
The deployment has outpaced the disclosure. According to a Cloud Security Alliance survey published in April, 82% of enterprises discovered previously unknown AI agents running in their IT environments within the past year, many of them appearing multiple times. That's an IT visibility problem. The employee awareness problem runs deeper.
Workday's 2025 global research, which surveyed nearly 3,000 decision-makers across North America, APAC, and EMEA, found that only 24% of employees are comfortable with AI agents operating in the background without their knowledge.
We can't call it a fringe concern when three out of four workers are drawing a clear line, and most of them are doing it without full information about where that line actually falls in their own workplace.
Transparency and Shared Assumptions
The ethical case for closing that gap is straightforward. Workers have a legitimate claim to understanding the systems that shape their task assignments, workload, and in some cases their performance evaluations. But the operational argument may be more legible to the executives who need to hear it.
Governance frameworks, such as NIST's AI Risk Management Framework, the EU AI Act's transparency requirements, and most enterprise AI policies drafted in the last two years, are built on a shared assumption that humans remain in the loop.
Not just executives or compliance teams, but the people doing the work. Human oversight means that when an agent makes an error, misroutes a case, applies a flawed rubric, or drifts from its intended behavior, someone catches it. That someone is almost always a frontline worker, close enough to the output to recognize when something is off.
When employees don't know an agent is acting, that oversight disappears structurally. The architecture of accountability depends on awareness that isn't there.
What Happens When They Find Out?
The breakdown, when it comes, tends to be personal. About eight months before speaking with People Managing People, Mike Rolfe, VP of Product at Outbuild — a construction scheduling platform — deployed an agentic system into customer success operations that could autonomously monitor account health, schedule calls on behalf of CSMs, and draft renewal risk reports without anyone triggering them.
Leadership was informed, but the broader customer service team wasn't, partly because internally everyone kept calling it an "automation" rather than what it actually was.
A CSM walked into a customer call and the customer mentioned a meeting they’d already scheduled, which the CSM had never set up.
The trust damage inside the team took weeks to repair. His rule now is that the moment an AI system takes an action that another person sees, the person whose name is attached to that action has to know about it before it happens.
His team also stopped using the word "automation" entirely, replacing it with plain descriptions. For example, "the system will book meetings for you and notify the customer under your name".
This is because in construction culture especially, vague tech language doesn't create informed consent. It creates the conditions for exactly the kind of moment his CSM experienced.
Accountability is a Human’s Responsibility
Steven Betito, COO and data protection officer at Elestio, has also seen this firsthand.
When his team integrated agentic AI into daily operations, they built disclosure in from the start. Each agent had a documented role and was introduced to the team explicitly as an AI, which meant no one was ever uncertain about what they were working with. What they didn't build early enough was a clear escalation structure.
In the first few months, we did not formalize escalation paths. A human teammate would email an agent with an ambiguous request, the agent would interpret as best it could, and the output sometimes landed in production without a clear human reviewer.
The fix was straightforward once they identified the problem. Every agent action above a defined threshold now requires a named human approver, but the lesson held. Transparency about what an agent is doesn't automatically produce accountability for what it does. Those are separate problems, and they need separate solutions.
Trust is Nowhere to Be Found
Asana's 2025 Global State of AI at Work report, which surveyed more than 2,000 knowledge workers in the US and UK, found that 62% view AI agents as unreliable.
More telling is what sits underneath that number, as 82% of workers said proper training is essential to using AI agents effectively, but only 38% of organizations have provided it.
The issue here is not so much that workers are rejecting AI agents, they're navigating a mismatch between what they've been handed and what they've been taught and they’re feeling less and less relevant as deployment accelerates.
When employees discover through a colleague, a process change, or an outcome they can't explain that AI systems have been running in their workflows, the reaction isn't usually curiosity. It's the specific kind of distrust that comes from feeling managed rather than included. And that distrust doesn't stay contained to the AI question.
Part of the problem is definitional. "Agentic AI" hasn't made it into most employee handbooks or manager training programs, even as the tools themselves have. Many organizations have deployed agents under vendor product names — workflow assistants, CRM integrations, scheduling platforms — that don't surface the underlying autonomy.
An employee following a case routing recommendation may have no idea that routing was made by an AI agent rather than a manager or a static rule.
WalkMe's 2025 workplace survey found that 78% of employees use AI tools not provided by their employer, and nearly half have avoided disclosing their AI use at work to sidestep judgment. Shadow adoption and institutional opacity are feeding each other.
Fixing this isn't a one-time communications exercise. What works is building disclosure into deployment by explaining what a system does and doesn't do before it goes live in a team's workflow, creating ongoing visibility into where agents are acting, and giving workers a clear channel to flag when something looks wrong.
That last part matters more than most AI rollout plans acknowledge.
Employees aren't just the audience for disclosure. They're the primary error-detection layer for any agentic system operating in their domain. When they know what's running and understand what to do when it misfires, governance functions the way it was designed to. When they don't, those frameworks are theoretical.