IT Pro's Rich McEachran asked me for a quote on a question I think most companies are getting backwards: should workers be preparing to manage AI agents rather than just work alongside them. My answer was yes, and the gap between what leadership expects and what employees are ready for is bigger than most organisations realise.
The numbers tell an uncomfortable story
Microsoft's own research, cited in the piece, found 67% of leaders say they're familiar with AI agents against only 40% of employees, a readiness gap that matters because Microsoft also expects roughly 1.3 billion AI agents to be operational by 2028. More strikingly, more than a third of leaders expect employees to be managing agents within five years, while a Workday survey found three-quarters of employees are comfortable working alongside an AI agent but only 30% are comfortable being managed by one. That's not a small gap to close before "agent boss" becomes a normal job description.
Trust is moving in the wrong direction
What I find more concerning than the readiness gap is the trend in trust. Capgemini's research showed confidence in AI agents among senior executives falling from 43% in 2024 to 27% in 2025, even as deployment accelerates. Rolling out agents faster while trust in them declines is a bad combination, and it's exactly the situation a lot of large organisations are in right now, driven by pressure to show AI progress rather than by genuine confidence the systems are ready.
Security is the part getting the least attention
The piece also drew on University of Maryland research showing AI agents are more vulnerable to attack than the underlying language models they're built on, susceptible to jailbreaking, being tricked into pulling malicious files, or being manipulated into leaking private information. That's a genuinely underappreciated risk in the rush to deploy agents: giving a system more autonomy also gives an attacker more surface area to exploit, and most companies are moving faster on deployment than on securing what they've deployed.
What I actually told IT Pro
My contribution to the piece was practical rather than alarmist: workers need to be fluent in how large language models work if they're going to manage systems built on them responsibly. You can't supervise a system you don't understand at some basic level, and that applies whether the AI agent is doing customer support, the way Salesforce restructured 4,000 roles around agents overseen by humans, or something more specialised. Investing in that literacy now is cheaper than fixing the trust and security problems after agents are already embedded in how a company runs.
Read the full piece on IT Pro.