Recently, I experienced what should have been a very simple interaction with a delivery company. Instead, it became an unexpectedly revealing example of where modern automation quietly begins to fail people.
I missed a delivery from Royal Mail, and a redelivery card was left through the door with instructions to rearrange the delivery online. Straightforward enough in principle.
The problem was the handwritten reference number on the card.
The handwriting was almost impossible to read. I tried entering different combinations multiple times, carefully checking whether certain letters were actually numbers, but every attempt failed. The online system repeatedly informed me that the delivery reference was incorrect.
At that point, the issue was no longer the missed delivery itself. The issue was that the system had no ability to handle ambiguity once the standard process broke down.
So I called customer services.
What followed felt increasingly familiar within modern digital systems. I was routed into an AI-driven automated phone system designed to efficiently process predefined requests. The system could handle tracking updates, delivery instructions, and routine enquiries. What it could not handle was my actual situation.
There was no option that matched the problem I was trying to explain.
The system could not understand that the issue itself originated from flawed human input into the process. More frustratingly, there appeared to be no meaningful route to speak to a human being capable of interpreting the situation and resolving it.
Eventually, I gave up and cancelled the delivery entirely.
The vendor lost the sale.
The parcel was returned.
Time was wasted across multiple parties.
Customer trust declined.
All because a highly efficient system had no meaningful mechanism for human adaptability once the workflow failed.
Ironically, the problem was not technology itself. The problem was the absence of meaningful human flexibility inside the technology.
That experience stayed with me because it reflects a much larger issue now emerging across modern institutions as agentic AI becomes increasingly embedded into operational systems.
Artificial intelligence is rapidly moving beyond simple automation into something far more influential: systems capable of acting autonomously, making decisions, initiating workflows, coordinating tasks, and shaping human experiences with increasingly limited human intervention. This is the emerging world of agentic AI.
Across healthcare, education, welfare, finance, customer services, human resources, and public administration, organisations are exploring how autonomous AI systems can reduce workload, improve efficiency, accelerate operational delivery, and lower costs. In many cases, the enthusiasm is understandable. Institutions are under enormous pressure. Demand continues to outpace human capacity. Administrative complexity is increasing. Staff burnout is rising across multiple sectors.
Within this environment, agentic AI appears not simply attractive, but necessary.
Yet an important question is beginning to emerge beneath the excitement:
What happens when institutions become highly efficient, but progressively less human?
This is the hidden risk of agentic AI.
The danger is not artificial intelligence itself. The danger lies in what institutions choose to optimise, what responsibilities they delegate, and whether human experience remains visible inside increasingly autonomous systems.
Because while AI can automate process, it cannot fully comprehend human complexity.
And the moment institutions forget that distinction, administrative dehumanisation begins.
Subscribe to continue reading
Become a paid subscriber to get access to the rest of this post and other exclusive content.
