AI Fatigue is Real

The Email I Did Not Ask For

It arrived in my inbox like any other piece of corporate communication: a cheerful announcement from my bank informing me that they were rolling out a new AI-powered assistant. Available around the clock, faster than any human agent, ready to answer every question I might ever have. The tone was breathlessly enthusiastic, as if the bank had just discovered fire.

I stared at the email for a moment. Then I felt something I have been feeling more and more often lately: a quiet, creeping exhaustion. Not at the technology itself, necessarily, but at the relentless pace at which it is being pushed into every corner of daily life, whether we want it there or not. There is a word for this feeling that is starting to circulate in conversations about technology: AI fatigue. It describes the weariness that sets in when a tool that was once exciting begins to feel ubiquitous, unavoidable, and at times, deeply unwelcome.

A Bank That Once Felt Different

I chose this bank back in 2019. It was one of those modern, digital-first institutions that seemed to genuinely understand what people were looking for. No bureaucratic queues. No mandatory appointments with a branch manager. No stacks of paper to sign and return. It was built for people who were always on the move, who valued efficiency and clarity, and who perhaps liked the idea of a bank that felt a little less like a traditional bank.

At the time, it was also making serious inroads across several European countries, positioning itself as the bank for people who travel, who live across borders, who do not want to be tied to a single country’s financial infrastructure. It offered genuinely useful features, many of them free, and it did so through a clean, intuitive app. For someone who had grown weary of the slowness and formality of conventional banking, it was a breath of fresh air.

But that email made me pause. Because thinking back on it now, I find myself wondering whether we have gone too far. The very qualities that made that bank appealing in the first place, its sense of responsiveness, its feeling that a real person was on the other end of the line, seem to be quietly disappearing. In their place is an assistant that gets a different name and a different profile photo each time you open a chat, as though the illusion of a human presence were somehow preferable to straightforward honesty about what the system actually is.

The Chatbot That Would Not Step Aside

I have had reason to contact that bank’s customer support more than once. A bug in the app that was preventing a transaction from going through. A payment that had been debited twice from my account. Situations that were not catastrophic, but that required a clear answer and a concrete resolution. Each time, the path to that resolution ran through the same obstacle: the AI agent.

What I discovered is that this agent is not simply a filter designed to route simple queries and pass complex ones to a human. It has been designed to act as a gatekeeper, one that has appointed itself the sole judge of whether your problem is serious enough to warrant a real person’s attention. On several occasions, I found myself completely unable to escalate to a human agent because the system had decided, without any apparent means of appeal, that my issue did not require human intervention.

This is where the problem becomes genuinely serious. AI agents work well as a first line of response for straightforward, repetitive queries. If someone wants to know their account balance, how to change their PIN, or what the bank’s opening hours are, a well-trained AI can handle that efficiently and accurately, drawing on a structured knowledge base to produce a reliable answer. These are what support teams call Level 1 requests, and automating them makes sense. It reduces waiting times, frees up human agents for more complex work, and provides users with instant answers to simple questions.

But the moment a situation moves beyond that structured, predictable territory, the AI’s limitations become apparent very quickly. A payment debited twice involves checking transaction records, cross-referencing with a merchant, potentially initiating a dispute process, and communicating with the user about what happens next and when. A bug in the app may require a support agent to flag the issue to a technical team, log a ticket, and follow up. These are not tasks that fit neatly into a decision tree. They require judgement, flexibility, and accountability. And yet the system was designed in a way that prevented me from accessing the human agents who could have provided exactly that.

The Transparency Problem

There is something particularly troubling about AI agents that are designed to pass themselves off as human. Giving a chatbot a human name and a friendly profile photo is not a design flourish. It is a deliberate choice to obscure the nature of the interaction, and that matters.

When people communicate with a support agent, they make constant, often unconscious decisions based on their understanding of who they are speaking to. They decide how much detail to provide, how formally to phrase their request, how much to trust the response they receive. If they believe they are speaking with a person, those decisions are made on one set of assumptions. If they know they are speaking with a machine, they make them on another. Deliberately blurring that distinction does not serve the user. It serves the institution, by reducing the friction that might arise if users understood that their query was being handled by an algorithm.

Beyond the ethical dimension, there is the practical question of data. Every message sent to an AI agent is processed, stored, and in many cases used to improve the model’s future performance. In the context of a bank, where conversations routinely involve account numbers, transaction details, personal circumstances, and financial concerns, this raises significant privacy questions. How long is that data retained? Who has access to it? Under what conditions can it be used for training? These are not hypothetical concerns. They are the kinds of questions that ought to appear in the enthusiastic announcement emails that banks send out when they roll out these systems. They rarely do.

Summaries Instead of Stories

Banking is not the only place where AI fatigue is taking hold. Search engines are increasingly delivering AI-generated summaries at the very top of results pages. The idea is that instead of clicking through to a source, you can simply read a few sentences at the top of the page and move on with your day.

On the surface, this seems like a convenience. In practice, it is something closer to intellectual shortcutting, and it carries real risks that are rarely discussed openly.

The way a writer structures an argument is not incidental. It is part of the argument. The order in which information is presented, the qualifications that are introduced at particular moments, the context that builds over the course of a piece, these are the things that allow a reader to genuinely understand a topic rather than simply collect a data point about it. When that structure is collapsed into three sentences generated by an algorithm, what is left is a skeleton. It may be accurate in its bare facts, but it is stripped of the texture that makes those facts meaningful.

More concretely, AI-generated summaries are not always correct. These systems can misrepresent the conclusions of a study, omit crucial caveats, or conflate the positions of different sources. Because the summary appears at the top of the page in a clean, authoritative format, many readers accept it without question and never open the original article. The error is never caught. The misunderstanding travels.

There is also a structural consequence for the web itself. When users stop clicking through to original sources, those sources lose traffic, and with it, the advertising revenue or subscription income that sustains them. If AI summaries become the dominant way people consume information, the incentive to produce the in-depth reporting and analysis that these summaries are drawing from begins to erode. We risk creating a system that feeds on the depth and rigour of human-written content while simultaneously undermining the conditions that make such content economically viable to produce.

Are We Becoming Intellectually Lazy?

There is a more uncomfortable question sitting underneath all of this, one worth taking seriously: are we allowing these tools to gradually weaken our capacity for independent thought?

When every article can be summarised before we read it, when every customer service interaction is pre-filtered by an algorithm, when every decision can be informed by an AI assistant before we have had a chance to think it through ourselves, what happens to the cognitive habits we would otherwise be exercising? Reading a long article in full, sitting with complexity, forming a view before receiving a verdict, these are not trivial skills. They require practice. And like any skill that goes unpractised, they can atrophy.

We seem to be perpetually busy. Too busy, apparently, to read a full article, to navigate a bank’s support system, to engage with a problem without AI assistance. But busy doing what, exactly? That is worth asking honestly. Because if the answer is that we are using AI to save time, it is worth asking what we are doing with that time and whether the trade-off is actually serving us.

The concern is not that AI assistance makes individual tasks easier. That is genuinely fine. The concern is what happens when AI assistance becomes the default mode for nearly all engagement with information and services, to the point where operating without it starts to feel unfamiliar. That shift, if it happens gradually enough, may not feel like a loss at all. It may simply feel like progress.

A Tool, Not a Replacement

None of this is an argument against artificial intelligence. These technologies, applied thoughtfully and with genuine transparency, can do real good. Automating repetitive tasks, making information more accessible, reducing friction in processes that were unnecessarily slow: these are legitimate and worthwhile goals. The problem is not the technology itself. The problem is the unexamined assumption that more AI, deployed everywhere, is inherently better than less.

When I think about my bank, I do not want it to roll back its digital ambitions. I want it to be honest with me about what I am interacting with. I want to be able to reach a human being when the situation requires one, without having to outwit a chatbot to do so. I want the institutions handling my financial life to treat my data and my time with the same care and seriousness that I bring to the relationship.

AI fatigue is not technophobia. It is not a rejection of progress. It is a reasonable response to the experience of watching a powerful set of tools being deployed without sufficient care for the people those tools are supposed to serve. We are not asking for less innovation. We are asking for more honesty, more transparency, and more genuine thought about where these systems belong and where, quite simply, they do not.