Merijn te Booij points out a new risk of AI. “We have been managing technological hypes for thirty years without truly improving CX.” One of the outcomes of even more technology and automation: even more interactions.

 

I am not a technologist. I want to make that clear right away. I am a graduate war lawyer, trained to provide interpretation in the field of international law, to find the flaw in legal arguments, and to remain healthily skeptical of anyone who tells you that something is simpler than it looks.

Dutch lawyers always look at my Master's degree with a bit of cynicism. A bit like when I wasn't musical enough to play the recorder and had to continue on the xylophone as one of the three. But my specialization in landmine issues turned out to be a great fit for the world of customer experience. A world currently gripped by a collective, industry-wide infatuation with artificial intelligence, so intense that it would make a teenager shy. Hail Agentic!

I also have two children studying at university. One is studying philosophy. The other English. Neither of them is particularly impressed by what the AI ​​industry is doing at the moment, for reasons I will return to shortly. I am, in their magnanimous judgment, “passionate but not dangerous”.

AI is a minefield

Making the right choices in terms of AI is not much different from walking through a field of landmines. Full of pitfalls and explosive errors. Not lethal, but still. There is an awful lot (a word chosen intentionally here) of technical information available, nothing, however, that actually neutralizes the landmines. We have been managing technological hype for thirty years without truly improving CX.

The only common thread running through CX that actually works is the agent, the human being. And that is why we have been trying to replace or help them for thirty years. Paradoxical. “Rethink not evolve” is my mantra at the moment, but rethink within long-existing frameworks. In other words, laws.

There are a considerable number of laws, not legal in the strict sense, that have been applicable to CX and the current state of affairs for years. It can be enlightening to know them and to make choices with those frameworks in mind. In this column, I will limit myself to five. None of them require a computer science degree.

Amara's Law: the hype has a name

Roy Amara of Stanford gifted us this gem: we overestimate the impact of technology in the short term and underestimate it in the long term. This is undeniably true, especially for AI. Every AI vendor presentation this year is Amara's Law in PowerPoint form. 'Outcomes' is the new magic word. For example, 'contact center transformed in Q3'. Or 'workforce halved by the end of the year'. The CFO finally smiling blissfully. Change at the expense of FTEs.

Short-term overestimation is the native language of our industry, spoken fluently at every conference between the keynote and the drinks. The underestimation, the long-term part, is quieter, slower, and considerably more expensive. There, the other four laws wait patiently for their turn.

Jevons' Paradox: The cruel joke of efficiency

In 1865, the British economist William Stanley Jevons noted something unsettling about coal. As steam engines became more efficient, Great Britain did not use less coal. It used considerably more. Efficiency did not reduce demand; it caused it to explode.

Every time AI makes a customer interaction cheaper and faster, something predictable happens: more interactions are taking placeMore channels are being opened. More customers are contacting you, more often, for more reasons, at more inconvenient times. The bar for what good service means is rising in real time, because that is simply how it always goes. You have deployed AI to reduce volume. Congratulations. You have just created more new volume. And naturally, AI costs are increasing exponentially as a result. That paradox does not punish a bad strategy – it punishes all strategies equally, with cheerful impartiality.

The Turing Trap: A philosophical problem disguised as a budget item

Erik Brynjolfsson’s Turing pitfall is the seductive tendency to build AI that replaces humans rather than enhances them. The test is named after Alan Turing’s famous question: can a machine mimic a human so convincingly that they are indistinguishable from the real thing? The latter is what the industry is doing. We shout very loudly for support regarding co-pilots, but those remain productivity tools. Agentic is not a technical term but a behavioral science term. Humans can be agentic just as well. Agentic Human Agents (AHA!), now that would be something. CX would look different.

My son, who studies philosophy, notes that Turing's question was a provocation regarding the nature of intelligence – not a tender document. The CX industry's decision to treat it as such, and then to jump into the trap en masse with cheers, is a form of supreme emulation confusion.

Goodhart's Law: pitfall squared

Charles Goodhart, with magnificent brevity: when a metric becomes a goal, it ceases to be a good metric. The moment 'containment rate' becomes the KPI, you are no longer measuring customer experience, but bluntness. The moment average handling time is the number everyone looks at, speed has become the product. The moment cost per interaction dictates everything, you have built a machine optimized for cost efficiency, which you then call 'customer service'.

It all sounds like a procedural error in court. In CX, we call them KPIs, until the figures on customer churn or sentiment come in and the room falls silent.

Martec's Law: Why smart people still fall for it

Scott Brinker's contribution: technology changes exponentially, organizations change logarithmically. The gap between them is not a planning problem. No company will be able to keep up in the avalanche of new technology through simple organizational changes. Making choices is enormously important, and constantly reinventing yourself as a company, no matter how drastic, is necessary. Evolution seems to be moving towards AI First, but for organizations, this makes making future choices much more difficult. Ultimately, you won't get anywhere by simply picking the low-hanging fruit. A true agentic enterprise that applies all the laws mentioned above is much more extreme, but certainly ready for the future.

Amara fills the room with hype. Jevons increases the demand. The Turing trap continues to automate people. Goodhart corrupts the measurement. Martec ensures that no one makes timely adjustments.

Every truly difficult customer—frustrated, complex, and emotionally charged—ultimately ends up with a human agent who is slower, less trained, and already at a disadvantage. This is not a hypothetical scenario, but the current roadmap.

The way out

Every law also points to the exit. Amara: invest for the long term, but not dogmatically. Jevons: build for more demand, not less. Turing: reinforcement beats replacement, always. Goodhart: measure resolution rate and customer value, not containment rate. Martec: consciously and drastically close the organizational gap before it closes around you.

My philosopher son remains convinced that AI will never fathom what it truly means to understand someone. He believes that the industry, generously estimated, is lagging behind that question by a few centuries. My daughter, who studies English, doubts whether AI can sense the difference between language that merely functions and language that truly resonates. Walking among the landmines, I find them both more convincing than the vendor presentations. Your customers don't want a Turing test. They want to be heard. Build for that.

(Ziptone/Merijn te Booij)

Featured, Opinions, Technology