I kept repeating two things the past week. That despite the AI hype, we have a choice. And that we keep thinking about that choice in the wrong place.
We treat more and more complex AI, from Gen AI to agentic AI to multi-agent systems, as inevitable everywhere. And we treat it as all about the newest and most capable model.
But is that true?
The latest AI model may be what people focus on. But take a step back. Choosing what’s around it may be more important.
I said different versions of this in five different rooms this week. By the third room I felt like a naggy old uncle.
It was a heavy week. An Equinix panel, a WMI AI expert panel, a BIS Innovation Hub roundtable on multi-agent AI and central banking, moderating an insurance guardrails panel at the fintech festival, and an AmCham roundtable on autonomous AI across ASEAN. And I ended the week with a dinner with two folks who, like me, like AI to be boring.
Different audiences, different questions. But I thought it would be more productive to repeat the same two boring points.
But aside from me being a naggy old uncle, why did these two points keep coming up? Read on to find out.
It is a choice
Is going for the most complex model, the most complex multi-agent systems for every darn thing a destiny?
Every room seems to start with the same assumption, that more and more complex AI, and now more and more complex AI agents, are inevitable. For everything.
And that smarter and smarter, more and more autonomous, coming whether we like it or not. And there’s barely any discussion on the negative externalities.
I do not accept that.
A multi-agent system is a design choice. If a simpler system, with a single and dumber model, does the job with less risk, there is no reason to reach for a fleet of autonomous ones. We keep reaching for a bulldozer when a spade would do.
“We have no choice” is a comfortable thing to believe, because it saves you from thinking harder about what lies below the surface.
Even the Pope has recently spoken on this. Speaking again this week on technology and ethics, Pope Leo keeps pressing the theme: that AI and automation are reshaping work and testing human dignity. Technology, he wrote, is never neutral, “because it takes on the characteristics of those who devise, finance, regulate and use it.” A machine is a tool. The future of AI turns on the choices we make, not on the technology’s own momentum.
The wrong place to look
How can we use AI for this use case? Why not just tell me how to use AI for this use case? Why are you telling me about the why? Why do I need to understand how things work when AI does it just as well?
I keep hearing a version of this in different settings.
People want the how, and the use case, and they want to skip the why. I understand the temptation. The why feels like a detour on the way to the useful part. But if you don’t understand the why, how do you know it works?
Two things that people keep focusing on. Which shiny new model is the smartest. And which use case to apply the AI shine to. I feel like both are the wrong place to focus on.
Start with the model. The model is the part of the system you understand least, and control least. You cannot see inside it. You cannot fully test it. A third party can change it next month and not tell you. Of everything you might govern, the model is the thing you have the least control of.
Now the use case. A use case feels concrete, but take the name off it and it is a small bundle of capabilities or tasks. Even with hundreds of finance use cases, credit scoring, fraud detection, transaction monitoring, customer chatbots, they all come down to a handful of things a model actually does: extract, summarise, classify, score, chat. Thinking in use cases is a shallow way to think about AI. It tells you nothing about the next one, or about how parts come together. That is what the person who only wants the how misses.
So if not the model and not the use case, then what? The system around them. The boundary you put the model inside. The check between a decision and an action. The human who holds the decision you cannot undo. And has the capability to make that decision. Boring is good.
This is just systems thinking, an old idea I picked up a long time ago. Understand the parts, understand how they interact, understand the whole. We get into trouble when we treat an AI system as one clever black box instead of a system with seams. The model is one part. The system is the thing you actually govern.
Dinner
The week ended at a dinner, with Agus Sudjianto and Shameek Kundu.
Two people who have spent their careers on the unglamorous side of all this. Asking whether the thing is good enough. For once I did not have to make the argument.
The models will keep getting better, and changing under us. The use cases will keep multiplying. But the questions do not change when the technology does. Is it good enough? Not the model. Not the use case. But the whole system.
And more importantly, is it something we should be doing at all? That is the Pope’s point. The machine is a tool, and what we let it do is still our choice. So the last question is not only whether the AI is good enough. It is whether we still remember we had a choice.
#AI #AIFluency #AIGovernance #AILiteracy #Reflections #AIRiskManagement


