Agent-driven systems thinking
I often see negative feedback loops forming around LLMs, and even this sub's rules say AI does not simulate thought in any meaningful sense. I agree with the intent. But I want to gently push on one narrow part of it, from experience: harnessed with the right tools and instructions, AI does simulate a thought process well enough to change what one person can do with a system model.
I remember few years ago, when I was learning about Futures Studies, the idea of mapping out systems and understanding its feedback loops really hooked me into thinking in maps. But even then, validating where the thought process had systemic flaws or utilizing mathematical calculations to simulate was never an easy process and often ignored. Now with AI, this has changed.
I'm a software engineer, so I experience this wave up close. The way I experiment now: I let agents build models using system dynamics tools I'm familiar with. I visually check the structure and ask it to simulate various scenarios. And this goes into a loop of its own: often into a subsystems where my interest hones in. None of this works with a raw model. It has to be guided thoroughly with tools and instructions — AI is bureaucratic by design, to borrow Harari's term, which is exactly why it's good at this kind of structured formalization work. Guided that way, it does well. Often beyond my expectation.
I remember the time in highschool (being korean who memorized multiplications, you know what I am talking about if you are Korean) knowing how calculate stuff in my head was thought to be gifted. It was like in test, if you use a calculator, you're cheating. But then the point was not about having the ability to memorize or do complex calculations in your head but to understand the system behind its mathematical journey which leads to a bigger system where it is almost impossible to reach without the help of scientific tools.
I think systems thinking is at the same moment. AI needs the right tools to do the right work, and people — including myself — are building those tools. My claim: bigger systems, whose modeling and simulation used to be delegated to a particular set of people, are becoming accessible to a much broader audience.
So I'm curious:
Has anyone here tried this — letting an agent do the model-building and simulation grind while you stay at the level of structure and judgment? Did it deepen your thinking or shallow it?
The calculator eventually became allowed in the exam room. What's the equivalent boundary for AI in systems work — what should still be done "in your head"?