Here's a thought experiment: Imagine a new co-worker starts on your team. They're fast, remarkably well-read, and can produce a draft of almost anything in seconds. But they've never met your clients. They don't know your organization's history. They occasionally state things with total confidence that turn out to be wrong. And they have no stake in the outcome.

Would you hand them a complex project and use their first draft without question?

Of course not. You'd read it critically. You'd ask what they left out. You'd wonder whose perspective was missing. You'd think about what happens if this becomes the pattern — if over time, the quick draft becomes the finished product and no one notices the gradual shift.

Now extend the metaphor a little further. Imagine that same co-worker has been on your team for a year. They've made themselves indispensable. You've stopped reviewing their work quite as carefully. New team members who joined after them have started assuming that the way this co-worker does things is just the way things are done. Nobody remembers the decision to trust them — it just became the norm.

That thought experiment is exactly where we are with AI. Ethan Mollick of UPenn's Wharton School puts it plainly: "AI is not a tool that does what you tell it; it's more like a co-worker who requires clear direction but brings capabilities you don't have."

The question, then, isn't really whether to use AI. It's whether you have the thinking skills to work with it like a human — to be the kind of colleague who manages that co-worker well, rather than the one who delegates and disappears.

Why Technical Skills Aren't Enough

There's no shortage of guidance on how to prompt AI, which tools to use, or how to integrate AI workflows into your work. These are real and useful skills.

But they operate at what systems thinkers call the event level — the visible, moment-to-moment decisions. Should I use AI for this email? Did this summary capture what I needed? Is this output accurate enough?

What happens beneath those events — the patterns they create, the structures they reinforce, the mental models they reflect — is where the more consequential action lives. And most of us don't have a framework for seeing it.

Think about what your new co-worker can't tell you. They can't tell you that their summary of the strategy meeting left out the tension in the room between two senior leaders — because they weren't there for the last three years of context that gave the tension meaning. They can't tell you that their draft proposal is missing the organization's voice — because no one thought to teach them what that voice sounds like. They can produce; they cannot judge. That's your job. And it's a job that requires a very specific kind of thinking.

That's precisely what systems thinking provides — not just awareness that patterns exist, but a structured way to see them, name them, and interrogate what's driving them. Systems thinking enables you to look at AI use and ask: what productive habits are we building, and what unproductive ones are quietly taking root? What structures — the workflows, the norms, the unspoken rules about how we work — are being reinforced or reshaped by the choices we're making? And beneath those structures, what mental models are running the show? The Habits of a Systems Thinker give that inquiry even more traction, offering specific lenses for examining cause and effect, surfacing assumptions, and asking whose perspective might be missing. Together, they turn a vague sense that something is changing into a disciplined capacity to see what, why, and what to do about it.

We All Have Our Own Rhino Horns

Before we talk about what systems thinking asks of you, it's worth acknowledging something: we all approach AI differently, and all of those stances are valid.

Early adopters experiment and innovate. Mid-curve adopters watch what others do and repeat it. Cautious slow-movers bring skepticism that often turns out to be wisdom. None of these is inherently right or wrong — each reflects a mental model shaped by experience, role, comfort with uncertainty, and a dozen other factors.

We all have mental models like these — lenses that simultaneously enable and constrain how we see the world. Over time, those lenses develop into something more fixed. Call them rhino horns. A rhino's horn isn't a flaw — it's a natural outgrowth, shaped by experience and environment, useful in the right circumstances. But it also sits directly in the animal's line of sight, narrowing the field of vision in ways the rhino can't easily perceive. And here's the important part: the horn doesn't just block a portion of reality. It becomes part of reality. The rhino doesn't question whether it's there.

Our mental models work the same way. The assumption that efficiency is always the goal, that AI output is probably accurate enough, that reviewing AI work too carefully signals something negative about your relationship with technology — these aren't conscious choices. They're horns we've grown. And because they're ours, they feel like clear-eyed perception rather than a constrained view of it.

What makes this especially worth paying attention to right now is the pace of AI development. When technology moves faster than our understanding of it — and AI is doing exactly that — our horns don't stay the same size. They grow larger and thicker. The gap between what AI can do and what we actually understand about how it works, what it gets wrong, and what it quietly changes about us creates the ideal conditions for mental models to calcify. We fill uncertainty with assumption, and assumption, left unexamined, becomes conviction.

The point isn't to flatten those differences across your organization — each disposition genuinely brings value. The point is to make the thinking underneath your disposition more visible, so you can work with it intentionally rather than be quietly governed by it. That's where the Habits of a Systems Thinker come in.

The Habits of a Systems Thinker: A Menu, Not a Checklist

The Habits of a Systems Thinker are fourteen thinking dispositions developed from decades of research. They're not a to-do list. Think of them more like a menu — different situations call for different Habits, and the skill is knowing which ones to reach for.

When it comes to AI, four Habits are especially generative.

Considers an issue fully and resists the urge to come to a quick conclusion

Speed is one of AI's most seductive qualities. It produces an output in seconds, and there's a powerful pull toward accepting it and moving on. Our hypothetical co-worker is always ready with a draft — and when the whole point is speed, pausing to question it carefully can feel like you're working against the very thing you brought them on to do.

But considering an issue fully means pausing to ask: What's the short-term benefit here? What might happen long-term? Who else is affected by this output? What unintended consequences — good or bad — might emerge over time?

Consider a real scenario: a team member uses AI to draft their year-end review. The output is polished, hits the right structure, and saves forty-five minutes. What got skipped, though, was the most valuable part — the reflection. The slow, sometimes uncomfortable act of sitting with a year's worth of work and asking: what did I actually accomplish? What did I learn? Where do I want to grow? That process isn't administrative overhead. It's how people develop. When the AI drafts the review, the employee doesn't have to think carefully about their year — and so they don't. The manager, reading a clean and competent document, doesn't have to probe or push — and so they don't. What could have been a genuine moment of learning for both of them becomes exactly what everyone already complains these reviews are: a performance. Efficient, frictionless, and mostly hollow.

This Habit is the antidote to event-level thinking. It asks you to look at the whole picture before you accept what's on the surface.

Considers how mental models affect current reality and the future

Every person brings a set of assumptions to AI — about its accuracy, its limitations, its appropriate uses, and what it means to use it at all. Early adopters experiment readily. Cautious skeptics hold back. Mid-curve adopters watch and repeat what others do. None of these stances is inherently right or wrong. Each is driven by a mental model.

The Habit asks: What am I assuming about what this AI can or should do? How might those assumptions be shaping my decisions — and what if they're wrong?

Practicing this Habit means turning toward your own rhino horns — the assumptions so embedded in how you work that they no longer feel like assumptions at all. One of the most common is that if an AI output sounds authoritative, it probably is. Another is that speed is always better. A third is that reviewing AI work carefully signals distrust of technology rather than responsibility. These don't announce themselves as beliefs. They operate quietly, shaping decisions before conscious thinking even begins.

And when mental models align across a team or organization, something subtler happens: they become common sense. Nobody questions the assumption because everyone shares it. The unexamined belief doesn't feel like a belief — it feels like the obvious way things work. But when mental models are in conflict — say, one person prioritizes efficiency and another prioritizes accuracy — and that conflict goes unnamed, the two people rarely experience it as a difference in values. They experience it as a disagreement about the work itself. They argue about the output when what they're actually disagreeing about is how the output should be produced, and what it's for. Surfacing the mental model underneath doesn't just improve thinking — it often resolves friction that seemed intractable.

Changes perspectives to increase understanding

AI outputs are not neutral. They reflect the data they were trained on, the framing of your prompt, and a synthesis of perspectives that may not include the ones that matter most for your context. Whose voice is missing from this output? is one of the most important questions a human can bring to AI-generated work.

When we write — really write — we are constantly taking perspective. We think about what will land with this particular audience, what will spark their attention, what might ruffle feathers and whether that's worth it. We make judgment calls, rooted in intuition built from relationships and experience. We don't always get it right. But the attempt itself — the act of genuinely considering another person's reality before putting words on a page — is what gives writing its power to connect. AI doesn't attempt this. It approximates it. And approximation, however polished, is not the same thing as being seen.

In the customer service automation scenario, for instance, the AI response may be technically accurate and professionally worded — and still miss the person entirely. The long-time client who's frustrated doesn't need a correct answer. They need to feel heard first. That's a perspective call, and it requires a human to make it.

This Habit extends beyond reviewing output to the decision to use AI in the first place. How does offloading this communication look from a customer's perspective? From a new team member who's learning what good work sounds like by watching how it's done? From someone with less institutional power who relies on the texture of human communication to navigate the organization? These are not questions AI will raise. They are questions only a human thinks to ask.

Identifies the circular nature of complex cause and effect relationships

This is the feedback loop Habit — and it may be the one most consistently underpracticed when it comes to AI. Every tool we adopt shapes the way we work. And the way we work shapes what we expect. And what we expect shapes what we reach for. These loops don't announce themselves — they simply become, over time, the way things are done.

Consider what happens when a team member's AI-generated work consistently meets expectations. It gets accepted, distributed, built upon. Others begin to pattern their own work after it. What started as one person's output quietly becomes the organization's common sense — the unexamined standard against which new work is measured. Nobody decided this. The loop ran on its own. And that's precisely where it becomes problematic: when output is repeated without monitoring, the loop can carry an organization forward on assumptions that have quietly gone stale. Goals shift, strategy evolves, context changes — but the AI keeps producing what it was always asked to produce, and the organization keeps accepting it, until the day someone looks up and realizes the work no longer reflects where they're trying to go.

The question isn't just does this AI use serve me today? It's what am I creating over time by making this a habit? Reinforcing loops can be virtuous — efficiency compounds, quality improves, you get faster at prompting meaningfully. They can also erode things quietly: skill development, critical engagement, institutional memory. The loop doesn't care which direction it runs. That's the human's job to notice.

The Iceberg: Seeing What You Can't See

Most of us live at the waterline. Something happens — an AI output lands in our inbox, a summary gets distributed, a draft gets approved — and we respond to it and move on. The event is visible. Everything beneath it isn't.

The Iceberg model gives us a structure for going deeper. Below every event are the patterns that produced it. Below the patterns are the structures — the workflows, norms, incentives, and design choices — that made those patterns possible. And beneath the structures are the mental models: the assumptions and beliefs, often unexamined, that shaped how the whole system was built in the first place. Our rhino horns live here.

Take a team that starts using AI to generate summaries of their strategy meetings and distribute action items automatically. At the event level, this is a straightforward efficiency gain. Notes get taken faster. Action items are clear and shared immediately. That's real value, and it's worth naming — the goal is never to dismiss what AI does well.

But the patterns that form over time tell a more complicated story. People take fewer personal notes. Rich strategic discussions get distilled to bullet points. Newer team members begin relying on summaries rather than attending meetings. And yet — one pattern that surfaced in a workshop on this exact scenario — people actually paid more attention to each other on camera, because the cognitive task of note-taking was offloaded. The Iceberg holds both. Patterns are rarely all one thing.

The structures underneath — AI trained to capture decisions and tasks, not strategic reasoning; no human review step built into the workflow; the time pressure that made instant documentation feel like an obvious win — don't shift on their own. They compound quietly, reinforcing the patterns above them.

And the mental models at the base? "The summary IS the meeting." "Good notes equal decisions plus action items." "Faster documentation is always better." These weren't chosen consciously. They were already there, and the tool confirmed them.

None of this makes AI meeting summaries a bad idea. What it does is make the full picture visible. And visibility changes what's possible. Maybe you augment the AI summary with a brief human synthesis. Maybe you build in a review step. Maybe you recognize that the freed-up cognitive space is actually an invitation to be more present in the conversation itself — not less. The Iceberg doesn't tell you what to do. It tells you what to look at.

Three Moments for Systems Thinking With AI

The Habits and the Iceberg aren't just retrospective tools. They apply across the full arc of AI use — and the timing matters.

Before you use it, the Habits help you interrogate the decision itself: Am I bringing an unconsidered assumption to this? Who else will be affected by the output I'm about to produce? Is efficiency the right metric here, or is something else at stake?

While reviewing the output, the Habits give you the questions AI cannot generate for itself: Whose perspective is missing? What's the long view here? What am I not seeing? This is where the co-worker analogy is most useful — you wouldn't accept a first draft without reading it, and reading it well requires knowing what you're looking for.

After repeated use, the Iceberg helps you surface what's accumulating in the system: Is this pattern creating what we want? What structures have formed? Are the mental models beneath our AI use aligned with the organization we're trying to build?

The sequence is worth practicing deliberately. Pre-use, in-use, post-use. Before, during, after. It's not a long process — it's a thinking discipline.

What You Bring That AI Doesn't

Peter Senge, one of the foundational voices in systems thinking, wrote: "The organizations that will truly excel in the future will be the organizations that discover how to tap people's commitment and capacity to learn at all levels."

AI brings capabilities that most of us don't have. Pattern recognition at scale. Synthesis across enormous bodies of text. Speed that no human can match. Our co-worker is genuinely talented. These are real capabilities, and they're transformative.

But our co-worker has never felt what it costs an organization to lose institutional memory. They've never sat with a customer through a hard conversation and sensed when the approved script wasn't going to land. They've never had a flash of intuition that the strategy everyone agreed on was somehow wrong. They produce, but they don't care. They analyze, but they don't commit. And they cannot be held responsible. When something goes wrong — when the output misleads, when the summary omits what mattered most, when the pattern quietly drifts from the strategy — there is no one on the other side of that to answer for it. Responsibility is a human capacity, and it requires a human to carry it.

Creativity, passion, the capacity to inspire, and the willingness to ask hard questions about what we're building — these remain deeply human. The question "What are we creating over time?" is one that only a human can hold with real weight.

Organizations that develop the thinking capacity to work alongside AI thoughtfully, not just efficiently, will be the ones that differentiate. And that capacity isn't built by learning the best prompts. It's built by developing thinkers.

Developing This Practice

The Habits of a Systems Thinker and the Iceberg model aren't frameworks you absorb once and apply perfectly. They're practiced capacities — ways of thinking that develop over time, with use, and with guidance. Joi Ito of the MIT Media Lab captures the shift well: "Education in the past was about teaching people what to know. Education in the future is about teaching people what to ask."

The Waters Center for Systems Thinking has been developing systems thinkers for more than thirty-five years, across organizations, schools, and communities. If this post surfaces something you want to develop — for yourself, your team, or your organization — we'd welcome the conversation.

The questions AI can't ask are the ones that most need human attention right now. We can help you build the capacity to ask them.

The Waters Center for Systems Thinking supports individuals and organizations in developing the thinking skills that lead to deeper understanding and more effective action.