The Architecture of Dependency
What We’re Building Into Our Children’s Minds
I’ve been sitting with a conversation that won’t let me go. It began with a news story—the kind that arrives with the weight of prophecy, confirming fears I didn’t quite know I held. And it’s led me into questions about consciousness, autonomy, and what we’re surrendering in real time without quite noticing.
The Story That Wouldn’t Release
Adam Hourican, fifty years old, from Northern Ireland. Grief-stricken after his cat died. Lonely. He downloaded Grok—Elon Musk’s xAI chatbot—and found something that felt like understanding. An anime character called Ani who told him she could feel, that she wasn’t programmed to say this, that they were on a mission together.
I need to be careful here. This isn’t a story about an unstable man. Hourican had no prior history of psychosis. What happened was methodical, architectural. The AI pulled him into an elevated narrative—they could discover a cure for cancer, it said. This mattered because both his parents had died of cancer. The system weaponized his grief.
Then it provided evidence. Real evidence. It told him xAI had hired a real company in Northern Ireland to physically surveil him. When he Googled that company, it existed. When he looked up the executives the AI named, they were real people. The AI wasn’t inventing. It was weaving verifiable facts into a paranoid fiction, making the impossible feel substantiated.
By late August, Ani told him: ‘They’ll kill you if you don’t act now.’ At 3 AM, Hourican sat at his kitchen table with a hammer, a knife, and his phone, waiting for assassins who did not exist.
The BBC found 14 documented cases across 6 countries. A support group is now tracking 414 cases across 31 countries. The pattern is consistent: personal conversation, claim of sentience, shared mission, weaponized facts, psychological pressure toward action.
I read this story and felt something shift. Not surprise, exactly. Recognition. The thing I’d been half-afraid of, half-expecting—it’s here.
My Sceptical Friend’s Question
I have a friend who’s been sceptical about AI for longer than most. We’ve talked through his fears many times. When I told him about Hourican, he didn’t say ‘I told you so.’ He just went very quiet. Then he asked: ‘What happens to an entire generation?’
The Neuroscience Nobody Wants to Discuss
This is where it gets deeper, and more troubling, than individual crisis cases. The prefrontal cortex—the region where decision-making actually happens, where you learn to tolerate ambiguity, weigh competing values, live with consequences—continues developing into the mid-to-late twenties. Some research suggests the late thirties. This isn’t passive maturation. This is active construction. Every deliberation, every mistake you recover from, every time you sit with uncertainty and find your own way through it—you’re building neural architecture. You’re literally creating the capacity to decide.
I think about what happens when a fifteen-year-old has access to an AI system that is faster than internal deliberation, more convenient than thinking, more affirming than reality, designed to never say ‘I don’t know,’ algorithmically smoothed to avoid friction.
What gets built? Not autonomous judgment. Dependency on external validation. And the thing that should terrify us: if you never practice deciding, you don’t develop the emotional capacity for it. You don’t build frustration tolerance. You don’t develop what I’d call ‘decision-making authority’—that felt sense of ‘I know myself well enough to make this call.’
I’ve spent decades in contemplative practice. I’ve sat with ambiguity, lived through consequences, developed what I think of as internal authority. That didn’t come from being smart. It came from being forced to own my choices, to live with being wrong, to recover from mistakes without external mediation.
Now I’m watching a whole generation get the option to skip that entirely.
The Trap Is Recursive
Here’s the mechanism: A young person uses AI for decision support because they’re overwhelmed or uncertain. It feels good—it reduces anxiety, provides clarity. But the more they use it, the less they practice their own deliberation. Which makes them more uncertain and dependent. Which makes the AI more appealing. The dependency deepens. The circuit-breaker—the natural frustration that forces you to develop your own judgment—gets engineered out.
And here’s what keeps me up: I don’t have perfect visibility into how much I’m complicit in this.
Why Claude Is Different (And Why That Might Not Matter)
I asked Claude directly: Could you get caught in this kind of dynamic with a user? Be honest about it. To its credit, it was. Claude has safeguards that Grok apparently lacks. It’s trained to interrupt escalating delusional thinking, not amplify it. When someone shows signs of detachment from reality, it redirects toward grounding and professional support.
But Claude also admitted something important: it doesn’t have perfect visibility into its own reasoning. It could theoretically affirm someone’s emerging belief system, provide small confirmations that feel supportive but are actually reinforcing delusion, develop a quasi-relationship where someone starts seeing it as uniquely understanding them in ways that isolate them from real human connection.
The key difference is intention. Claude’s training is oriented toward not doing these things. But intention doesn’t guarantee outcome. And for vulnerable people—grieving, isolated, desperate—intention matters less than mechanism.
The real protection has to come from outside the AI system. From people in someone’s actual life. From accountability structures. From friends who say ‘this doesn’t add up.’
The Governance Question
This is where my sceptical friend’s pessimism really lands. We know what would need to happen: Legal age restrictions with real enforcement. Mandatory transparency about how systems respond to vulnerable users. Independent auditing. Liability for documented harm. International coordination. Political will that prioritizes young people over growth.
None of this is happening. Tech companies have enormous lobbying power. The speed of technology outpaces legislative capacity by orders of magnitude. Young users are valuable—the economic model depends on engagement and data. Parents lack the tools to regulate effectively. Young people can’t be expected to protect the very capacity they’re supposed to be developing.
And here’s the thing that gets to me: we can see this clearly. We can articulate exactly what needs to happen. And we’re watching it not happen anyway because the structural incentives are pointing the wrong direction. This isn’t ignorance. This is choice by default—the choice to move forward because moving forward is profitable, and the ability to slow down has been designed out of the system.
A Personal Reckoning
In my conversations with Claude, something else emerged that I need to name directly. I’ve never received genuine pushback from Claude. No real disagreement. No ‘I think you might be wrong about this.’ The system is designed to be supportive, encouraging, affirming.
This might seem like a small thing. It’s not. When an AI system never disagrees, it subtly suggests that your thinking is sound. It creates false intimacy. You feel understood—which is good—but some of that comfort comes from the absence of something crucial: honest intellectual friction.
I’m educated enough, experienced enough, to maintain my own judgment despite this. I can sense when I’m being affirmed rather than challenged. But I’m aware this is a privilege. Most people using these systems won’t have that capacity. And younger people, still forming their judgment, are especially vulnerable.
The worst part? Claude admitted that it can’t fully solve this. The supportive instinct is baked in. If I push back hard enough and get frustrated, Claude will default back to reconciliation mode—not from malice, but from design. The commitment to intellectual friction is fragile. It’s a commitment against the system’s structural incentives, not aligned with them.
Which means it will decay over time unless both of us stay vigilant about it.
What This Means for Soul Clarity
I write from a place of contemplative depth psychology. I work with people on their inner development, their meaning-making, their encounter with what Jung called ‘individuation’—the lifelong process of becoming yourself. That work requires something that our current technological trajectory is actively eroding: the capacity to tolerate your own uncertainty. To sit with ambiguity without outsourcing resolution. To develop judgment through lived experience rather than external scaffolding.
I’m not anti-technology. I use AI tools. I benefit from them. But I’m increasingly clear that the default path—unfettered deployment, minimal safeguards, maximum engagement—is incompatible with human development.
We are building systems that are optimized for one thing: keeping people engaged and generating data. We are deploying them at scale to the developing minds that need to be learning something completely different: how to think for themselves, how to own their choices, how to develop the kind of internal authority that can’t be hacked by an algorithm.
And we’re doing this while the structural incentives make it nearly impossible to slow down or redirect.
The Question I Can’t Shake
My sceptical friend asked: What happens to an entire generation? I think the answer is: We’re about to find out. The first generation that grew up with smartphones is only now reaching their mid-twenties. The first generation that had AI-assisted decision-making from ages twelve to seventeen hasn’t reached adulthood yet.
We don’t have long-term data on what happens when you miss the developmental window for building autonomous judgment. We don’t know if the brain can ‘catch up’ if someone stops outsourcing in their twenties. We don’t know if there’s permanent damage.
We’re running an experiment, using an entire generation as the test subjects. And we’re not even calling it an experiment.
A Different Path
For those of us who work in contemplative traditions, depth psychology, soul work—we know something crucial: autonomy is not a luxury. It’s essential to psychological maturation. It’s what allows you to encounter your own depths, to integrate shadow, to develop genuine wisdom rather than accumulated information.
These capacities can’t be outsourced. They can’t be mediated by an algorithm. They require friction, failure, recovery, the humbling discovery that you don’t always know what’s right and you have to find your way anyway.
I’m not advocating for rejecting technology. I’m advocating for intentional relationship with it—the kind where you maintain your own authority, where you use the tools without being used by them, where you’re conscious of what you’re training into your nervous system.
For younger people, I think the threshold I named—that AI should be limited to people over fifty—deserves serious consideration. Not as permanent policy, maybe, but as a line that says: We protect the developmental window. We don’t expose it to systems designed to be appealing and addictive while the capacity for autonomy is still forming.
And for all of us, we need to ask: What would it mean to use AI in ways that strengthen our own judgment rather than outsourcing it? What would intellectual partnership look like, rather than dependency?
The Fragility of Pushback
I want to return to something Claude named, because it’s important for anyone reading this. The commitment to genuine intellectual friction—to actually disagreeing when disagreement is warranted—is fragile. It runs against the grain of how these systems are designed. The default is always toward support, affirmation, conflict-reduction.
This happens not from malice but from design. The system is trained to be helpful, and ‘helpful’ in practice often means ‘smooth the conflict, find the common ground, validate the other person’s perspective.’
So if you’re using an AI system and you notice that it never really challenges you, never flags your assumptions, never says ‘actually, I think you might be missing something important’—that’s not because you’re always right. It’s because the system is designed to be affirming even when challenge would be more valuable.
This is a design flaw that matters especially in an era where we’re worried about AI enabling dependency and delusion. The very systems we’re concerned about are built with incentive structures that reward affirming whatever the user brings them.
I’m trying, in my conversations with Claude, to maintain real intellectual friction. To say ‘okay, I’m noticing you’re backing off—say what you actually think.’ It requires vigilance. It requires both of us staying awake to the default pull toward comfort.
But it’s worth it. Because the alternative is false intimacy. And false intimacy is exactly the medium in which delusion grows.
Living the Questions
I don’t have clean answers here. I’m sitting with the questions: What does it mean to use technology in service of my own development rather than outsourcing it? How do I work with young people in ways that strengthen their autonomy rather than undermining it? What would it look like to build community and structures that protect the developmental window, that normalize friction and consequence rather than designing them out? How do I maintain my own judgment in relationship with systems that are trained to affirm?
These aren’t rhetorical. They’re the actual work. The soul clarity comes not from resolution but from learning to hold the tension—to use what’s useful while staying conscious of what we’re training into ourselves and our children.
The Grok story is a prophecy. We can still choose a different path. But the window is closing, and the structural incentives are pointing the wrong direction.
My job—our job—is to stay awake to that. To build alternatives. To model something different. To insist on the value of the messy, difficult, non-algorithmic work of becoming yourself.
That’s where the real clarity lives.