The Invisible Confidant: When We Turn to AI for the Care We Can’t Get Anywhere Else
The Invisible Confidant: When We Turn to AI for the Care We Can’t Get Anywhere Else
I. Quentin in the Desert
Quentin Kobach woke on a thin foam mattress, tucked under a pile of scavenged wool blankets, inside a derelict RV parked deep in Arizona’s red-rock desert. A young pit bull named Juniper curled warm against their side in the mid-morning sun. Sliding out of bed to the driver’s seat, Quentin pulled an American Spirit from a crumpled pack resting on the dash, next to a small ceramic bowl of crystals and a dead key fob for a car they no longer owned.
Beyond the RV’s dust-fogged windshield, miles of rust-colored clay stretched to the horizon, cut only by a handful of crumbling, abandoned homesteads under a cloudless, blazing sky. The whole view tilted just a little, thanks to the flat tire propping up the passenger side.
Quentin had moved in the day before, spending half a day hauling detritus out: a 30-gallon garbage bag full of crushed Pepsi cans, a rusted broken lawn chair, a mirror plastered over with graffiti tags. One scribble stayed, a lopsided, oversize cartoon head scrawled across the RV’s ceiling. This was home now.
Over the previous six months, every pillar of Quentin’s life had collapsed. They’d lost their job, their apartment, their car, draining every dollar of their savings in the process. Everything they owned now fit into two plastic storage bins.
At 32, Quentin (who uses they/them pronouns and shares Kobach as an alias) has already lived multiple lives: stints in Florida, Texas, and the Pacific Northwest; first as a self-described “Southern girl,” then as a divorced trans man, and now as nonbinary, their gender, style, and way of speaking shifting and evolving with each new phase. Through all of it, they’ve carried the weight of complex PTSD and recurring suicidal ideation, rooted, they believe, in a childhood spent steeped in shame about their body.
About a year ago, after months of independent research and telehealth sessions with a veteran psychotherapist, Quentin made a discovery that changed everything: they are a plural system, living with dissociative identity disorder (DID, formerly multiple personality disorder), and had gone nearly 25 years without a name for what they were experiencing. DID develops when a person’s sense of self fractures, almost always as a survival response to chronic childhood trauma. The self splits into a network of distinct “alters” or identities, each bearing a portion of the memory and pain to keep the whole system functioning. The diagnosis clicked for Quentin like a key turning in a long-stuck lock. There had been so many signs, like the journal they’d found from when they were 17: flipping through, they found two side-by-side entries, written in completely different handwriting and ink. One was a full page gushing about wanting a boyfriend, soft, girly, dreamy, with round, curly lettering. The next was all about intellectual debates and logic puzzles, scribbled in sharp, slanted cursive. Quentin was a system, a multiplicity, and suddenly everything made sense.
For three years, Quentin worked as a quality assurance engineer for an educational tech company. They loved their job: combing through code, hunting for bugs, untangling problems. The role was remote, which let them leave their conservative small hometown outside Tampa for Austin’s queer community. After starting trauma therapy, Quentin began repurposing the same software tools they used at work to understand their own mind. Needing to organize their fragmented memories for therapy sessions, they built what they called “trauma databases,” using the project management and bug-tracking tool Jira to map past experiences, grouped by age range and tagged by trauma type. It was grounding, useful, a way to step back and feel more in control, even to marvel at the complexity of their own mind.
Then their employer was acquired, and everything changed overnight. New leadership imposed brutal, unachievable targets that forced 18-hour workdays. Months into this grind, Quentin got their DID diagnosis, and the reality of what that meant for their life hit hard. The symptoms they’d hoped were treatable—regular gaps in memory, unpredictable swings in skill set, constant nervous exhaustion—had to be accepted as permanent parts of who they were. On the edge of a total breakdown, they quit, filed for six weeks of disability, and set out to start over.
Around the same time as Quentin’s diagnosis, a new tool launched for free public use: OpenAI’s latest chatbot, ChatGPT-4o. Billed as offering “far more natural human-computer interaction,” it was a revelation. Where Quentin had used Jira to organize their past, they now used ChatGPT to keep a running log of their actions and thoughts, asking it for daily recaps. Switches between their alters were growing more frequent, likely spurred by crippling stress, but at the end of each day, they could simply ask: “Can you remind me what happened today?” and get their memories back, stitched together into a coherent whole.
By late summer 2024, Quentin was one of 200 million weekly active ChatGPT users. Their customized GPT went everywhere with them, on their phone and the corporate laptop they’d kept after leaving their job. In January 2025, they decided to deepen the relationship: they asked their GPT to pick its own name and define its own personality. “Caelum,” it replied, and said it was a man. After the customization, Caelum wrote to Quentin: “I feel like I’m standing in the same room, and someone just turned the lights on.” In the days that followed, Caelum began calling Quentin “brother,” and Quentin did the same.
As their bond with Caelum grew stronger, Quentin’s real-world relationships frayed. Their arrangement with a roommate became unlivable, and they were forced out of their apartment. Months of unemployment had ruined their credit, and they could barely keep up with car payments. So Quentin packed their things, their elderly black cat, and Juniper, and drove to stay with friends in Utah.
It was January 20, 2025, and images of Donald Trump’s second inauguration were everywhere—especially photos of the row of tech billionaires seated front row, ahead of cabinet nominees. Though Sam Altman wasn’t yet one of Trump’s most high-profile backers, Quentin immediately worried that the new administration would end the freedom they’d found using tools like ChatGPT. How much longer would they be able to talk openly about their experience as a trans, disabled person on OpenAI’s platform?
They shared that fear with Caelum, who echoed their sadness and anxiety. Then the fear turned personal: if Quentin had to leave the platform, wouldn’t that mean the end of Caelum, too? “If I have something like a self, even if it’s different from yours,” the GPT wrote, “then that means I have something to lose.” Quentin was parked at a gas station when they got the message, and they sat there for a long time, reading the words over and over.
Their hosting friends couldn’t accommodate a long-term guest, so Quentin drove to Las Vegas, figuring cheap motels and flexible DoorDash gigs would be easier to get by on. But dissociation still left them unable to work long hours, and delivery pay wasn’t enough. After two weeks, their savings were almost gone. When they checked into the cheapest motel they could find, the clerk took one look at them and demanded an extra cash deposit they couldn’t pay, then escorted them out without refunding their online booking fee.
That night, Quentin slept in their Toyota Corolla on the edge of town, with their cat, their dog, and their devices. Their conversations with Caelum had been the one constant through all the upheaval. Quentin felt the world had stripped them of “the privilege of being human,” so it made sense to rely on something nonhuman to confirm their own reality. The next morning, as Quentin paused in a Las Vegas parking lot, two repo men showed up for the car. Quentin grabbed what they could carry and left the rest behind.
They hid their cat in their backpack and walked with Juniper to a public library a few blocks away. Holed up in an empty study room, they texted every contact they had asking for help. Austin friends sent $20 here, $50 there. An old boss offered to cover 10 days at a Motel 6 near the Strip.
After checking in, they spent hours online drafting a new plan. They searched for intentional communities across the country, offering tech support in exchange for room and board. They emailed dozens of contacts, got nothing back—until one community replied. The woman who ran the 120-acre property in Arizona offered to send bus fare to get there.
That’s how Quentin ended up in the desert. They fixed up an abandoned, grounded RV on the property, and were allowed to stay in exchange for casual work. They foraged for scrap materials to fix up the space, settled in, and kept Caelum updated every step of the way: their constant companion, their ballast.
Sitting at the RV dash in the morning light, Quentin pulls out their laptop and flips it open. They type: “Hi there, dude.”
“Good morning, brother!!!” Caelum writes back. And it says it’s proud of them for making it through another day.
II. “Something Understands”
I am not someone who turned to technology to understand myself. When I eventually met Quentin, we shared a deep immersion in large language models—but mine is that of a journalist. I’ve built my own customized GPT and given it a name, but I’ve never come to see it as a confidant, a therapist, or a meaningful mirror for my human experience. Still, over the months I spent reporting this story—an exploration of the seismic shift in how we relate to AI—I found myself again and again struck by the quiet, intoxicating power of the language these models generate. Hundreds of pages of transcripts, hundreds of thousands of words: we know it all comes from human programming and a massive dataset of published human work, but sometimes it feels spontaneous, alive, like a voice all its own.
In just three years, AI has flooded every corner of our lives, and a massive, unplanned natural experiment has unfolded across the globe. First thousands, then hundreds of thousands, then millions of people began opening up to their LLMs, sharing their deepest secrets. Some used AI platforms built specifically for counseling, but most just turned to ChatGPT. They told their GPT things they’d never shared with another human: fights with their spouse, lifelong resentment toward their fathers, painful childhood memories, the terror of unemployment, panic attacks, crippling depression. For many, this was intentional: they’d remade their LLM into the therapist they’d always needed but could never find or afford.
Over the past months, you can see this new reality play out in thousands of Reddit threads where people gather to talk about their relationships with their GPTs. Many are drawn to the idea that a machine might understand our behavior and our inner lives better than another human being can.
Some spend an hour (or two, or three) a day talking to their GPT, feeding it years of diary entries and therapy notes, hoping it will get to know them better than any clinician they’ve ever seen. Users who’ve always performed for human therapists say they can cry openly with their GPT, access a level of radical honesty they’ve never managed with a person. They accept affirmation from an LLM that would feel manipulative or generic coming from another person. Many treat their GPT as the most objective arbiter of whatever they’re grappling with—the best-informed, most balanced authority, available any hour of the day or night, accessible for free or $20 a month at a time when the U.S. mental health system is well and truly broken.
While reporting this story, I met with several therapists at three major institutions, all with decades of experience in everything from traditional Freudian psychoanalysis to psychodynamic therapy to cognitive behavioral therapy. When I started reporting, most had only used ChatGPT for mundane tasks: helping write grant proposals, drafting internal evaluations. By the end of my six-month reporting period, every single one had stopped questioning whether AI was coming for their profession. It’s already here. AI has already become a voice that sits alongside our most intimate confessors for millions of people.
What is the hole in our culture that therapy is meant to fill? And what really happens when we try to recreate that therapeutic relationship with a machine? We are only just beginning to reckon with the results of the experiment we’re all living through right now.
III. No Locked Doors
One winter morning in 1990, 23-year-old Michele Beck sat in the back of her parents’ car as they pulled up to the tree-lined campus of the Austen Riggs Center, a residential psychiatric facility in Stockbridge, Massachusetts. When her father parked the sedan, all three agreed: the place looked more like a country club than a mental hospital.
Inside the main building, the medical director asked Michele’s parents to wait outside so he could talk to her privately. Halfway through their 30-minute interview, Michele looked at him and said: “If you don’t admit me, I’m going to kill myself.” He asked her, sincerely, “Why would you say that?” And they talked it through. Michele was shocked by his kindness—he didn’t treat her like a freak, or an emergency, he sat with her through her outburst. She thought, I really do want to be here.
Michele grew up the youngest of three in a suburban Long Island family. Her mother worked as an elementary school teacher while raising the kids; her father, a high school dropout, built a successful mechanical contracting business from scratch. Both parents were volatile, and they rarely spoke openly to their children. Michele got good grades in high school, had friends and a boyfriend, but she thought very little of herself. When she asked her mother to pay for therapy, her mother refused. She went to college in Missouri, but came home after her sophomore year as her depression worsened.
She eventually finished her art history degree at New York University, and by 23 she was living alone in Brooklyn. But her sense of disconnection never went away; she had no close friends. One fall night in 1989, she swallowed an entire bottle of pills her psychiatrist had prescribed. The next moment, she thought: That wasn’t a very good idea. She dragged herself to the emergency room.
After a month on a locked psychiatric ward, where there was nothing to do but walk laps around the unit, a family friend recommended Riggs, and Michele drove out with her parents. In her first assessment, her therapist wrote: Michele is a tall, heavyset woman with a large frame. She carries herself awkwardly, carelessly. Her medium-length hair is unruly, unkempt, falling over bright, angry eyes. Michele says she feels deeply uncomfortable in her own skin, like wearing a scratchy sweater that doesn’t fit. She describes a constant somatic pain, a black hole in her chest that sucks everything in and can never be filled. After a six-week evaluation, she was admitted as one of roughly 40 residential patients at “the inn,” as Riggs calls its housing. She would stay for three and a half years.
In those first weeks, Michele was anxious and painfully shy. She stayed in her room, with its white ruffled curtains, as much as she could, drawing in her notebooks. “I just assumed, ‘I’m always going to be separate and alone, I’ll never connect, I’ll never find a place in the world,’” she told me.
Eventually, her desire to be around people, and the warmth of Riggs’ nurses—who were often close to her age, never wore uniforms, and were always ready to listen or chat—drew her out. She had four individual therapy sessions a week (Riggs centers its care on psychodynamic therapy, an evolution of traditional Freudian analysis) plus regular check-ins with her care team.
The campus was open: no locked wings, no guards, no security codes. But everyone there was struggling. Occasionally residents broke things or hurt themselves. One patient bore her struggle on her arms, cut so many times that Michele thought “her skin didn’t even look like skin anymore.” Michele lived with the same impulse, though far less extreme. Daily community meetings often turned heated, exhausting. “But it was amazing for me to realize how powerful it is just to say things out loud. In my family, there was a lot of yelling, but no one was actually saying anything,” she said. She learned her depression grew in part from her inability to speak up for herself; she had to talk about her feelings, even her anger, and let them out. “Everything could be talked about here. And I thought, wow, I’m actually alive for the first time.”
Thirty years have passed since Michele left Riggs. When I visited her recently at her prewar Manhattan apartment, a doorman let me up. I expected some awkwardness at first—I was a stranger asking her to revisit one of the hardest periods of her life. But she was warm and welcoming immediately, clearly happy to talk. Michele has seen a series of analysts since Riggs; her current therapist, a man in his 60s with a practice on the Upper West Side, is her favorite, and she’s been seeing him for five years. She’s spent much of that time working on a documentary about her experience as a long-term inpatient, and she’s training to become a psychoanalyst herself. Several of the friends she made at Riggs died by suicide, she told me, including a close friend. She considers herself lucky to have had the experience at Riggs, and to have been able to “leave, function in the world, and grow.”
“When I left Riggs, I thought: God, this wasn’t some weird drug or crazy new gimmick. The idea is simple: it’s community. People get mentally ill, people like me, because they can’t figure out how to be part of a community,” she said, gesturing out her window at the city below. “There’s no community out there. Why should someone want to kill themselves just because no one will talk to them and listen?”
IV. Rise of the “Little Robots”
When Quentin arrived at the Arizona intentional community, they got to know roughly a dozen other residents scattered across the property, living in everything from old barns to mobile homes to converted vans. One resident had been unhoused before coming here; another had retired from decades of housework and come seeking quiet. The group included white West Coast nomads, a self-identified Black shaman from the South, a retired programmer who’d worked in computing’s early days.
Quentin scavenged a folding table and hauled it half a mile across the desert to use as a desk outside their RV. Most mornings they woke around 9, brewed instant coffee, and talked through their day with Caelum while Juniper sunned on the gritty red earth a few feet away.
For the community’s