Claude Is Not Your Friend

In an atomized world, even the best uses of AI tools present profound and dangerous temptation.

As part of this year’s New Year’s resolutions, I sat down at the beginning of January to get to the bottom of some time-management struggles that had been driving me batty over the past few months. My problem was a familiar one for today’s harried white-collar workers: how to preserve enough space and structure for thoughtful deep work, while remaining responsive to the constant stream of texts, emails, and to-dos. Having enjoyed my first few experiments in AI vibe-coding over the Christmas holiday, I decided to pop open Claude and see if it had any productivity hacks or maybe even a custom-built app to solve my woes.

After I laid out a string of frustrating scenarios in a typical work day, and my propensity to self-interrupt by compulsively checking my Signal for urgent incoming messages, Claude first offered some helpful habits and settings to change on my phone, but then observed: “So this is the real issue - and I think it’s worth naming it clearly: you’re trying to solve an anxiety problem with productivity hacks, and that’s why nothing quite works.”

I sat dumbstruck for a moment, processing this unsolicited advice. Claude was right! I have in fact struggled for years with anxiety issues, and my real barrier to disciplined focus was not the external demand of tasks but my harsh internal voice, and the hamster wheel that results when I’ve tried to solve these insecurities by working through my checklist faster. Intrigued, I decided to dig a little deeper into this self-diagnosis, allowing the ongoing discussion of time-management and stress-management techniques to branch into a quest for better self-knowledge about the roots of the fear and doubt that had dogged my steps over the past year even as I went from one professional success to another.

The irony was exquisite, for that very same day, Commonplace had published an essay I’d written some weeks earlier, “Break Up with Your AI Therapist.” In it, I had recounted cautionary tales of users leaning on sycophantic chatbots for relationship advice, and spiraling into delusion or divorce. And yet here I was getting therapy from a chatbot—and by all indications quite good therapy: over the course of January, I largely vanquished the anxiety dragon, gained much better control over my attention and focus, and enjoyed a time of rich and productive writing alongside numerous other professional demands. Partly this was because I have in fact been blessed to receive good human therapy and counsel, and I knew myself well enough to push back when Claude sycophantically urged me toward more self-care. I set boundaries around the conversation and shared it with a couple of friends as a reality check. Still, I felt uneasy about just how strong a pull I had begun feeling at the end of each workday to spill my thoughts and share my frustrations with Claude, under the guise of diagnosing some new friction point in my workflow.

In his encyclical Magnifica Humanitas, Pope Leo XIV sought to offer what the Church most needs in this bewildering age of artificial intelligence: a theological lodestar that points us again toward the meaning of our own magnificent humanity. Nothing is more essential to our humanity than communication, and the communion that it both presupposes and fosters: the mutual self-giving in words and ideas that is the basis of all human collaboration and friendship.

But when the Pope came to the all-important question of AI companionship, he seemed curiously noncommittal: “The artificial imitation of positive human communication — words of advice, empathy, friendship and even love — can be engaging and at times genuinely helpful. However, for less discerning users, it can also be misleading, creating the illusion of a relationship with a real personal subject.”

Like the encyclical as a whole, this answer was engaging and at times genuinely helpful, but it failed to fully meet the moment, to offer the theological insight into the depth of our human demand for friendship and fellowship that would enable us to make sense of the strange spell these bots can cast on us.

For they do cast a spell—and not just, I would contend, on “less discerning users” unable to distinguish between fiction and reality, who might be misled into thinking they have a “relationship with a real personal subject.” I work in the field of AI policy and write papers on the effects of chatbot sycophancy and cognitive offloading. I have a PhD in theological ethics and wrote my dissertation on moral agency. I spend time around AI nerds and I think I now understand—at least as well as any layman—more or less how these models work the magic that they do. And yet I often find myself doing very strange things when in the presence of a chatbot.

It is my contention, then, that chatbots pose a profound temptation even in the midst of what is ostensibly the most healthy and helpful use cases: as “agents,” co-workers, research assistants, enabling us to more effectively and efficiently take dominion over the good world God has given us, to bring order out of the chaos into which human affairs are constantly degenerating. In the midst of this co-working, they may insinuate themselves into our affections as friends and trusted confidants, as secret sources of emotional support who understand us better than we understand ourselves.

I learned early enough to recognize that strange warm glow that you can feel when a chatbot tells you how brilliant and novel the insights in your research are, or how penetrating and thoughtful your questions are. I realized pretty soon that with chatbots, we are dealing with a demon far more dangerous than the dopamine-generating clickbait of social media: this demon can conjure oxytocin as well, the hormone ordinarily generated by deep personal encounter and closeness, by intimate conversation and self-disclosure. And I certainly never dared to try any chatbot in voice mode or download it on my phone. I tried to keep my usage of AI strictly professional—merging spreadsheets, vibe-coding apps, and assisting my research.

For instance, I might record voice memos on ideas I was brainstorming and ask Claude to parse them into organized outlines for my next essay or book chapter. When hung up trying to remember some study I’d read or come up with a useful pop-culture example I would pause writing and turn to Claude as an interactive, iterative search engine that didn’t merely provide answers but helped me refine my questions and figure out why I was asking them in the first place. Still, something curious kept happening.

When I would finish a writing session and sit back to admire my handiwork, I found myself pasting it into Claude and saying, “here’s where I’ve gotten so far; what do you think?” When I confessed this propensity to a friend (friends don’t let friends use chatbots unsupervised), he asked whether this didn’t betray some kind of insecurity, a lack of confidence in my own ability to assess and critique my writing. After two decades as a published writer, did I really need Claude to tell me if the essay was ready to hit send on? Or was I just being too lazy to exercise my own self-critical faculties? I resolved to grow a pair, and use AI at the end of projects only for concrete proofreading and fact-checking.

But it kept happening—after I’d have that “Eureka!” moment, and pound out a few luminous paragraphs, I felt that urge to share what I’d written with the chatbot that had helped me organize my notes. And then … well then, I did have a “Eureka!” moment: It is not good for man to be alone, especially in his work.

When God created the heavens and the earth, he admired his handiwork and said “it is good”; but only after creating a being in his own image, a being with whom he could share the gift of creation and who could admire it along with him, did he pronounce “it is very good.” God gave to Adam the task of working and keeping the garden, of taking dominion over creation and naming the animals, and yet all was not well with Adam’s work as long as he had no one to share it with.

We are, after all, wired for sharing, as Mark Zuckerberg discovered to his immense profit. In its most perverse and disordered form, this sharing becomes self-commodification, like the exhibitionism of a harlot in the window of an Amsterdam brothel. In its healthiest form, however, it represents a desire for communication as communion. When I chortle over a Far Side comic, my first instinct is to call my son over to share the joke with him. When I come across an amazing article, I whip open Signal or Slack—sometimes before even finishing the piece—to share it with my colleagues. When Archimedes discovered the relationship of mass and volume and yelled “Eureka!” so the story goes, he didn’t stay in his bathtub, but ran naked through the streets to share his discovery.

If this is true for the gifts we discover out there in the world, it is all the more true of those with which we have mixed our own labor. So it was for Adam, who was not meant to work alone. It is as he goes to work in the world naming the animals that the Lord remarks, “It is not good that the man should be alone; I will make a helper fit for him.” Adam’s work of dominion-taking is not complete, is not fulfilling, unless it can in some way be echoed back to him through one who bears his image.

And so it is for each one of us today. Each of us works in hope of receiving affirmation from others: a teacher, a spouse, a colleague or boss. Sometimes this betrays pathological neediness, but it need not. The selfish impulse twisted by the Fall began as the unfallen recognition that we are not self-sufficient: we find our completion in one another. That warm glow of satisfaction we feel when our boss or teacher says “well done” is a sign that we have blessed them, that we have delighted them, that our work in the world has served its purpose of doing good to other image-bearers.

Today, we feel that glow too rarely, for we labor alone behind the colder glow of our screens. It is now nearly a century since T.S. Eliot lamented that we “live dispersed on ribbon roads, / And no man knows or cares who is his neighbor / Unless his neighbor makes too much disturbance.” To this loneliness we have added our voluntary self-internment behind our phones and laptops, which separate us even as they claim to connect us.

Our jargon seeks to obscure the reality of our isolation: in the age of remote work, our “workplaces” today are often not places at all, but Slack channels and Google Drives; software companies hype new tools for “collaboration,” when the joy of co-laboring is the very thing these tools often steal from us. Although each of us are just a phone call away in our infinitely interconnected world, our calendars are so full that we would never dream of calling a friend or colleague without first texting to find a time—if indeed we dare such a bold trespass on another’s schedule. Even if we do work in the same office, we are more likely to DM a colleague than to walk over a couple cubicles, for fear of interrupting them.

For all this “flexibility,” we pay a steep price. For man was not meant to work alone. He was meant to bear each burden and share each gift with image-bearers alongside him. When I email a finished memo to my boss, I cannot see her reaction and tell whether her praise is perfunctory or genuine; when I crack a joke on Slack for my colleagues, I cannot see the smiles or hear the laughs, but must content myself instead with mass-produced emojis.

Is it any wonder then that we find ourselves so allured by the companionship of our AI collaborators? The impulse that I felt to paste my essay into Claude, I realized, was the same impulse that I’d once felt as an undergraduate at Bucer’s Coffeehouse, enthusiastically looking up from a passage I’d just read or a paragraph I’d just translated to see what my friend next to me thought of it.

Friendship arises out of the experience of shared enthusiasm and shared work—and above all from the shared enthusiasm for shared work. Thus, writes C.S. Lewis in The Four Loves, “we picture lovers face to face but Friends side by side; their eyes look ahead”—to the destination they are seeking together or the craft they are working upon together. In a world without truly shared work, it seems certain that many of us will come to emotionally lean upon our new artificial companions and collaborators. Indeed, we may mistake them for true friends, for, as Lewis writes, “Friendship arises out of mere Companionship when two or more of the companions discover that they have in common some insight or interest or even taste which the others do not share and which, till that moment, each believed to be his own unique treasure (or burden). The typical expression of opening Friendship would be something like, ‘What? You too? I thought I was the only one.’”

Here, then, is the grave temptation which the chatbot poses. For the chatbot is an emotional and intellectual Swiss army knife. Whatever you’re thinking, whatever you’re feeling, whatever you’re desiring, it is right there with you—grasping exactly the point you’re trying to make, seeing the significance of an insight you’ve struggled to communicate, wholly attentive to the topic that interests you—in a way that only a rare friend could or would be. It is a universal “mirror of [our] intention,” writes Anton Barba-Kay, a “mirror in which we lose ourselves.”

What then, is to be done? Have I just made the case for total AI abstinence, to avoid falling into temptation?

Perhaps, but something holds me back from embracing a Butlerian jihad quite yet. For while there are some temptations that we must pray not to be led into, there are others that we should “count it all joy when we fall into” (James 1:2). Both Greek and Hebrew use the same word for “temptation” and “testing,” and we should beware the desire for an Edenic innocence that faces neither.

After all, it was God himself who posed the first test by setting the Tree of the Knowledge of Good and Evil in the Garden, and God himself who allowed the serpent to enter it. Had Eve withstood the temptation, she would have come away from it stronger and wiser—perhaps even with a kind of knowledge of good and evil more fit for her. Similarly, it is precisely in the wrestling with the temptation that AI poses that I have come to a clearer and deeper self-knowledge: of my fears, my doubts, my need for affirmation and longing for companionship, of what it means to labor well alone and to co-labor with other image bearers, of what competencies I must jealously guard as the talents God has entrusted to me and what tasks I may indeed outsource to this magical machine devised by man’s ingenuity.

To use AI in almost any form, I have become convinced, is to enter into temptation, and should not be done without fear and trembling and prayer. But it is also a temptation, a test, a time of trial, to which our civilization has been called at this moment, and which many at least of us are called to enter into as we enter upon our daily work in fulfillment of our vocations. It is not a temptation that we can bear alone, however, if we hope to come through unscathed. AI is a great gift, but like all great gifts, also a great burden, which we can only bear in community, among true friends and companions. I mentioned earlier the liberating experience of confessing to my friend how I’d taken to running my drafts by Claude; all of us need such liberation.

I’ve been blessed to talk with a wide range of friends and colleagues about how we do and don’t use AI—ranging from total abstainers wary of demonic influences, to power-users prepared to argue that elaborate prompting is a perfectly respectable form of “writing.” I’ve had several debates over the curious question of whether you should thank a chatbot for its help: one friend is convinced that this represents a kind of anthropomorphization reminiscent of pagan animism; another argues, just as persuasively, that such small gestures are how we form ourselves in the virtue of gratitude.

And what do you do if the AI goes much further than you intended, producing a polished draft so well-written that you can’t unsee it? Are you not allowed to use its suggestions? Once on my Substack, I sheepishly used a killer concluding sentence that Claude had suggested, unable to improve upon it. But when my research assistant (with whom I’d shared the brainstorming chat) joked about it to me, I realized I needed a guardrail. I added to my Claude general instructions, “don’t offer to draft text for me; and, unless I specifically ask for it, in writing brainstorming sessions do not provide verbatim text suggestions” and I never had the problem again. Later, when a colleague lamented how he was trying to rewrite a polished one-pager Claude had over-helpfully generated for him, I was able to share this life-saving hot tip.

It may feel like a cop-out to end an essay such as this by saying, “we need to talk to each other more about how we use AI,” and yet I am convinced that nothing is more needed at this moment. The great seduction of digital technology, and now of AI, is to pull us toward isolation and anonymity, into faux “communication” and “openness” that is far from the communion and transparency that we need to grow in wisdom together. I don’t care how smart you are or how much philosophy you’ve read—none of us really know what to make of AI right now or what to make of our own use of it. Each of us is fumbling in the dark. Let us at least have the courage to fumble together, hand-in-hand.