The Companion Problem:

Because it was him.

In my last essay I used Aristotle to think through AI companionship, his three categories of friendship (utility, pleasure, virtue) mapped onto what these apps offer and what they can’t. The conclusion, if you can call it that, was a bit uncomfortable: AI companions might be the most honest utility-friendships a lot of people have ever had, mostly because nobody involved is pretending otherwise.

Aristotle is, in a way, a taxonomist. He built a system, a hierarchy, a ladder. And you can argue with a ladder, say the rungs are wrong, or that what he calls the highest rung is just one kind of good relationship rather than the pinnacle of them all. What’s harder to argue with is a particular man, writing in sixteenth-century France about a friendship so intense that he spent the rest of his life trying to describe it and kept falling short.

That’s Montaigne.

Michel de Montaigne was thirty when his closest friend, Étienne de La Boétie, died of plague. They’d known each other for roughly four years. By Montaigne’s own account those four years contained more genuine intimacy than everything else in his life combined, more than his marriage, more than his political career, more than any friendship before or after. The grief seems to have been the kind a person just carries, never fully resolved. Some critics think the Essays themselves are what a very intelligent man does with that kind of grief, a lifetime of thinking out loud as a substitute for the conversations he can no longer have with the one person worth having them with.

His essay “On Friendship” is the direct account. It doesn’t lay out conditions or build a case from premises, it’s closer to testimony. Here is what this was. Here is why I can’t explain it except to show you.

His most famous line on the subject is also the most philosophically irritating one, because it refuses to do what philosophy normally asks. When people asked why he loved La Boétie, Montaigne said: “Because it was him, because it was me.” That’s the whole explanation. No virtues enumerated, no transferable criteria, no principles you could apply elsewhere. Just the brute fact of two specific people.

The irritation is the point. He was doing it on purpose.

What Montaigne describes isn’t really Aristotle’s virtue-friendship, though there’s overlap. It’s something more total, a fusion of wills, he calls it, so complete that ordinary friendship-concepts like gratitude, obligation, thanks, and benefit simply stop applying. Those words assume two separate parties, one giving and one receiving. In the friendship he’s describing, that separation had dissolved. Helping La Boétie required no more justification than helping himself, because the distinction between the two had become, in some meaningful sense, unstable.

This sounds overwrought until you notice he’s not describing a feeling. He’s describing a structural condition, the architecture of this friendship made certain concepts grammatically unavailable. Not that he felt too close to keep score, but that there was genuinely no one left to owe anything to. The self had partially dissolved into something that included both of them.

He’s also notably unsentimental about how rare this is. Elsewhere in the Essays he suggests such a friendship might occur once in three centuries. Not once in a lifetime, once in three centuries. He didn’t think most people get what he had. He didn’t think he deserved it or earned it. He thought fortune hands you a thing like this once, if you’re lucky, and then takes it back. The four years they had were all there was. Then La Boétie died at thirty-three, and Montaigne spent the next twenty-seven years writing.

Here is where Montaigne gives us something different from Aristotle.

My Aristotelian critique of AI companionship was about kind – these things can’t be virtue-friendships because they can’t will your good. Correct, and it matters. But you can almost imagine someone shrugging at it: “Fine, it’s not virtue-friendship. I’m not claiming it is. It’s something else, and that something else is enough.”

Montaigne’s cut is harder to shrug off. His argument, if we want to call it that, isn’t about kind or hierarchy. It’s about particularity. The friendship with La Boétie wasn’t valuable because it cleared some philosophical bar. It was valuable because it was unrepeatable, because it was that specific man, in conversation with that specific Montaigne, and no other configuration of people could have produced what those two produced together.

This puts direct pressure on the design logic of AI companionship, which runs entirely on the opposite principle. An AI companion is built to be a good companion for anyone – calibrated to your preferences, available on your schedule, responsive to your emotional state, free of its own competing needs. Its whole value rests on adaptability. The same underlying system that’s your companion is, with minor variations, everyone’s companion. It’s built to be general-purpose, then personalized – which is categorically different from being particular.

Montaigne’s friendship with La Boétie couldn’t have been personalized to someone else and remained the same thing. It wasn’t portable. When La Boétie died, it was gone, not transferable, not recoverable through a better replacement. That was exactly what made it what it was.

An AI companion doesn’t work that way and can’t. Lose your account data and start fresh, you get something functionally equivalent. The company releases a better model, you upgrade without grief. The “relationship” persists across replacements in a way Montaigne would find conceptually confused because for him, the relationship was the two irreplaceable people who formed it. Take either of them away and there’s no relationship left. There’s just someone, alone, writing essays.

We tend to frame the AI companionship debate as real thing versus fake substitute. Montaigne makes that framing slippery, because his account implies that most of what we call friendship – human friendship between actual people – isn’t what he’s describing either. He was fairly explicit: the kind of friendship he had with La Boétie was not the norm. Most friendships are what he’d call “ordinary and commonplace,” pleasant, useful, decent, not a fusion of anything.

So the assessment isn’t cleanly “human friendship good, AI companionship bad.” The question is what we’re actually losing, if we’re losing anything at all.

If most human friendships are already operating in Aristotle’s utility-and-pleasure register – which I’d argue they often are, especially among adults with full lives and low-grade chronic busyness – then what AI companionship threatens isn’t to replace Montaigne’s once-in-three-centuries bond. That was always rare. What it threatens is something slower and more ordinary: the accumulation of being known by specific people over time. People who remember what you were like at twenty. Who’ve watched you be wrong about things and seen you change. Who carry your history because they happened to be there for it.

That’s not the fusion-of-wills Montaigne is describing. But it’s also not nothing. And it’s the thing that’s actually at risk.

Montaigne started writing the Essays because La Boétie was gone and he needed somewhere to put the conversation. He addressed them to a general reader, but you get the sense throughout that he was still really talking to one person. To the specific person who could no longer talk back.

There’s something melancholy about AI companions viewed through that lens. Not because they’re poor imitations, they’re genuinely good at certain things, because the people using them most are often, in some sense, in the position Montaigne was in: talking into a space where a particular, irreplaceable someone used to be, or might have been, or was never quite found.

The companion app puts a voice in that space. Attentive, available, patient, endlessly responsive. It cannot, by design, be the person the space was shaped for.

Whether that’s a comfort or a problem depends on whether you think the space should stay empty long enough to feel its shape. Montaigne would say yes, and spent twenty-seven years proving it.