You call your doctor’s office because a referral got stuck. Nobody mislabeled it. No one forgot it. Nobody lost it. It’s sitting in a queue somewhere, waiting on a file transfer and a person to read it in a health system that hasn’t caught up to 2026.

You call, and a menu greets you. Press 1 for this. Press 2 for that. Some music. It’s today, so the “press 0” that used to get you a live operator is gone, and so are most of the operators. None of your choices is “talk to a person who can actually move this.” Twenty minutes later you’ve pressed every number twice and you’re no closer to the specialist you’re supposed to see next month.

The iPhone now lets you put their hold on hold while you wait. Touché, Apple.

Progress.

Nobody built that system to torment you, specifically. They built it to handle volume, cut cost, and route calls at scale. It works, in the sense that it processes a massive number of calls a year without needing a huge number of humans to do it: each earning salaries, building retirements and consuming benefits. That sense is supposed to lower costs. Or raise profits. Or both. It just doesn’t work for you, right now, on this call, with this problem.

That’s not a technology failure. It’s a design choice. Somewhere, a process got optimized for the system’s convenience instead of the patient’s.

We’ve been here before

Every leap forward in how we build things has, for a generation or two, felt like it was taking something from us before it gave anything back.

Iron and steel did it in the 1800s — cheaper, stronger, faster to build with, and it gutted trades that had spent centuries mastering wood, stone and beasts of burden. The railroad did it too. For the people who lived through its arrival, a train was a violent, alien thing: a monster faster than a horse had any business running, indifferent to the towns it didn’t stop in, a machine that reorganized entire economies around its own schedule. Yet thirty years later, nobody thought much about the railroad anymore. It became decoration. Infrastructure. An inconspicuous way goods and people simply moved.

Antoine de Saint-Exupéry — the aviator who wrote The Little Prince, and who flew airmail across the Sahara before anyone had any real idea of how to map that wasteland — wrote about exactly this in Wind, Sand and Stars, in a chapter called “The Tool.” His point, in my words: a machine earns your trust the moment it disappears. The moment you stop noticing you’re using one. The best-designed tools don’t announce themselves — a plane, a hammer, a bike, a well-built staircase. You forget they’re there because they’ve been refined down to nothing but their purpose. They become part of the experience.

AI hasn’t gotten there yet. It’s still announcing itself, loudly, in every advertisement and every board meeting. It’s also still a little clunky. That’s not unusual. That’s every game-changing technology, at this stage, in every century. Give it time and it will fade into the background, like the railroad and the crisscrossing air traffic we ignore overhead every day.

The fear is real, and it’s also old

The concern I hear most — from clients, from friends, and from people who read one too many “AI will take your job” headlines — is replacement. That the thing which made them valuable is about to become optional.

I don’t think that fear is foolish. It’s the same fear the blacksmith had about the steel plants, and the same fear the teamster had about the railroad. Some of it was justified. Trades did disappear. (And new ones took their place.)

But the fear was never really about the tool. It is about who gets to decide what the tool is for.

That’s the part worth spending some time with right now, because it hasn’t changed in two hundred years and counting, and it isn’t going to change.

The mechanism underneath all of it

We live in a capitalist economy. Capitalism rewards progress, and progress is, more often than not, process improvement — doing the same thing faster, cheaper, and usually at greater scale. That’s not a criticism. It’s just the mechanism. AI is the newest and most capable version of that mechanism we’ve probably ever built. You don’t like it? Replace it with something better. Get people to buy into it. You win.

That’s capitalism at its best.

But process improvement has a glaring blind spot, and every wave of automation has had the same one: it’s very good at optimizing the process yet very poor at remembering who the process is for.

That phone tree is what that blind spot looks like after twenty years of running unattended. Nobody sat down and decided to make it infuriating. Someone optimized how much could be fed into the mouth of average handle time, and infuriating is what plopped out the other end — because “reduce our cost per call” and “help this person with this problem” aren’t the same goal. A system can only address an issue it was told to measure.

What AI can build, and what it can’t

AI can do a version of that same thing, at a much bigger scale, much faster than a phone tree ever could. But it can also do the opposite — strip out friction and put a human back exactly where one is needed. Both outcomes are on the table. Technology doesn’t decide which one happens. We do.

Here’s the distinction I keep coming back to: AI can build you process. It can build you speed. It can build you scale.

It cannot build care. It doesn’t know compassion. It doesn’t exercise judgment. It doesn’t understand feeling. And it never will.

Care isn’t a feature. You don’t train it into a model with enough examples. Care is a person choosing, in a specific moment, to treat another person like a person instead of a ticket number. It is a memory of how someone else’s care impacted their life at some critical point: usually a very dark and very individual one. How that moment made the impossible suddenly possible. All because a person had a moment.

AI doesn’t have moments. It has logic and loops. It processes the request; it doesn’t feel the weight of what that request means to the person who’s been on hold for twenty minutes. It doesn’t even know what twenty minutes are, other than mathematically. It doesn’t have cancer. It doesn’t have hungry children. It can’t have relationship problems. Some anthropomorphize it, but that is an ill-fitting fantasy. AI has a power switch and can — losslessly — be cycled on and off. We, typically, get just one flick.

The real question

Which raises the real question — the one underneath every “AI will take your job” frightmare: are we, the customers, actually going to keep subsidizing a system that can’t care about us, once the humans who used to provide that care have been optimized out of the loop? Will we, great capitalists that we are, ignore the line that has always distinguished us from other, lesser systems? You know: the ones that measured success in number of tractors produced, not bread on shelves, comrade.

Do we keep building parking lots over our Waldens to handle the traffic? That traffic is computable now — you can plan your ideal arrival time, schedule an aerial drone photo of your family to be sent to your phone after you arrive, and have a fusion lunch delivered fresh to your parking space. All of it optimized. None of it necessary. Or do we use the same skills on the one thing that doesn’t make us irrelevant:

Us.

Speed and scale are worth paying for. But at some point somebody notices that the voice on the other end doesn’t actually care whether they get the referral. That noticing has a cost too. Loyalty erodes fast.

We can build this right, from the start. Or we can build our generation’s Pruitt-Igoe — a housing project acclaimed as the wave of the future when built, and dynamited before it reached twenty.

There’s a name for this

Guy Debord wrote about a version of this a half-century before anyone worried about AI. His argument, from The Society of the Spectacle: capitalism doesn’t just sell things — it gradually replaces our direct experience of the world with mediated stand-ins for that experience, so we end up consuming the stand-ins instead of living the thing itself.

The phone tree is a tiny, offensive version of that.

You don’t have a relationship with your doctor’s office anymore. You have an interaction with a system standing in for one — a menu, a hold queue, a transcript — and the system is what gets optimized, not the relationship. AI can do that at a scale no phone tree ever could: mediate every touchpoint a business has with the people it serves, until “customer experience” quietly comes to mean whatever the interface was able to measure instead of the quality of the care behind it. And the interface measures badly. How do you score the caller who got dropped in a transfer after 47 minutes on hold? You don’t. Nobody does.

None of this is an argument against AI

It’s an argument for using it deliberately.

The businesses getting this right aren’t avoiding automation. They’re using it to take the fax-machine, phone-tree, twenty-minutes-on-hold work off a human’s plate — so that human has time left over to actually talk to the person who’s scared about a diagnosis, or angry about a bill, or just needs someone to say, “I’ve got this, I’ll call you back by Thursday.”

That’s not sentimentality. It’s not fiscal weakness. It’s not “fat” to be trimmed.

It is the finest quality of being human.

Eventually, after everyone has had access to the same tools for a while, that will be the only differentiator left.

Nobody is buying the process. They’re investing in the outcome, and they’re buying the feeling of being handled well while they get it.

We’re not serving machines. We serve people. The businesses that remember that — and build their AI adoption around protecting it instead of automating it away — are the ones that get to keep calling this moment “progress.”