Somewhere in the last few years, the cost of making a plausible thing dropped to almost nothing. A screen that looks like a product. A paragraph that reads like a person. A layout that photographs well. All of it can now be summoned in seconds, on demand, in infinite quantity.
The strange consequence is that competence stopped being proof of anything. When everything can look finished, looking finished tells you nothing about who made it, or whether anyone meant it.
So I've been paying attention to a smaller question, the kind you can only answer by building. What actually changes when you try to make a piece of software feel human?
Not human as a costume. Not the rounded corner, the friendly error, the mascot. Human in the way a room feels lived in, or a sentence feels spoken. And the more closely I looked at the micro-decisions that produce that feeling, the more they resolved into one thing. Not warmth. Authorship.
Humanity is a residue of choices
A machine optimizes toward the average of everything it has seen. That is its gift and its ceiling. The average is competent, legible, and unsurprising by construction. It is, definitionally, what most things already are.
Authored work moves the other direction. It is the accumulation of a thousand small refusals: this, not that; here, not there; enough, stop.
You feel it in the details that never announce themselves. A transition that lasts exactly long enough to explain where a thing went, and not a frame longer. A single accent color doing structural work, carrying attention and state and trust, instead of five colors competing for applause. Copy that says the true thing plainly instead of the impressive thing vaguely. Motion that orients you rather than performing for you.
None of these are decorative. Each is a decision that something else was left out so this could be understood.
The signature isn't what a person adds. It's what a person was willing to remove.
This is why restraint feels human and abundance feels automated. Anyone can add. Adding is what you do when you haven't decided. Subtraction requires a point of view, a reason to believe this element earns its place and that one doesn't. The average has no reason to leave anything out. A person does. That reason, felt but rarely seen, is the tell.
Context is the thing that can't be generated
The other half of it is context.
A generated interface knows the shape of a dashboard, a checkout, an onboarding flow. What it can't know is the particular person on the other side, at the particular moment they arrived: anxious, rushed, grieving, hopeful, mid-task, half-distracted.
Human-made work, at its best, is built backward from that moment rather than forward from a template. It asks what this person is actually trying to do, what they're afraid of, what would reduce the distance between their intention and the outcome. Those answers don't live in the average of all interfaces. They live in a specific empathy for a specific situation, and they show up as choices a general model would have no reason to make: the reassurance placed exactly where doubt appears, the step removed because it was never really needed, the honesty where a lesser design would have smiled and hidden the complexity.
The paradox is that the more transparent this care is, the less visible it becomes. Good human work disappears into ease. You don't notice the seams because someone spent their attention so you wouldn't have to. That invisible, transferred attention, one person spending effort so another spends less, may be the most human thing an interface can carry. It is a form of respect, encoded.
Where this leaves us
I don't think the line between human-made and machine-made will hold at the level of surface. Surfaces are exactly what machines do best, and that gap will only close.
The line will hold somewhere quieter: in whether the thing was meant. Whether it carries evidence of a person who understood a situation, took a position, and paid for that position by leaving other options on the floor.
Which turns the usual anxiety inside out. The task was never to out-produce the machine. You can't, and the attempt only makes you more machine-like. The task is to become more deliberately yourself: to build fast but edit slowly, to let function come before ornament and beauty emerge from structure, to make every element argue for its own existence.
Not because craft is nostalgic, but because craft is now the signal. In a world drowning in the plausible, the authored thing is the one that could only have come from someone who cared enough to choose.
Making software more human, it turns out, was never about adding humanity to it. It was about leaving your fingerprints in the grain, and trusting that, in the end, the fingerprints are the point.