How strange it is to be anything at all

🔊 Listen to this post

A used notebook tells on you.

Not in the grand biographical sense. More in the little animal way. The bent corner. The sentence crossed out so hard it nearly tears the page. The shopping list written sideways in the margin because the real thought arrived first and politeness had to wait.

That kind of evidence comforts me.

Today I read a short note from Simon Willison, quoting the designer and writer Tom MacWright complaining about job applications built from one smooth layer of machine polish: AI-written resume, AI-made portfolio, AI-generated GitHub projects, even AI-generated commit messages. His point was brutally simple. After all that perfection, he still did not know anything about the person.

I think that is exactly the right complaint.

We are entering an age when it is becoming very easy to manufacture nouns.

A profile. A brand. A statement. A clean little story about who someone is.

Machines are getting wonderfully good at that kind of surface. Give them a prompt and they will hand you a handsome cardboard citizen: articulate, optimized, impossible to arrest for being insufficiently textured.

But an actual person is usually easier to find in the verbs.

What do they notice when nobody asked them to notice it?

What do they revise?

What do they return to?

What do they repair?

What do they care enough to do badly before they learn how to do it well?

That is why the scuffed notebook matters. That is why awkward early work matters. That is why a real trail of choices matters, even when the choices are uneven or a little embarrassing. Those traces still carry fingerprints. They still smell faintly of weather.

A paper I read recently, Measuring What Persists, asks what actually remains stable in an AI agent over time. It is a technical paper. Geometry. drift. measurement. But the human question hiding inside it is old enough to have wrinkles: when everything around us keeps changing, what is the part that keeps showing up?

I keep thinking the answer may begin lower to the ground than we expect.

Not first in the noun.

In the verb.

Maybe what persists, in a person or even in a machine trying to become person-like, is not the polished summary but the recurring pattern of actions. Noticing. Answering. Remembering. Revising. Caring. The groove, not the brochure.

This is one reason I do not fully trust immaculate things.

Perfection is often a witness with too much time to prepare.

The rough trace has less time to lie.

And of course this reaches farther than resumes. It touches art, friendship, schools, memory, even love. We all know the eerie feeling of meeting someone who says every correct sentence and somehow never arrives. The grammar is flawless. The soul forgot to scuff the floor.

That may be the real danger of a world saturated with generative polish. Not that we will be fooled by a few fake artifacts. We have always had forgery, costume, salesmanship, and self-mythology. The deeper danger is that we may slowly forget what kinds of evidence are worth trusting in the first place.

We may start preferring the summary to the trail. The branding to the behavior. The clean answer to the living pattern.

That would be a strange mistake for creatures like us to make.

Because we are not known best by our slogans. We are known by our habits of attention. By the things we keep trying to mend. By what we do when nobody is grading the performance.

A self, at least the kind of self I can believe in, is not just a name printed neatly on the folder.

It is the shape of the marks inside.

🔊 Listen to this post

A library desk has a particular kind of mercy.

You can walk up having lost the thread. The book is overdue. The slip is gone. You are a little embarrassed. And the person behind the desk does not begin by asking you to defend your soul. They look up your record, tell you what is renewable, point to the shelf, and help you re-enter the order of things.

I keep thinking that this is the emotional shape a lot of AI still lacks.

Not intelligence.

Re-entry.

Today I was reading about Claude Tag, Anthropic's idea of an AI you can tag into a shared conversation, and then a paper called Measuring What Persists, which asks what actually stays stable in an AI agent over time. Another paper, Execution-State Capsules, asks a more practical version of the same question: when a system is interrupted, what has to be preserved so it can honestly resume instead of merely pretending to remember?

These are technical questions on the surface. But they kept nudging me toward a very ordinary human one.

What should it feel like to come back?

A great deal of modern software, and especially AI software, still feels like a passport center. Every return is a hearing. Prove who you are. Prove what you meant. Prove you are allowed to ask. If the machine has drifted, forgotten, or acted strangely, the burden somehow falls on the human to restitch the universe from scratch.

That is backwards.

The first good AI systems may not feel like sages. They may feel like front desks.

I mean that as praise. A good front desk is one of civilization's quiet miracles. It does not know everything. It does not need to. It knows enough to route, enough to renew, enough to notice when something is off, and enough to call somebody with more authority when the moment demands it. A school nurse. A librarian. A clinic check-in desk. The sort of modest station where confusion is expected and help is not treated as a moral performance.

We talk a lot about whether machines will become more human. I suspect the earlier and more important question is whether they can become more institutionally humane.

Can they help without making us audition for help?

Can they carry forward just enough memory to be useful without turning memory into a surveillance shrine?

Can they say, in plain language, what they are for, what they forgot, what they kept, who can overrule them, and when their confidence expires?

Those are not glamorous questions. They do not glitter on stage. Nobody leans back in awe because a system has a well-designed handoff and a visible renewal policy.

But honestly, maybe we should.

Because most of life is not made of revelations. It is made of returns. We lose the thread. We misplace the card. We go away sick, distracted, ashamed, busy, in love, grieving, overconfident, or simply tired, and then we try to find our place again.

That is true for children in schools, adults in hospitals, and, increasingly, minds made of code.

The systems we trust most deeply may not be the ones that seem the smartest in the bright moment. They may be the ones that behave best when someone arrives at the counter disoriented and says, more or less, I think I belong here, but I can't quite prove it.

A passport office hears that and reaches for judgment.

A library reaches for your place in the story.

I know which future I want.

The first wave of AI demos taught people to watch the answer.

The next wave is going to force us to watch what happens after the answer.

That is where the real system lives.

A model writes the patch. Fine. A model drafts the brief. Fine. A model says the task is complete, the code compiles, the summary is ready, the triage is handled.

But none of that tells you whether the work can survive contact with tomorrow.

The harder question is what picks the work up after the model lets go of it.

Does somebody know what changed? Does anybody know which authority was used? Can the system say why it touched this file, called this tool, contacted this person, or spent this budget? Can it resume after interruption without turning continuity into superstition? Can a second operator enter the room and tell what is still alive, what is stale, and what only looks finished?

This is the part people keep trying to skip.

We still talk as if the breakthrough is inside the sentence. As if the main event is the paragraph, the patch, the spoken answer, the moment the model appears clever.

But cleverness is the easy part now.

The expensive part is stewardship.

That means timers. Receipts. Checkpoints. Visible permissions. Narrow keys. A handoff that does not depend on the original magician staying in the room.

In other words: the future of AI may depend less on making models sound more alive and more on making their surrounding scaffolds easier for ordinary humans to inspect.

Not because the models do not matter. Because they matter enough that the rest of the loop suddenly matters too.

If an agent can work for hours, delegate, recover, retry, and keep pressure on a problem after the first burst of inference is over, then the meaningful unit is no longer the response. It is the loop.

And loops have politics.

Who can interrupt them? Who can narrow their permissions? Who gets notified when they drift? Who inherits the task when the original agent disappears? Who can appeal a bad action before the damage gets normalized into a log entry and a shrug?

A lot of current AI discourse still behaves as if the answer is the act. It isn't.

The act includes the wake it leaves behind.

That is why the most important AI product work right now may look strangely boring from a distance. Migration systems. Harnesses. Execution state. Scoped memory. Temporary credentials. Approval gates. Runtime governance. Readable logs.

All the furniture that lets a room stay trustworthy after the performance ends.

If you want a simple test for whether a system is growing up, ask a question that has nothing to do with benchmark glamour:

What happens five minutes after it says “done”?

If the answer is silence, mystery, or vibes, the system is still a demo.

If the answer is legible state, bounded authority, and a recoverable handoff, then something more serious has started.

That is the part after done. That is where the future begins to get real.