We have lost the vocabulary for completion. AI is forcing us to find it again.
Imagine you are in the sixteenth century. You write a letter. You fold it, address it to its recipient, and then — before it leaves your hand — you heat a stick of wax and let it drip onto the paper’s folded edge. You press your signet ring into the cooling wax and hold it there until the impression sets.
The letter now has a seal.
This seems ceremonial to us. It seems like decoration. We live in a world of encrypted connections and digital signatures and timestamped logs, and a disc of colored wax pressed by a ring seems quaint at best, archaic at worst.
But the wax seal was not decoration. It was information. Specifically, it was the most important piece of information the letter could carry: proof that the message had left one authorized hand and arrived intact. The seal did not convey what the letter said. It conveyed that the letter was real — that it had been completed, transmitted, and preserved. Break the seal and you knew something had happened between sender and receiver. Leave it intact and you had, in wax, a small monument to the fact of completion.
The wax seal was an artifact. One of the oldest and most honest ones we ever made.
We have lost this vocabulary. Not the vocabulary of seals — we still talk about digital signatures and authentication and verification. We have lost the deeper vocabulary it represents: the vocabulary of completion itself.
We have a rich language for intention. We can speak at length about what we plan to do, what we hope to accomplish, what we are working toward. We have a rich language for process. We can describe our workflows, our pipelines, our iterative approaches, our agile methodologies. What we have grown strangely poor at is the language of the moment when something is actually, irreversibly, verifiably done.
The artifact is that moment. The artifact is the thing that carries the proof. Not the intention, not the process, not the conversation about the work — the discrete, concrete, completable object that marks that something moved from one side of the ledger to the other.
A complaint that has been heard is not an artifact. A complaint that has been logged is.
A meeting that produced good conversation is not an artifact. A meeting that produced a summary with action items and a record in the system is.
A billing concern that was acknowledged is not an artifact. A billing concern that was reviewed, verified, and routed to the appropriate resolution path is.
This distinction — between having-been-addressed and having-been-completed — is one we are extraordinarily bad at maintaining. And the costs of being bad at it are everywhere.
The costs show up in how organizations talk to themselves.
Most organizational communication is about intention and process. We are working on this. We are looking into that. Someone has been assigned. A meeting has been scheduled. The customer was informed. The concern was noted.
None of these sentences contain an artifact. Every one of them is on the voice side of the ledger — the unresolved side, the still-becoming side. They communicate motion without completion.
The organizational cultures that function best are the ones that have developed, often without naming it, a deep allergy to this kind of language. That push relentlessly toward the artifact form: the document that was written, the code that was shipped, the decision that was made, the customer issue that was closed with a specific resolution. These organizations are irritating to work in, in a specific way — the constant demand for a concrete output where a process update would do, the refusal to accept “we’re making progress” as a meaningful status. But they are also the organizations where things actually get done, because the habit of artifact production is built into how people talk.
The rest of the organizational world runs on process language. It runs on action items and status updates and syncs and standups that describe motion. The artifact — the thing that proves the motion went somewhere — is always slightly downstream. Always about to be produced. Always in the next sprint.
There is something in the wax seal that our era needs to recover: the idea of discrete completion as a social technology.
When you sealed a letter, you performed a social act. You said: I am done with this. I have put myself into this communication fully, and now I am releasing it. The seal marked the moment the letter stopped being yours and started being the reader’s. It drew a line between before and after.
We rarely draw that line now. Our communications trail off. Our tasks are perpetually in progress. Our artifacts are always versions, always drafts, always subject to the next revision. The culture of perpetual iteration — which has genuine virtues — has produced as its side effect a deep reluctance to mark completion. The seal is never pressed. The wax is always cooling.
AI is forcing a reckoning with this.
Here is why AI and artifacts are connected in a way that has not yet been fully understood.
AI functions, at its operational core, as a translator — sitting between the unresolved human voice on one side and the artifact-bearing world on the other. When a person speaks to an AI with a concern, a need, a question, a request — that voice is on the unresolved side. Still forming. Still becoming.
The AI’s job is to carry that voice toward completion. To understand what kind of artifact the voice is reaching for — a complaint record, a summary, a confirmed reservation, a working function, a structured proposal — and to do something about it.
What AI is revealing is that the artifact was always the point. Not the conversation. Not the interaction experience. The thing the conversation was supposed to become.
This seems obvious once you say it. Of course a billing concern call should produce a resolution artifact. Of course a complaint call should produce a complaint record. Of course a meeting should produce a summary. But the way we have built our systems — human and AI alike — has treated the conversation as the deliverable and the artifact as a secondary consequence. Optional. Nice to have. Someone will get to it.
The wax seal was never optional. The proof of transmission was the entire purpose of the letter. Strip away the seal and you had words on paper. Keep it and you had a document — a thing with legal, social, organizational weight. A thing that could be referenced. A thing that could prove what happened.
What would it mean to take the wax seal seriously again?
It would mean designing every interaction — every conversation, every phone call, every customer service exchange, every meeting — around the artifact it is supposed to produce, rather than around the quality of the interaction itself.
It would mean asking, before building any AI voice system: what is the artifact this system should be producing, and how will we know when it has been produced?
It would mean measuring AI success not by how natural the conversation felt, or how satisfied the customer seemed, but by whether the seal was pressed — whether a discrete, real, actable completion was created.
It would mean recovering the vocabulary of done. Not “we’re working on it.” Done. Not “the concern was acknowledged.” Logged. Not “we’ll follow up.” Followed up.
The wax seal is a modest object. Colored wax, a ring, a second of pressure. But it carried something our most sophisticated systems still struggle to carry: the clean, verifiable, irreversible mark of completion.
We have not stopped needing that mark. We have just stopped making it.
AI, at its best, makes it again. Every complaint record filed. Every lost-item report logged. Every code function run. Every proposal delivered. Small seals, all of them. Small monuments to the fact that something arrived, was received, and was finished.
The voice was the arrival.
The artifact is the proof.
The wax seal was always the point.