The slow death of a marketer's voice

Pull up anything a marketer published two years ago, then anything they published last week, and read them back to back. Often there’s a drift you can hear but can’t quite point at, the voice has gone a little smoother, a little more like everyone else’s, a few degrees toward the center. Nobody decided that. No single piece did it. It happened one translation at a time.

Here’s how it happens. You write one good thing, a clear argument, a real example, a line you were proud of. The writing took a morning. Then the second job starts, and it takes the rest of the week: the same idea as a six-tweet thread, a LinkedIn cut for founders, a 90-second script, a newsletter intro, three captions, a cold-email hook. Same idea, ten surfaces, and every one of them still has to sound like you. Each one is a hundred small acts of translation, each easy to do slightly badly, and the badness is invisible until it’s already published in your name.

That second job is the one marketers quietly dread, and it’s not because the thinking is hard, the thinking is done. It’s because you don’t have ten things to make; you have one idea that has to reach ten rooms still sounding like you, and nothing on the market was actually holding the you part steady while it traveled. That’s the pain we built from. Not “we’d like a repurposing feature.” A voice slowly averaging itself out of existence, five hundred low-attention translations a year, with no one and nothing standing guard.

A voice doesn’t break, it erodes

The reason this pain has no real answer in the market is that nobody experiences it as a crisis. A crisis you’d fix. This is erosion.

Voice drifts one word at a time. A “leverage” here, an “in today’s landscape” there, a rocket emoji nobody asked for, a hook that’s structurally a hook and emotionally nothing. None of it is wrong. Each piece is individually defensible, “it’s about the right topic, isn’t it?”, and collectively it’s a stranger wearing your name as a costume. The reader can’t articulate why they trust the feed a little less than they did last year. They just never catch you sounding like you twice in a row.

The numbers make the erosion lethal instead of merely sad. By one widely cited estimate, the average person sees somewhere between 4,000 and 10,000 ads and brand messages a day. In a flood that size, “structurally correct but voiceless” isn’t neutral, it’s invisible. The whole reason to have a voice is to be the one thing in the scroll that sounds like a specific person and not the average of everyone. Erode the voice and you’ve quietly deleted the only asset the work was building.

A voice doesn’t fail in one bad thread. It dies by averaging, five hundred small translations a year, each one a little more like everyone else.

So the pain isn’t “repurposing is tedious.” It’s that the act of reaching every room is the same act that slowly erases the person trying to reach them. That’s the thing we couldn’t find a tool to stop, because the tools all treat repurposing as a formatting problem, and it was never a formatting problem.

Why “make ten versions” is the wrong physics

Watch what nearly everyone reaches for, because it reveals the misunderstanding cleanly. You paste the finished post into a chat window. You type “turn this into a thread and a LinkedIn post.” Out come a thread and a LinkedIn post, numbered tweets, a hook, a “here’s what I learned.” Structurally fine. The boxes are filled, and filled feels like done.

Then you read them a day later, next to your real writing, and they’re off. The thread explains the idea to someone who already read the post. The LinkedIn version uses three words you’d never say. The caption is technically about your topic and emotionally about nothing. The tool reformatted your idea. It did not re-aim it, and re-aiming was the entire job.

That gap isn’t a quality bug you fix with a better prompt. It’s the wrong physics. A post-into-versions tool treats the post as the source and every other piece as a degraded photocopy: a thread is “the post, chopped up,” a caption is “the post, shrunk.” But a copy is not a translation. A photocopy keeps the wrong things and shrinks the right ones. You don’t have ten things to make; you have one idea that has to reach ten rooms still sounding like you, and a photocopier can’t do that no matter how good it gets, because it’s optimizing the wrong object.

So the thing the situation needed never was a “make ten versions” button. It was somewhere for the idea to actually live, separate from the first post it arrived in, and a set of surfaces that each pull from it in their own shape without losing the voice. That isn’t a feature you bolt onto a content tool. It’s a different center of gravity. And it turns out you only get it for free if the system was already built to be that kind of thing.

What falls out when the substrate is right

Here is the part worth sitting with: we didn’t design three clever repurposing tricks. The way this works is a consequence of what the OS already is, something that’s on, that remembers, and that’s permitted to do the same job a hundred times without getting tired. Point that at one painful job and three things fall out that a paste-and-reformat tool structurally cannot do.

The idea gets separated from the prose it first arrived in. The hardest part of repurposing isn’t writing the thread; it’s that the idea is trapped inside the sentences of the post, and a reformatter can’t tell the load-bearing part from the connective tissue. So before anything reshapes, the system pulls the post apart into what must survive every translation: the one thesis, the proof that earns it, the single example a reader will remember, the line too good to lose. That’s the canonical idea, not the post, the spine of the post. Now the thread is built from the spine, not the paragraphs; so is the newsletter, so is the script. Each is free to throw away everything else, the transitions, the throat-clearing, the second example that fit the long form but drowns a caption. A good thread stops reading like a chopped-up essay and starts reading like someone who understood the essay telling you the one thing that mattered, in thread-shaped sentences.

Each surface gets aimed at the room it’s actually in. Even with a clean idea, reformatting fails a second way: it speaks to every room as if it were the same room. It isn’t. The person scrolling at a stoplight is not the person who opened your newsletter on purpose with coffee. A thread reader wants the payoff in tweet one or they’re gone; a newsletter reader already raised their hand and will follow you for four paragraphs; the cold-email recipient owes you nothing and gives you one sentence to earn the next. Same idea, three different rooms, and a format that ignores the room is why repurposed content reads like a press release got loose on the internet. So each surface carries its own rules about its room: where the payoff goes, how much patience it can assume, what a good open is here versus there. The idea stays fixed; the delivery bends to the audience. A tool handed a post and a format was never told who was on the other side, so it structurally cannot make this move.

The voice is held as a thing, not a vibe. This is the move that actually stops the erosion, and it’s only possible because the system remembers across every piece instead of treating each one as a fresh blank prompt. “Sound like me” is a vibe, and a vibe doesn’t survive contact with ten surfaces a day. So the voice gets made explicit and persistent: the words you’d never use, the sentence rhythm you actually have, the hype adjectives that are an automatic tell, the way you open and the way you land. Held as a thing, the voice becomes something every translation is checked against, not “is this roughly on-brand,” but “is this a sentence this person would publish.” The result is ten pieces that feel like one person had ten thoughts, not one model had ten outputs.

Notice that none of those three moves is a “content feature.” Extraction, room-awareness, and a remembered voice aren’t things we built for marketing. They’re what a memory-bearing, always-on, trustworthy substrate does the moment you point it at a job where one decision has to be applied faithfully many times without drifting. You don’t have ten things to make; you have one idea that has to reach ten rooms still sounding like you, and that’s a sentence about the substrate, not about a feature.

The breadth is a consequence, not a boast

Which is exactly why this isn’t really a story about content at all.

The shape of the repurposing job, take one good decision and apply it a hundred times without fatigue or drift, is not unique to marketing. It’s the renewals desk applying one pricing policy across two hundred accounts. It’s onboarding applying one good welcome across every new user without phoning it in on the fortieth. It’s support applying one true answer consistently instead of re-deriving it, a little worse, each time. The same spine runs all of them, and it runs them for the same reason: humans are worst at applying one standard a hundred times without erosion, and a remembering, untiring system is best at it. Drift is a tax humans pay at 4pm on the tenth caption; the substrate doesn’t get tired on the tenth caption, or the four-hundredth.

So if you ever see this thing carry a dozen jobs that resemble a dozen point tools, that breadth isn’t a checklist we’re proud of. It’s the opposite of impressive, it’s just evidence that we found the right substrate, because once you have it, the jobs fall out for free. We didn’t add ten capabilities. We built one kind of thing, and the capabilities are its shadow.

The turn: what can’t be translated

So hand off the ten translations. But look hard at what you’re keeping when you do, because it’s the whole point.

The system can extract the idea, aim each surface, and hold the voice steady across all of them. What it cannot do is have the thought in the first place, decide that this is the argument worth making this week, find the example that makes it land, write the one line too good to lose. Everything downstream is faithful translation of that origination. Nothing downstream replaces it. The machine is extraordinary at making your idea reach ten rooms still sounding like you, and completely silent on whether the idea was worth having. That silence is the deal, not a gap left to close.

And that’s the world we’re actually building toward, which isn’t on the market yet, because the market is still shipping faster photocopiers. We’re not building a better repurposing tool; we don’t think repurposing should cost you your voice at all. The thing we’re building is an AI-native company OS: one substrate that’s on, that remembers, that’s allowed to act, so that the parts of the work that were never human, applying a decision faithfully across a hundred low-attention surfaces, stop falling on a person at 4pm. It doesn’t fully exist yet. We’re building it out of this exact pain, not cloning anyone’s feature list.

The marketer reading this has a specific dread in their head, the tenth caption, the voice they can feel thinning. The answer was never a tool that makes ten copies faster. It’s a substrate that holds the one thing software can never originate, the thought, the example, the line, completely yours, and stops charging your voice a tax to reach every room. Write the one thing only you could have thought of. Let the other nine reach their rooms still sounding like you anyway, and let that be the first proof of a company where nobody’s best work erodes just to keep the lights on.