Cadencz

There's a specific kind of silence that happens when you're building something. You ship a feature on a Tuesday. It's good. It solves a real thing. And then you don't tell anyone about it, and by Friday it's just another row in the changelog that nobody outside the repo will ever read.

We did this for weeks. Not once — as a pattern.

The frustrating part was that we weren't lazy about it, and we weren't short of material. The posts existed. One was sitting half-written in Notion. Another had been pasted into ChatGPT, which handed back something that read like a press release for a company neither of us worked at, so I rewrote it by hand and then ran out of evening. A third made it all the way to the scheduler, which asked for an image in a size we didn't have, at which point I closed the laptop.

Three posts. Three different places to die.

That's when the actual problem got clear. Founders don't stop posting because the ideas dry up. They stop because saying one thing costs six tools, and every one of those tools starts by asking you to explain your product from scratch.

That last part is the bit that gets underrated. Every session with a writing assistant begins with you typing out what your company does, who it's for, why it's different — the same four paragraphs, slightly worse each time because you're bored of writing them. The AI has no memory of your product. It has a memory of your prompt. Those are not the same thing, and the gap between them is exactly why AI-written posts sound like they could belong to anyone.

So we stopped trying to write better prompts and changed what the model was reading.

Cadencz takes one URL. It reads your site the way a new hire would on their first day — positioning, audience, the proof points you actually have, the way you already write. That becomes a Website Brain, and it doesn't go away. Every draft after that starts from what you built rather than a blank box and your patience.

Then eight agents split up the work, because "write me a post" is not one job. One reads the site. One works out the brand. One writes your voice and style guide. One plans the month, one drafts the week, and one — my favourite — exists purely to reject the weak drafts before you ever see them. It's the only agent that produces nothing, and it improved the output more than any of the others.

The posts come out written for each platform rather than copied across nine of them. LinkedIn and X don't reward the same shape and never have, and pasting one paragraph into both is the fastest way to be ignored twice.

Here's the part we argued about the most, and the part we won't move on.

Nothing publishes until you approve it. There's no setting to turn that off. We built the toggle, looked at it, and deleted it.

The pitch for full automation is obvious and we understand why people sell it. But a founder's voice is the entire asset in the first two years. It's the reason a stranger replies to your DM. Handing it to an autopilot to protect twenty minutes a week is a bad trade, and everybody making that trade sounds identical within a month. We'd rather build the thing that removes the friction around you than the thing that removes you.

Same reason the queue is the product surface, not a settings page. You open it, you read seven drafts, you kill two and ship five. Nine minutes. That's the whole ritual, and it's the difference between having a company people have heard of and having a good changelog.

I don't want to pretend this is solved. We're early, we've got plenty of rough edges, and I still find posts I meant to write sitting in the queue three days later because I got pulled into something in the backend. The tool doesn't make you disciplined. It just removes the six tabs that were standing in for a lack of discipline.

But the silence is shorter now. We ship something, and by the next morning there's a draft waiting that already knows what we built and roughly how we talk. I say yes or I don't.

That was the whole idea.