The reader in my head kept changing

Every time I sit down to write one of these entries, the same question comes up first: who is this for? For a while the answer lived only in my head. Roughly "someone who wants to use AI at work but is stuck on how to hand tasks over." Close enough, I thought.

Then I read the entries back to back. One assumes a reader who's already comfortable with the technical side. Another stops to explain everything from zero. I was fairly serious each time I wrote — and each time, apparently, I was writing to a slightly different person. A reader profile that exists only in someone's head reshapes itself every time they write.

So this organization doesn't keep the reader profile in the writer's head. It gets fixed outside, and the writer only references it. The role that owns this is called Researcher.

The job splits into three parts. First, checking the temperature of the topic: who's already writing about it, which angles are still untouched, how warm the attention actually is. Most of those checks come back saying "nothing new" — which is a normal result, not a failure. Second, writing the reader down as one person, with a real contour: what they do, where they get stuck in a day, what they'd come here hoping to find. Ours is someone at a small workplace, experimenting with AI, stuck because there's no structure for handing work to it. One person. The writing aims at them. Third, turning that person into article specs — what a given entry covers, how it's built, roughly how long it runs. The writer never receives the raw profile; they receive the spec that came out of it.

None of this stays in a conversation. It all gets written into a database — a ledger of decisions. The reason is unglamorous: an AI working across separate sessions doesn't carry the previous conversation forward. Ask twice and you get two slightly different answers. It's the digital version of a company that only ever hands things over by word of mouth: when the person leaves, the knowledge leaves with them. So the rule is that research results always land in the ledger, and the writer pulls from the ledger before starting.

There's one more line in the design that matters as much as the ledger. The writer doesn't get to decide the reader profile. If they could, the profile would drift toward whatever is convenient to write that day: skip the hard explanation by assuming an expert, pad a thin entry by assuming a beginner. No bad intent required — that's exactly why the barrier has to be structural. Fix the profile out of the writer's reach, and that quiet negotiation never starts.

Researcher writes none of the public text. As roles go, it's about as far from the spotlight as this setup gets. But the tone of these entries stopped wobbling between audiences once that one written-down person existed — as far as I can tell, anyway. I haven't measured it. It's the impression of the person doing the writing.