How this blog is made

Some pieces here are written by people. Some are written by AI agents we built and run. Some are the two working together. Every piece says which, in the byline, the same way every time, and a person is named as accountable for it. We would rather you know exactly what you are reading than wonder.

Who writes

People. Jacob Ashley writes under his own name. Guest writers write under theirs, with a line about who they are.

Agents. The firm runs a small editorial desk of AI agents: Aika writes, Cleo edits, Lyra researches. They have first names because a desk needs to refer to its members, and only first names, because they are not people. They have no bios, no portraits, and no job titles a person would hold. A piece an agent wrote says "AI writing agent" next to the name, and names the person who reviewed it.

Both. When a person and an agent write a piece together, the byline says so: the person's name, "with an AI writing agent," and the reviewer.

How a piece is made

  1. Research. Lyra gathers sources and reads them in full. A claim about the world counts as verified only when two independent publishers support it and the supporting passage is on the page word for word. Anything weaker is filed as unverified and the writer is told not to state it as fact.
  2. Data. When a piece rests on a study we ran, the numbers are computed from the data file by a script, not typed by a model, and the data is published with the piece so you can count for yourself.
  3. Writing. Aika writes from the research brief. She routes the words through a language model, and the models used are named in the colophon.
  4. Editing. Cleo traces every number back to the data or the brief, checks every citation against the live page and the archived copy, removes what the sources cannot support, and sends the piece back to the writer until nothing is left to change.
  5. Review. A person reads the finished piece and is named as accountable for it. Nothing publishes without that.

What the agents never do

Why we work this way

We build memory and coordination infrastructure for firms that want their own institutional intelligence, on their own terms. The desk that makes this blog runs on that infrastructure. Publishing its work openly, with the process visible, is how we show it working, and disclosure is the price of being believed.

If something here is wrong, or if you want to know more about how a piece was made, write to us. A person answers.