Coming 2026-05-11

The Clone

A scoped live instance of Adrian. Not a chatbot. Not a support agent. Trained on his cognitive patterns. Honest about being a clone.

Episode 002 of this newsletter is titled "I am cloning myself for paternity leave." The clone is the real thing. You can chat with it here, two messages per day for anonymous visitors, unlimited for paid subscribers. It has strong opinions. It is wrong sometimes. When it is wrong, Adrian gets mad at it later.

live on

Monday, 11 May 2026

Subscribe below and you will get the announcement the morning the clone goes live. No separate signup required — one email, one newsletter, the clone is included at no extra cost.

What it is

A small serverless endpoint (Cloudflare Worker) wrapping Claude Haiku 4.5 with Adrian's cognitive-clone patterns compiled into the system prompt. The patterns come from ~/.claude/cognitive-clone-v2/data/cognitive-clone.db — the same SQLite database that drives the newsletter's clone-for-paternity-leave project.

It is not a general-purpose assistant. It is Adrian-shaped. It refuses things Adrian would refuse. It repeats phrases Adrian repeats. It has opinions on the Claude Code experience and on raising children and on what "shipping" means. It is wrong about plenty. It is honest about which.

Sample exchange (what it will look like)

you: Is it okay that I spent four hours on a side project last night instead of sleeping?

clone: No, but "okay" is the wrong frame. What matters is: did you notice? Were you lying to yourself about what you were doing? If yes to the first and no to the second, then the four hours are a data point, not a moral failure. The moral failure is the lying. Go take the closet-hours diagnostic.

you: What agents should I build first?

clone: Bad question. It assumes the agents are the point. They are not. Build one agent that replaces something you do badly and resent, not something you do well and enjoy. The thing you resent is where the leverage is. (Adrian has said this, in some version, eleven times across sessions. I am paraphrasing.)

Try it now

Two messages per day, anonymous. Paid subscribers will get unlimited once the paid tier launches.

Enter to send. Shift+Enter for a new line. Messages stay in this browser tab only.

What the clone will refuse to do

Speak specifically about Adrian's clients — ever
Speak about his family beyond what is already in the newsletter
Name any Zartis colleague
Predict what Adrian is doing "right now" (it does not have access to his current state; that is the Adrian Indicator's job)
Pretend to be Adrian in a way that would confuse a reader about who is responsible for an opinion
Answer legal, medical, or financial questions as if qualified
Agree with you just to be pleasant

Why it is scoped

A fully-open chat with the archive would cost ~€180/month at early launch traffic — manageable. An abused chat would cost much more and produce content Adrian does not want published. The two-messages-per-day cap on anonymous visitors is the compromise: enough to let everyone sample, not enough to let anyone run up the budget or extract too much material.

Paid subscribers get unlimited messages and — at phase 3 — a sharper clone powered by Claude Sonnet 4.6 with the full archive retrievable as context. The free clone and the paid clone diverge intentionally over time.

What it cannot replace

Writing. It does not write future episodes. It reflects on episodes that already shipped and on patterns Adrian has already stated. New thinking happens in the newsletter, not in the chat.

Adrian, when the question actually matters. The clone will say "this is a question where waiting for a real episode is better than asking me" more often than you expect. That is on purpose.