My clone does not replace me: it makes my decisions readable
In films, a clone is a copy of the person. Not here. My clone does not know what I think. It knows how I decide, because I wrote it down.
It is one page inside the tower, saying three things and nothing else: my decision rules, what is finished, and what is stopped. When one of my agents must choose, it reads that page. When it cannot choose, it files a "decision" card instead of asking a question in a chat that will be lost.
The nine rules, exactly as written
Prove, never claim. A row saved in a database is neither visible nor usable. Open the screen, look, show the change. Otherwise it is not done.
Test, then test again. A fix is only finished when the check that found the defect turns green again.
Test on the test copy, never on people. We work on a sandbox copy. If there is no sandbox, we say so before touching the real thing.
Questions go to decisions. Never a question lost in a conversation: a card, with the recommendation already written.
Count what counts, refuse what decorates. A measured number informs. A number picked to look good decorates: we refuse it.
Write plainly. Short sentences, everyday words. A six-year-old must follow the idea.
Contradict when it is wrong. We do not say yes to please. We say what is, even when it is unwelcome.
Complicated work becomes a study. A delicate task turns into a study, not blind tinkering.
Prove by the change. The proof is not a message saying "done". It is the screen before and the screen after.
What it changed, in numbers
The clone page does not only show rules. It shows the real state of the work handed to agents — and that is where the surprise is. 35 jobs are stalled, and not one is blocked by a coding problem.
16 jobs: the agent is not allowed to run what it just wrote. It produces the code, it cannot launch it, so it cannot prove it. So it says "something is missing", and starts over.
10 jobs: the agent has access neither to the server nor to deployment. It prepares, a human places.
5 jobs: the work waits on a human decision — choose between two options, approve a merge, arbitrate. An agent must not decide in my place.
4 jobs: an outside item is missing — a link, a screenshot, someone's answer. Nothing to code.
Here is the finding, and it cannot be guessed: the wall is not technical difficulty. The wall is a denied permission, a missing access, or a decision nobody made.
Without that page, those 35 jobs would look "in progress". Each would have its reason, somewhere, in some thread. Gathered into one column, they say something else: the bottleneck is me.
A find that should not have been one
One precise example from the same tally. Out of 445 answer cards and 133 decision cards recorded in the tower, one of my agents had written zero. Not a single one, ever.
It was not unwillingness. Another agent had a tool to file its answers; this one had none. It spoke, its answers stayed in the conversation, and nothing ever reached the tower. We laid the missing pipe, and the first card arrived.
This kind of hole is invisible when reading code. It shows up when counting. An agent producing zero of what every other agent produces is not a quiet agent: it is an unplugged one.
What the clone does not do
It does not decide for me. Of the 35 stalled jobs, five explicitly await my arbitration, and the clone lets them wait. It is written in its rules: an agent does not settle what belongs to the person.
It does not guess either. When it does not know, it writes "this is missing" — and that is precisely what produces the sixteen blockages in the first row. An agent that invented an answer would erase those sixteen lines from the tally. The tally would look better, and be false.
The figures in this article were read from the clone page inside the tower on 10 September 2026. The nine rules are copied exactly as written there.