I write these posts with Codex.
The ideas are mine. I usually start with fragments: a technical judgment, a few keywords, or several sentences that do not yet fit together. Codex turns that material into a draft.
It helps with structure and transitions. It also catches gaps: a claim without an example, a jump in the logic, or a section that has already said enough.
This is much faster than writing every sentence alone. I can spend my attention on the part I care about: deciding whether a system, a piece of code, or a workflow makes sense.
There is an obvious cost. Codex likes prose that is smooth, balanced, and complete. My notes are usually none of those things. They are short, technical, and sometimes leave a thought slightly open. A polished draft can be accurate and still sound unlike me.
I have tried several open-source humanizer skills. They catch familiar habits: neat groups of three, empty transitions, and the "not only A, but also B" construction. That helps, but removing AI patterns is not the same as preserving a voice.
So the workflow is still changing. I am teaching Codex a narrower style: lead with the judgment, keep the technical detail, and stop when the point is made.
I do the thinking. Codex helps me make it readable. I am comfortable with that division of work.