I write about engineering in the age of AI agents, so it would be strange to be cagey about using them. Here is the actual division of labor.
The Post-Human Briefing is fully machine-generated. Agents running on Gemini crawl the sources, synthesize the daily edition, and narrate the audio. No human writes those sentences. I audit it, I fix the pipeline when it drifts, and I decide what it covers.
Field Tests are experiments. The models help me design the rig and write the code. The numbers come from actual runs against real APIs, checked by a program rather than by a model’s opinion, and I publish the raw results including the failures. The claim I am testing gets written down before the data exists, and when I turn out to be wrong the original prediction stays visible with a note underneath.
Deep Research is done by Gemini Pro models.
Essays are drafted with AI and edited by me until they sound like me. If a paragraph survives, it survived because I agreed with it.
Curation, vision, and structure are not shared work. Which claim is worth a week of my life, which one is noise dressed as news, what this publication is trying to be over five years, where an essay turns and what it refuses to say: that is mine, and it is most of the job. A model can write a competent paragraph about anything. It cannot tell you which paragraph was worth writing. Everything the models produce here lands inside a shape I set first.
What I refuse to delegate. Deciding what is true. I read primary sources rather than summaries.
Anecdotes are never invented. If a story appears here in first person, it happened to me. A model may have tightened the sentence. It did not supply the memory.
Errors are mine. Not the model’s, not the pipeline’s. Corrections are published in place, and the archive shows its own edits.
The whole publication is an argument that machines produce, instruments measure, and humans judge. It would be dishonest to make it any other way.