"You don't weep at a stranger's thought. You weep at a reunion."
Writing with AI, as Kenneth practices it, is neither dictation nor autocomplete. It is a division of labor with a hard rule at the center: the human writes the foundation, and the AI extends it. The rule came from code first, a hand-written codebase is a style guide written in the most unambiguous language possible, and AI can read that embedded philosophy and extend it faithfully, where an AI-written first version leaves nothing to match and drifts toward the mean of everything ever written. Then the realization that the same thing had already happened in prose, accidentally: eighteen years of essays were V1. An AI with no corpus regresses to the mean of the internet; an AI steeped in one writer's corpus regresses to the mean of that writer, which is more or less the definition of voice.
What the AI contributes is not words so much as room to think. It amplifies thinking itself, cognitive scaffolding that lets you hold ideas too complex for biological working memory, explore tangents without losing the thread, and iterate without the fatigue that usually kills the third revision. The process is recursive: another intelligence reflecting your thinking back at you, and the reflection must be policed. When a draft said "force me to see," Kenneth caught it, I didn't say that, because the sentence sounded smarter than him and felt less true. Every suggestion is a choice about identity as much as content.
The deepest reason for the practice is access. The bridge between knowing and saying is where a lot of minds get stuck, and a tool that can read fragments and return prose can open it. The dam was never empty: it was unarticulated. The discourse has categories for "slop" and "cheating"; this is a third thing, an accessibility device, and the answers to the ethical objections are disclosure, a supplied corpus, and a human who refuses to abdicate. The same division of labor scales past the essay: the human provides the what and the why, the AI provides the how. And it can run continuously through a vault, a second brain that stopped being write-then-search and became write-then-think.
In Practice
Arrive with a spark, a sentence, three fragments, a correction. Let the AI read you first: the corpus, the vault, the previous entries in the series. Let it draft. Then do the actual authorship: catch the facts only you can catch, redirect the structure, supply the material that happened in your kitchen and not on the internet, and enforce the rules that keep it honest, no invented anecdotes, disclosure always, and when a piece is wrong at the root, delete it whole. The AI holds the architecture, the continuity, the syntax, and the stamina. You hold the spark, the facts, the taste, and the kill switch.
Related
- Write It First, Then Let AI Drive, the foundation principle, proven in code
- How I Write Now, the full mechanics, the access argument, the ethics
- Idea Amplification and Writing with AI: AI as amplifier of thinking itself
- The Art of Writing with AI, the recursive mirror, and catching it drift
- The Becoming, the division of labor at book scale
- Obsidian Vaults & Claude Code, the vault as thinking partner
- Collaboration Patterns, where this sits among the modes of partnership
- The Recursive Loop, the loop this practice runs inside
- Cognitive Convergence, the homogenization risk the corpus answers
- Cadence: what the corrections are listening for
- Writing, the writing hub