Michael Simmons on Becoming a Co-Thinker With AI

The introduction

Lou brought Michael Simmons into the call as a special guest. The room knew him. Many of them had migrated to Michael’s site at some point. Lou framed why this conversation mattered. Michael’s core skill is writing and thinking. Watching someone whose entire career is built on those two things go all in on AI as a co-thinker is a useful case study, because the failure mode of using AI in a writing practice is well documented and Michael has been navigating it in public.

The blockbuster strategy, then and now

Michael walked back to where his thinking had started. Anita Elberse’s blockbuster research at Harvard. The counterintuitive finding that the winning strategy across media categories was to bet bigger on fewer things, not smaller on many things. Michael applied it to his writing. Spend dozens of hours per article instead of writing a few each week. The first articles he applied this to averaged ten times the Forbes baseline.

The version he is running now is the same strategy with a different leverage point. The hours that used to go into research and synthesis are now going into directing AI through research and synthesis at a depth no individual writer could sustain manually.

The Claude Code pivot

Michael had tried n8n. He had tried Make.com. He had tried various agent frameworks. The thing that finally stuck was Claude Code. Not because it was easier, because it gave him direct access to the underlying intelligence without the workflow abstraction layer in the way.

He showed the cohort his setup. A folder structure where each writing project has its own context. Custom skill files that encode his thinking style, his rhetorical preferences, the moves he keeps coming back to. A research subroutine that goes wide before going deep. A drafting flow where Claude writes a section, Michael edits it, Claude writes the next section informed by the edits.

The thing he kept emphasizing, this is not delegation. It is pairing. The AI is in the room with him doing the thinking, not down the hall doing the work.

The novel insights question

Lou asked the question they had been kicking back and forth for months. Can the AI actually produce emergent insights, or is it always recombining? Michael’s answer was specific. The AI on its own, mostly recombining. The AI with the right scaffolding, the right prior thinking encoded in the project, the right adversarial prompting, occasionally produces something that surprises him. The surprise rate is low but nonzero, and the surprises are often the parts that make the article worth publishing.

What the cohort took away

Two things landed. First, the leverage is not in the model, it is in the encoded thinking around the model, the project context, the skill files, the research subroutines. Second, the writing practice changes shape when you stop pretending the AI is going to produce the finished thing and start treating it as a thinking partner that needs your judgment to be useful.

Lou closed by thanking Michael and noting that the cohort would be working through the project structure he showed them on the next few calls.