Writing

The same engine runs in public.

The writing is not marketing. It is the reasoning with the work removed: read the signal in public data other people were already looking at, name the assumption holding the consensus together, and say what would have to be true for it to hold.

What it covers

Four recurring threads.

  • CPG, pricing, and market signals

    Where the tape stops agreeing with the forecast, and what a commercial team should do about it before consensus moves. An oil thesis that called $100 Brent ahead of consensus. A beef piece that named producer-level incentive failure before the trade press.

  • Commercial and operating systems

    How pricing, planning, and margin decisions actually get made, and why the loop matters more than the model. Usually written from the inside of a mess rather than about one.

  • AI, product, and workflow thinking

    What separates an AI product from an expensive demo. Evaluation, human judgment, deployment economics, and the workflow assumptions automation quietly locks in.

  • Experiments and what changed the model

    Pieces that started as a test and ended with a revised view. Including analogies borrowed from outside the category, which is usually where the useful method comes from.

Where to read it

Everything published lives on Substack. Archive, subscribe, and recent pieces are all there.