Writing
Notes on governed generation.
Essays on what can be specified and what stays human when AI produces content — written while building the systems they describe, and including the results that did not support the design.
Start here
- VocabularyConceptsEvery term I use, in a sentence or two — with where to read more.
- Series · 6 partsThe admissible spaceShort pieces on how an explicit boundary makes creative AI work possible. Start at part 1.
- Series · 4 partsApplied cognitive architectureLonger essays on knowledge, websites, creativity and agents.
All writing, by date
2026
- And this is what the measurement saysI tested whether AI dials actually steer output. First result: no effect at all. What fixed it, what still doesn't move, and why the gap never closes.
- Who holds whatMachines hold the stated part of a brand, people hold what resists statement. Why human judgement isn't a temporary gap, and the loop that shrinks it.
- The edges are softThe best work often sits just outside the line. Why writing a brand boundary down is what makes deliberately crossing it legible as craft, not an accident.
- Same space, different regionInforming and inspiring aren't two brands. They're the same axes with different ranges. Why a brand boundary should be coordinates per purpose, not a page.
- Every picture of a boundary liesBrand guidelines are a flat picture of a many-dimensional boundary. AI doesn't create the gap; it's the first reader honest enough to show it to you.
- The boundary makes it possibleA pilot flies freely inside a boundary they can't cross. An explicit brand boundary doesn't make AI output dull; it makes delegating to AI reasonable.
- Your Agent Is an Actor Without OrgansEveryone is building agents. Without an instrument, substrate, contract and conscience, an agent is an actor without organs. A four-part admission test.
- The Edges of Creativity Are the Part That Becomes ComputableCan creativity be computed? Half of it. Separating a design's identity from its manifestation makes creativity's edges computable and safe to delegate.
- The Corporate Website Was Built for a Different CenturyAI answer engines don't browse, they parse. Why the navigation-first corporate website is outdated, and what replaces it: content units plus smart routing.
- Every Organization Has a Knowledge Fingerprint. Most Never See Theirs.Your content audit feels random because it lacks a structure. Five layers, from differentiator to proof, reveal your knowledge fingerprint and its gaps.
- You can't scale AI on unstructured contentAI doesn't read documents, it assembles fragments. Why scaling AI needs explicit, governed building blocks of meaning instead of pages, slides and folders.
- SEO optimized for matching words. AEO optimizes for matching meaning.SEO matched words; AEO matches meaning. Why clear language, synonyms and accessible structure make content easier to find for AI search and for people.
2025
- On brand, on intent, on risk: why AI content needs governance, not just promptsBetter prompts won't keep AI content on brand. Intent, brand and risk need explicit rules the system can check — and unlike prompts, those rules compound.
- The goal is not to be read — it is to be understoodContent is understood only when it is true, well expressed and right for this reader. Miss one and you get noise, silence or manipulation.
- G-core: Governing Meaning Before GenerationG-core decides how something may be said before the model looks up what to say. Why that order makes generative AI governable, auditable and safe to scale.