Writing

Essays

Four essays are open; the rest are shared on request: email me for the password.

• 15The door where the designer gets inSep 2026 The six doors an agent can take into your product, from headless to bound to the UI. Only WebMCP lets a designer own the encounter, and building one on live Cortellis data taught three lessons about guardrails, confidence and refusal.
• 14Stop selling UXSep 2026 Advocacy that never ends has failed. The three-rung ladder (teach, institutionalize, demand pull), the calendar-share metric nobody tracks, and the comfortable failure of the perpetual evangelist.
• 13The assistant is scaffoldingSep 2026 The chat assistant bolted onto a product is the m-dot site of the AI era: legitimate as intent telemetry and organizational training, a failure when declared the destination.
• 12The org chart will have agents on itAug 2026 Lessons from a workspace where humans and AI agents are teammates: membership has terms, trust has chemistry you can design, and accountability stays human.
• 11AI lowers the bar; the ceiling stays oursAug 2026 Taste rescued from mystique: fast defect detection against an agreed standard, held by mechanisms rather than heroes. And the apprenticeship problem: we automated the reps that used to train it.
• 10Leverage before headcountJul 2026 Demand doubled; the team did not. The sequencing argument (leverage makes each hire worth more), why headcount-shaped capacity is fragile, and the incentive system that pays for the wrong thing.
• 09An MCP wrapper is table stakes, not strategyJul 2026 From the builder of 50+ MCP servers: wrapping your API makes you reachable by agents, not chosen by them. Conway's law over JSON, what agents actually select on, and where the moat moved.
• 08Knowledge that compoundsJun 2026 AI sessions are amnesiac by default. What changes when every analysis compiles into a wiki the next session starts from, and why memory turns out to be an operations problem.
• 07Synthetic users you can trustJun 2026 Every research team is about to be offered the same deal: an AI persona that answers instantly and costs nothing per session. Most will take the deal without reading the terms. On grounding, provenance, and designing the ignorance.
• 06Your next user is an agentMay 2026 The command line turns out to be the best interface for agent users. Why that is, what it has to do with MCP, and why designers should claim this surface before it ships undesigned.
• 05When the method becomes codeApr 2026 The qualitative research pipeline, packaged as executable skills. What that changes about rigor, what it changes about the researcher's job, and what refuses to be automated.
• 04Design for the user who has to defend the answerMar 2026 In regulated work the deliverable is a defended answer. The defensibility stack (says who, said when, how sure, what it refuses), what it does to the interface, and why regulated users were early, not edge cases.
• 03Design systems are political projectsMar 2026 Design systems do not die of bad tokens; they die of politics. Redistributed authority, legitimacy over artifacts, governing as a service, and why we staff a diplomacy job with component engineers.
• 02Instrument your org like a productFeb 2026 Design leaders demand analytics from products and run their orgs on vibes. The telemetry that changed real decisions, and the strict line that keeps instrumentation from becoming surveillance.
• 01Reorganizations are infrastructure windowsDec 2025 Steady state starves infrastructure and reorgs briefly suspend the immune system. The discipline: keep a plan on the shelf, claim infrastructure not empire, and prove it before the window closes.

Coming: Strategy documents should compile · Where taste will come from