FZ*

Writing

Essays on cognitive design and working with AI — from the newsletter Through the Context Window.

Prompting Is Really Pointing Prompting advice says be exhaustive. But past a point, the harder task isn't describing what you want — it's pointing at it through shared references. June 2026 The Work Is Mostly In Our Heads A new hire who forgot everything not written down, but read the entire codebase, the manuals, and a two-year-old lunch order. The easiest onboarding ever — until it wasn’t. May 2026 Making Work Is Good After All You sit before a genius chatbot and type 'summarize this PDF.' Something feels off — and better prompting isn't the fix. May 2026 Meetings Aren't So Bad After All We hate bad meetings, not meetings. As AI handles the doing, your role shifts to the one running the room. February 2026 What Kind Of Problem Even Is This? Some delegation fails even when your instructions were clear and correctly executed. The problem is which mode of delegation you chose. January 2026 Understanding Your Own Feedback You know instantly when AI output is wrong but can't say why. That inarticulate reaction is the most valuable signal in the exchange. January 2026 In The Artist's Own Hand Did you write that, or just approve it? Neither 'AI is a tool' nor 'you're a manager of agents' captures what's really happening to expertise. January 2026 The Document is the Interface When AI collaboration goes wrong, we blame the prompt. Often the real problem is having the conversation in the wrong format. December 2025 Four Summaries And A Decision Most people use AI like a slot machine, rushing to converge on the answer. The valuable move is to deliberately diverge first. December 2025 New Work Demands New Training Professional education produced people who could do the work. The AI era needs people who can get work done — a different capability entirely. December 2025 Prototyping With Words When AI writing disappoints, the fix usually isn't a better prompt — it's switching formats. Each format exposes a different kind of problem. November 2025 Genie in a Model The apocalyptic AI story imagines a single on-off switch to AGI. That framing gets two things badly wrong. November 2025 Claude Code. Minus the Code. Chat tools pass messages; coding tools read and write files. That difference — persistence and naming — is what lets work stick around and compound. November 2025 From Farm To Factory To Office To…What? AI isn't 'either superpowers or replacement.' Like farm-to-factory-to-office before it, it will change both how we work and what work we do. November 2025 When the Price Tag Stops Making Sense We price professional work by the hour because we're paying for effort. When AI makes the effort disappear, the whole mental model breaks. October 2025 Be A Critic: Learn to Evaluate Even Things You Can't Do School taught us to specialize and execute. AI inverts that: value moves upstream to people who can evaluate work they could never produce themselves. October 2025 Stop Trying to Get the Right Answer From AI That frustrating back-and-forth with a chatbot isn't indecision. It's how you discover what you actually meant to ask. October 2025 The Jagged Frontier Is About The Problem, Not The Model Whether AI can handle a task depends less on how smart the model is and more on the kind of problem you give it — convergent or divergent. October 2025 AI creates scale. Make sure you have headroom to grow. Automating the tasks you hate just makes you more efficient at fixed demand. The real payoff comes when you have room to grow into the freed-up capacity. October 2025 Knowledge work is a bundle of activities and AI is unbundling them Most office jobs bundle a 'worker' who produces and a 'manager' who decides. AI is pulling those two halves apart. October 2025 From Thinking in Its Head to Thinking with Its Hands Modern AI is getting better not just at reasoning internally but at doing — running code, using tools, learning through trial. It's thinking by doing. June 2025 A Busy Week in AI OpenAI, Anthropic, and Google each made big news the same week — and each revealed a different bet: consumer, enterprise, and platform. May 2025 What Is a Context Window, Anyways? The real magic of AI isn't inside the model — it's what you put in front of it. Treat the context window like a whiteboard: if it isn't written down, it doesn't exist. May 2025 Building Judgement Funnels In a world flooded with cheap thinking, human judgment becomes the scarce resource — and the task is to architect when, where, and how it lands. May 2025 Decisions, Not Documents AI can now read a library in an hour, but we still can't. As thinking gets cheap, the bottleneck shifts from research to strategic discernment. May 2025 Hybrids, Shepherds, and Centaurs When reasoning gets cheap, work doesn't evolve chaotically — it reclusters into distinct leverage models, each pairing human judgment with machine scale in its own way. April 2025