A personal field log about building useful systems, making things with care, and noticing what changes the work.https://johnsondeng.pages.dev/https://johnsondeng.pages.dev/zettels/smaller-loop/https://johnsondeng.pages.dev/zettels/smaller-loop/When a process feels stuck, shrink the distance between action and feedback. A small loop produces evidence while a large plan is still defending its assumptions.Thu, 30 Jul 2026 00:00:00 GMTWhen a process feels stuck, shrink the distance between action and feedback. A small loop produces evidence while a large plan is still defending its assumptions.systemspracticezettelhttps://johnsondeng.pages.dev/posts/small-systems/https://johnsondeng.pages.dev/posts/small-systems/The useful operating systems are usually the ones nobody needs a meeting to explain.Tue, 28 Jul 2026 00:00:00 GMTThe systems that matter most rarely announce themselves as systems. They are the short checklist beside the door, the naming convention that makes a folder obvious, and the weekly note that keeps a decision from becoming folklore. I used to reach for bigger machinery first. A new dashboard felt decisive. A new workflow promised order. Then the work would change slightly and the machinery would become another thing to maintain. ## Build for the ordinary Tuesday A durable system should make the common path easier without making the uncommon path impossible. That usually means fewer required fields, one clear owner, and a visible place for exceptions. > The best process leaves enough room for judgment while making neglect difficult. I now look for a smaller loop: notice, decide, record, revisit. If that loop works, software can make it faster. If it does not, software only makes the confusion arrive sooner. The goal is not perfect order. It is enough shared memory that the studio can keep moving without asking the same question twice.Systemsposthttps://johnsondeng.pages.dev/zettels/clarity-speed/https://johnsondeng.pages.dev/zettels/clarity-speed/Fast work is not always quick work. A precise name, a visible constraint, or one honest sentence can remove days of motion that was never progress.Wed, 22 Jul 2026 00:00:00 GMTFast work is not always quick work. A precise name, a visible constraint, or one honest sentence can remove days of motion that was never progress.writingteamszettelhttps://johnsondeng.pages.dev/posts/taipei-morning/https://johnsondeng.pages.dev/posts/taipei-morning/A field note on shutters, scooters, breakfast counters, and the city assembling itself.Sun, 19 Jul 2026 00:00:00 GMTAt 7:12 the city is already halfway through its first shift. Metal shutters lift one uneven meter at a time. A breakfast shop stacks warm paper bags beside a row of soy milk. Scooters fill the spaces that looked impossibly narrow the night before. The useful thing about walking early is that the city has not finished hiding its work. Deliveries happen in public. Buckets, extension cords, folded tables, and handwritten signs reveal how much improvisation sits underneath an ordinary day. ## A city made of adjustments Taipei often feels less designed than continuously tuned. A planter blocks the rain from one doorway. A mirror solves a blind corner. A strip of tape tells a regular customer exactly where to wait. None of these changes are grand enough for a case study. Together they make the place legible. The lesson I keep taking back to software is simple: people do not need every surface to be new. They need it to notice where life is already happening.Field Notesposthttps://johnsondeng.pages.dev/zettels/keep-door-open/https://johnsondeng.pages.dev/zettels/keep-door-open/An early idea needs enough structure to be testable and enough looseness to surprise you. Close every decision too soon and the work can only confirm its first draft.Mon, 13 Jul 2026 00:00:00 GMTAn early idea needs enough structure to be testable and enough looseness to surprise you. Close every decision too soon and the work can only confirm its first draft.makingtastezettelhttps://johnsondeng.pages.dev/posts/useful-definition-of-taste/https://johnsondeng.pages.dev/posts/useful-definition-of-taste/Taste is less about preference than the ability to notice which choice changes the whole.Wed, 08 Jul 2026 00:00:00 GMTTaste is often treated like a private weather system: some people have it, some do not, and nobody can explain why. That makes it sound mysterious enough to avoid practicing. I prefer a more useful definition. Taste is the ability to notice which differences matter, then to make a choice that clarifies the whole. ## Attention before opinion The practice begins before preference. Look long enough to describe what is happening without reaching for praise or dislike. Which edge carries tension? Where does the rhythm break? What is the one element doing too much work? Good criticism makes the next attempt more possible. It names a relationship, not merely a reaction. Over time those relationships become a library: proportions, pauses, contrasts, and tiny violations that give a thing its character. Taste grows when attention becomes specific enough to act on.Cultureposthttps://johnsondeng.pages.dev/posts/room-for-judgment/https://johnsondeng.pages.dev/posts/room-for-judgment/Automation is most useful when it carries repetition and keeps consequential choices visible.Wed, 24 Jun 2026 00:00:00 GMTThe tempting version of automation ends with a button labeled "Do everything." The useful version is usually more modest. It carries the repeated parts, shows its work, and stops where context begins to matter. I think of a good tool as a workbench rather than a vending machine. The state is visible. Materials stay within reach. A person can interrupt the sequence without fighting the interface. ## Keep the consequential choice legible When a tool makes a recommendation, it should also preserve the evidence that made the recommendation sensible. When it changes a file, the diff should be easy to inspect. When confidence is low, the uncertainty should shape the interaction. ```ts const nextStep = confidence > 0.85 ? 'apply' : 'review'; ``` This is slower than pretending every decision can disappear. It is much faster than repairing trust after the tool quietly makes the wrong call.Makingposthttps://johnsondeng.pages.dev/posts/notes-before-ready/https://johnsondeng.pages.dev/posts/notes-before-ready/Fragments are not failed essays. They are the raw format some ideas need at the beginning.Thu, 11 Jun 2026 00:00:00 GMTMost ideas arrive without an introduction. A sentence appears while walking. A comparison feels promising but cannot yet support a claim. A question keeps returning without becoming urgent. I keep these fragments in a single dated stream. No folders at first, no pressure to title them well. Once a week I mark the ones that still have heat and link the ones that seem to be speaking to each other. The method works because it separates capture from performance. A fragment only needs to be recognizable to my future self. An essay can ask for coherence later. Small notes make writing less like starting and more like continuing.Systemsposthttps://johnsondeng.pages.dev/posts/personal-site-for/https://johnsondeng.pages.dev/posts/personal-site-for/A personal site can be a home for unfinished thinking, not another feed to keep full.Sat, 30 May 2026 00:00:00 GMTA personal site is useful because it does not need to behave like a platform. It can be uneven. A long essay can sit beside three sentences. A project can remain visible after the launch conversation ends. The archive matters, but so does the act of arranging it. Categories become a record of attention. Links reveal a set of neighbors. The gaps are allowed to say that life was happening somewhere else. I want this site to be less of a publication schedule and more of a worktable: current notes near the front, older tools within reach, and enough empty space for the next thing to arrive.Culturepost