More rules should mean better output. That's the intuition. I spent weeks building a comprehensive CLAUDE.md — 200 lines covering naming conventions, security rules, error handling, architectural patterns, import ordering, type safety requirements, and more. I was proud of it. I'd thought through every scenario. Then I scored the output. 79.0 / 100. My carefully crafted documentation was actively
The previous two posts covered how events flow from the SDK to the UI. This post focuses on visualizing one specific type of event: tool calls. Tool invocations are the most frequent operations in an Agent application. A typical task might call tools twenty or thirty times—reading files, writing files, executing commands, searching code. If every tool call renders as the same gray block, it's hard
Post 1 covered how AgentBridge converts the SDK's AsyncStream<SDKMessage> into [AgentEvent]. This post looks at what [AgentEvent] becomes — how TimelineView renders 18 event types, handles scroll behavior, and stays smooth when the event count gets large. TimelineView is the main body of the workspace, filling all the space between the sidebar and the input box. Its view hierarchy is shallow: Time
I'm a software engineer in Japan. I've been using AI coding assistants — Claude Code, Cursor, Copilot — for about one years now. At some point I started keeping informal notes on how many prompt revisions it took to get production-quality output. After a few months, a pattern was hard to ignore. For tasks I described in Japanese: 4–6 revisions on average. 1–3. Same AI. Same model. Roughly similar
"Write a function to fetch the list of users." — same prompt, same codebase. Yesterday: getUsers(). Today: fetchUserList(). Tomorrow: loadAllUsers(). Six months of AI-assisted coding and I kept hitting this wall. My initial reaction was "maybe I need to write better prompts." I wrote better prompts. The functions got slightly better. New inconsistencies appeared elsewhere. The problem wasn't the A