Coding with AI agents has quietly flipped my read/write ratio. A firehose of generated output gets written into me across three terminals, and I almost never sit with a hard problem long enough to read any of it back. A power user's look at what that does to the brain — with the research that says it's not just nostalgia.
A bookmarklet for United's flight search led to a broader idea: users no longer have to accept default web UX as fixed. AI makes local, runtime personalization much more practical.
Claude Code silently truncates lines longer than 2,000 characters — no error, no warning. Minified HTML, bookmarklets, and SVGs all hit this. I watched it loop for 20 minutes on a five-file project it could never fully read. Diagnostic, repro scripts, and workarounds included.
I tried to test Claude Code's multi-agent team mode with split panes in iTerm2. The it2 CLI was broken, the docs were confusing, and I spent the whole session yak-shaving. Here's what actually works.
AI coding assistants amplify both velocity and bottlenecks. Teams seeing 10x gains already had fast feedback loops. Here's why your infrastructure is now the bottleneck.
I had 10 years of workout data and wanted AI coaching for my strength training. Instead of building AI into my app, I built something simpler: a way to let any LLM generate workout configs. Zero AI code, zero costs, perfect privacy.
Deep dive into why Google's Gemini CLI takes 35-60 seconds to start. Analyzing 129,681 syscalls, MCP server initialization, and fundamental architectural issues affecting developer productivity.