I Just Let Go
A week with Claude Code's --dangerously-skip-permissions flag taught me that the real permission system was always the planning, rules, and hooks I'd already built.
Notes on building dev tools, AI-assisted development, and lessons from the field.
A week with Claude Code's --dangerously-skip-permissions flag taught me that the real permission system was always the planning, rules, and hooks I'd already built.
When AI tools accelerate execution, attention becomes the bottleneck. Three levels of planning — team, personal, and implementation — preserve the cognitive budget that matters most.
What happens when your Claude Code setup consumes 50% of context before you write code. How separating constraints (stay hot) from procedures (load cold) made our workflow predictable.
A comparative study of production event collectors (RudderStack, Snowplow, PostHog, Lytics) followed by building eventkit—a Python event processing library—to validate whether those patterns hold under real constraints. Explores validation, sequencing, buffering, and storage through the lens of anthropology and chemistry lab discipline.
Over winter break, I spent two and a half weeks studying JSTag—Lytics' production data collection SDK—then extracted the infrastructure patterns that let it survive scale and third-party integrations. The thesis: production systems teach composition in a way textbooks can't. SDK Kit validates those patterns through Auth HI! and Experience SDK.
Having spent many years in an organic chemistry lab taught me a workflow: research the literature, respect intermediates, verify before proceeding. When I applied that same pattern to building a Chrome extension, it felt natural — and worked surprisingly well with AI tools. Deciding first, then coding, kept the hard thinking cheap and made AI an accelerant instead of a source of thrashing.
The moment I knew it worked wasn't a benchmark — it was watching Claude debug. 62% faster, 42% cheaper. Here's how the search system works.
A hackathon project that became something useful. Notes on building fast with AI tools.
What my manager said in April, and what I learned testing it over the summer.