I’ve been working this way for the last few months too. I challenged myself to develop side projects in Claude Code without writing any code by hand. Ended up have some of the same epiphanies you did 💪🏼
The reasoning trace habit is underrated. If the logic makes sense, the code usually does too. I found the same shift applying to code review. Instead of reviewing PRs myself, I built a system where specialist agents review different domains in parallel: frontend catches the accessibility issues, DevOps checks the k8s configs, and I just look at the summary. Covered the OpenCode setup here: https://reading.sh/one-reviewer-three-lenses-building-a-multi-agent-code-review-system-with-opencode-21ceb28dde10
I’ve been working this way for the last few months too. I challenged myself to develop side projects in Claude Code without writing any code by hand. Ended up have some of the same epiphanies you did 💪🏼
The reasoning trace habit is underrated. If the logic makes sense, the code usually does too. I found the same shift applying to code review. Instead of reviewing PRs myself, I built a system where specialist agents review different domains in parallel: frontend catches the accessibility issues, DevOps checks the k8s configs, and I just look at the summary. Covered the OpenCode setup here: https://reading.sh/one-reviewer-three-lenses-building-a-multi-agent-code-review-system-with-opencode-21ceb28dde10
I went through the same journey. I have shipped more features in last 2 months of coding than last 3-5 years combined.