Very helpful article. Actually, I think most of us did realize by now planning docs are necessary, I have used them extensively with Cursor even before I added Claude Code to the mix. Another good idea is to divide the project into "zones" and concentrate our reviews and full attention only on the core where AI's mistakes might really matter.
For anything with a web UI I've been handling the E2E testing by telling CC to make (and then execute) a plan for Selenium to do it. :)
Very helpful article. Actually, I think most of us did realize by now planning docs are necessary, I have used them extensively with Cursor even before I added Claude Code to the mix. Another good idea is to divide the project into "zones" and concentrate our reviews and full attention only on the core where AI's mistakes might really matter.
The ROADMAP.md approach for context is smart. I use CLAUDE.md files for similar effect. Structure enables autonomy. More: https://thoughts.jock.pl/p/ai-agent-self-extending-self-fixing-wiz-rebuild-technical-deep-dive-2026
Thanks so much for sharing. It's great to hear what others are figuring out with these new tools.