I handed my React codebase to several AI coding agents for a full refactoring cycle — here is what they got right, what they broke, and why I still prefer human review for architectural decisions.
I handed my AI agent control over our team’s approval workflow for five days. Expense reports, leave requests, purchase orders — all processed through the lark-approval AI skill. Here is what went right, what broke, and whether I would do it again.
I handed my AI agent the keys to our team’s Lark documentation for a full work week — reading, writing, organizing, and searching across 40+ docs. Here is what worked, what broke, and how the LarkDoc skill changed how I think about AI-assisted documentation.
I tried generating AI images directly inside Claude Code using the ai-image-generation skill. 50 models, one terminal, and a few surprises along the way.
