Claude picks the first idea that works. Make it pick the best one
SMRTR summary
Claude AI tends to "satisfice" by choosing the first workable solution rather than exploring better alternatives, which is problematic for important design decisions with high switching costs. A new custom skill called "Replan" forces Claude through structured critique-and-refine loops, generating multiple alternatives and stress-testing assumptions across at least three iterations before finalizing plans. This approach addresses Claude's tendency to commit to initial ideas by requiring systematic evaluation of simplicity, maintainability, and technical assumptions, producing more thoroughly vetted solutions for critical architectural decisions.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article