Dexter Horthy discusses the evolution and lessons learned from using coding agents, particularly the Research Plan Implement (RPI) methodology. He admits initial mistakes, such as not emphasizing code reading and over-relying on lengthy plan files. Horthy introduces CRISPI, a refined approach involving distinct stages like Questions, Research, Design, Structure, and Implement, to enhance instruction following and leverage. He stresses the importance of not outsourcing thinking and advocates for human-agent alignment through design discussions and structural outlines. The goal is to achieve a 2-3x speedup in development while maintaining code quality and team alignment, contrasting with the risks of unaudited "slop" from faster, less controlled AI code generation.
Sign in to continue reading, translating and more.
Continue
