Adding some fuel to the SDD + AI debate 🔥 Lots of opinions flying around on X, but I think the best way to have this discussion is to look at actual research and data Just read an interesting paper from researchers at CMU and Google analyzing how developers use Spec Driven Development They studied 266 repositories and interviewed 21 practitioners A few interesting findings: - 77.6% of specs include implementation details - Developers report more reliable agents, but also more planning and review overhead - They couldn’t separate SDD benefits from improvements in AI models. What I found most interesting is that SDD might help simply because it forces us to think more clearly about what we want As coding agents and LLMs get better, how much of SDD will we actually need?