DSPy Hits Adoption Wall

DSPy feels like overengineering until you’ve built a few brittle prompt systems; then it looks like the missing layer you should have had from day one.

• –The post’s strongest point is architectural: typed I/O, composable modules, separated prompts, and early evals are exactly the pieces serious LLM systems need.
• –Its “DSPy at home” examples explain the adoption gap well: shipping pressure makes prompt spaghetti feel cheaper until model swaps, retries, and maintenance get ugly.
• –The article cites production users like JetBlue, Databricks, Replit, VMware, and Sephora, which suggests DSPy already has proof points even if it lacks mainstream mindshare.
• –The practical takeaway is framework-agnostic: make model swapping, testing, and optimization first-class from day one, even if you never adopt DSPy itself.