March 23, 2026
Hype train, empty seats
If Dspy is so great, why isn't anyone using it?
Big brands use it, devs still wary—smart fix or slick sales pitch
TLDR: DSPy claims big wins and lists major users, but adoption still looks small. The crowd is split between fans who say everyone ends up reinventing it badly and skeptics who fear lock‑in and call the post a sales pitch, turning a niche tool into a full‑blown comment‑section brawl.
DSPy promises to tame the chaos of building AI apps, and the blog touts real names—JetBlue, Databricks, Sephora, VMware, Replit—saying their systems got easier to maintain and faster to test. The author even drops a new “law”: build anything complex enough, and you’ll unknowingly rebuild half of DSPy yourself. But the download charts still look lonely, and that’s where the comments turn into a cage match.
One camp swears the tool is gold but admits the learning curve is steep. sbpayne says the quiet part out loud: everyone ends up making a janky “DSPy-at-home” anyway, wasting time and pain. On the other side, TheTaytay bailed after trying it “in earnest,” spooked by the idea of fully buying into a framework and letting a computer spit out a mysterious, uneditable prompt—“an opaque blob.” Then the spice hit: stephantul calls the post a straight-up ad for the author’s consulting, while tinyhouse shrugs that the whole DSPy/RLM vibe feels “more marketing than problem-solving.” Even the ops nerds got a meme in—QuadmasterXLII roasted the example of adding a database to avoid redeploys, calling it “hair-on-fire” deployment hygiene.
Result: a split-screen drama—true believers versus framework-phobes, with a side of “is this content or commerce?”
Key Points
- •The article highlights a gap between DSPy’s promised benefits and its relatively low adoption compared to frameworks like LangChain.
- •Several companies (JetBlue, Databricks, Zoro UK, VMware, Sephora, Replit) are cited as using DSPy in production.
- •The author attributes low adoption to DSPy’s unfamiliar and harder abstractions, not to incorrectness.
- •A humorous “Khattab’s Law,” inspired by Greenspun’s Law, argues complex AI systems end up re-creating parts of DSPy ad hoc.
- •A stepwise example shows how teams evolve from simple OpenAI API calls to externalized prompts with versioning and structured outputs via Pydantic.