March 8, 2026
Vibes > Vectors?
You Don't Need a Vector Database
Dev community says skip the math box—just make search smarter
TLDR: The article says most teams don’t need a vector database—just smarter search—since vectors are storage, not understanding. Comments roast the hype as marketing, joke about using simple math, and push Postgres hybrids, making this a cautionary tale: don’t overbuild infrastructure when you just need better results.
Today’s hottest tech fight: do you really need a vector database, or are we just chasing shiny math boxes? The article argues these tools only store numbers and don’t magically understand “affordable hiking boots.” Cue comment fireworks. noemit blames the craze on RAG (retrieval-augmented generation), saying it was a marketing fix to calm fears about AI making stuff up. sirfz drops the meme of the day: “you don’t need a vector db, you just need np.dot,” translating to “try simpler math before buying infrastructure.” Meanwhile, kenforthewin suspects this is really a sales pitch for an API that just wraps a vector database—“you don’t need the box, you need our box for the box.”
The thread dishes practical alternatives, too: cpursley says just use Postgres with classic keyword search and a vector add-on, a.k.a. hybrid search, via postgresisenough.dev. And the moderation drama pops when exhost warns, “Please don’t use HN primarily for promotion,” as the crowd side-eyes vendor vibes. The article concedes niche teams—ML tinkerers, serious RAG builders, and researchers—do need vector DBs. But the audience’s mood? Stop building infrastructure first and give users better search that understands plain language. Less hardware flex, more helpful results.
Key Points
- •A vector database stores and indexes vectors but does not create embeddings or understand domain semantics.
- •Making a vector database useful requires substantial infrastructure: embedding pipelines, data sync, a separate primary database, query resolution, and model lifecycle management.
- •Starting from infrastructure often leads to weeks of setup and iteration for results comparable to using a simpler search API.
- •Vector databases are most appropriate when teams need fine-grained control over the vector layer, such as in custom ML retrieval systems, specific RAG pipelines, or research settings.
- •For many teams seeking natural-language-aware search, a managed search API may be sufficient and faster to implement than operating a vector database.