October 28, 2025
SQL slap-fight, now streaming
Text2SQL is dead – long live text2SQL
Commenters clash: keep queries in-house or ditch SQL entirely
TLDR: Exasol says cloud-based Text-to-SQL risks leaking data and pitches a local, tightly controlled “Governed SQL” setup. Comments explode: some say metadata isn’t secret, others argue SQL should be replaced, and many warn Text-to-SQL is hype that’s “dangerous” without clear business value.
Exasol’s blog tried a stunt: declare “Text-to‑SQL” dead because cloud AIs can leak database details, then shout “Long live!” if you keep everything on‑prem with Governed SQL—read‑only queries, local models, and a gateway between the AI and your data. Cue comment chaos and popcorn.
Skeptics immediately rolled their eyes. lateforwork argued, “metadata isn’t secret,” calling the whole privacy scare a niche issue. bob1029 went harsher: Text‑to‑SQL “works just enough to be dangerous,” a shiny toy that sucks teams dry, with the German‑to‑SQL demo getting meme treatment. Others just roasted the headline—“Horrible title and premise,” snapped dcreater—while accusing the post of clickbait gymnastics.
Then the futurists stormed in. memelang declared SQL is doomed, pitching a new AI‑friendly query language and dropping a link like a mixtape. Meanwhile, the SQL faithful clapped back: sschnei8 asked why people keep bashing a language that’s simple and powerful at crunching tables. Jokes flew: “Schrödinger’s Query—dead and alive,” “SQL is the cockroach of data,” and “Du hast SELECT?”
Privacy hawks cheered the on‑prem angle; pragmatists called it overkill. Folks who actually run budgets warned: local GPUs aren’t cheap. Verdict from the crowd: bold vision, spicy headline, and a classic internet split between practicality and moonshots today.
Key Points
- •The author describes three Text-to-SQL experiments: initial testing via Hugging Face, a partially on‑prem one‑shot transformer, and a final fully on‑prem setup.
- •Cloud-hosted LLM services pose data governance risks because metadata and even result data may leave company premises.
- •The article claims Text-to-SQL is effectively “dead” for organizations that cannot allow data egress due to confidentiality or regulatory constraints.
- •A viable on‑prem Text-to-SQL solution requires a local LLM server (ideally GPU-accelerated), a Text-to-SQL processor, an MCP server gateway, and a compatible AI desktop application.
- •The proposed “Governed SQL” approach enforces read-only queries, leverages learning from prior questions, and enables prompt reformulation to ensure translation quality and safety.