The right question for LLM data pipelines: which stages keep determinism guarantees

aronchick · x · 2026-09-26

Part 2 of aronchick's thread on LLMs in data pipelines: classical pipelines trade flexibility for reliability, needing a code path per edge case. LLM stages change the contract — the model classifies unknown records by reading them, but may classify differently tomorrow, making retries unsafe. The real question isn't "should we use LLMs" but which stages get determinism guarantees and which don't.

Related event: aronchick: In LLM data pipelines, draw the determinism boundary first(7 posts)→

Original post →

More from coding & agent

coding & agent channel →