Hacker News Reader

Aren't these abstractions often so complex, company-specific and leaky that most companies are better off building their own in-house solutions? Atleast that's what I've seen at most FAANG companies but then again they're notorious for having NIH syndrome and have the resources which your regular IT shop might not.

View on HN

IME the issue is that startup data modelling ends up in some ad-hoc framework that enforces a specific paradigm. When you try to adopt a standardized tool they don't ask "does this produce the right answer" but "can this do exactly what the old tool did", and usually it's not the case. Nobody can actually reason about the whole set of data from first principles anymore so they're forced to mechanically repeat the same exact process to get consistent results. Hell, the results might be wrong, or nobody uses them, but you don't have the tooling to detect that and you're too afraid to adopt it

Reply on news.ycombinator.com