Harshit Omar, CTO of FluidCloud, gave me the sharpest answer I've heard.
FluidCloud makes multi-cloud migration easier. The hard part isn't tooling. It's that every cloud models the world slightly differently, and the gap between "looks equivalent" and "is equivalent" is where migrations quietly break.
They use AI to read messy infrastructure and map one cloud's concepts onto another. Good fit. Models are strong in that ambiguous middle.
Then Harshit told me about the four months of AI-based mapping they built, watched produce plausible-but-wrong output, and deleted. Rebuilt the harder way, by hand.
His reasoning is the part I keep thinking about, and it flips how most teams talk about AI in production.
(FluidCloud raised 8M USD seed from Unusual Ventures and others, which tells you how real this problem is.)
I wrote up the full conversation, including exactly why he pulled AI out of the layer that has to be correct every time.
If you're building in this space, where have you drawn your line between the AI layer and the deterministic one?
Where LLMs Belong in Cloud Migration — FluidCloud's Determinism Boundary | goBIGai
Why FluidCloud rebuilt its multi-cloud migration engine without AI — deleting four months of LLM-based mapping for deterministic rules. A field guide to where LLMs belong in infrastructure.
gobigai.org
Top comments (0)