Ask HN: What would make an AI builder trustworthy enough to use daily?

I’m building a multi-model AI workspace (thetank.io) and the hardest part isn’t “can it write code” — it’s trust: - preview reliability - saved state/persistence - smooth iteration - fewer hallucinated integrations

For people who’ve used v0/Bolt/Lovable/etc.: What specific UX or reliability behaviors made you trust (or stop trusting) an AI builder?