Given that reddit threads on Claude's performance over the past couple of months are making their way to HN as 'news', I thought this post was interesting enough to add (yes, I see the irony).

If you are willing to rule out mass hysteria (which may or may not be reasonable) - the question is whether continual internal tweaks - for whatever reason - affects performance to a degree that whether or not you are able to use the platform for 'real work' (define it as you wish) can fluctuate from month to month or even week to week. - should you off-load key work to an industry that has not yet found a success business plan? I'm just an AI hobbyist, but this seems insane to me. Of course, if you are on an enterprise plan there is some buffer. But there is no guarantee.

So how do you mitigate this risk?