In the last couple of days, the prototyper AI is freaking out. It clames it did actions he never did, it answers half answers, it makes beginers mistakes and I find myself boosting his responds with Claude to close the gaps. Is that normal behaviour or something is happening behind the scenes?