Have also realized with the whole personal agent experience is you think less about model choice and more about whether it’s a limitation of form factor (which isn’t actually a bad thing) I’ve off loaded a lot of stuff to Grok Bot / Instinct and despite getting better outputs from GPT 6-Astra for example for most nearly everything, not really worth the effort/convenience/time to route it there I still focus my research, writing, thought intensive tasks into the most capable models but there’s plenty that can and should be handled by lower models that were frontier not too long ago. Not having the ability to flip up or down compute solves it for you and make it less top of mind vs chatbot where you notch the effort or model higher and immediately see improvement which reinforces the psychology you always want top model top effort for even simple tasks Interesting anecdotal experience