Coming at this from a design perspective, ultimately, I think it’s because we’re following and repeating the same patterns we see at larger companies or trendy startups that look interesting or useful, but when when the “end product” is repeated across domains, it’s either too broad of a solution to be useful within a specific context, or it’s too narrow to be applied to multiple contexts. AI (across the different ways we’ve defined it over time) really only enables 5 things, either individually or combined: perception, organization, inference, production, or action. If you use those enablements as starting points to ask questions about whether AI needs to be used at all in a particular context, and you can articulate WHY a system needs to sense something, and WHY it needs to act on what its sensed, then you can start to see where it can actually be useful in the context you’re building. Not everything needs to be a generator, or if it is, it doesn’t always need to be a chat interaction, etc.