When did “AI research assistant” start meaning “chatbot that summarizes papers”?

Maybe this is just me, but every time someone says “just use ChatGPT for the lit review” I feel like we’re talking about two completely different problems.

Summarizing a paper is useful. I do it too.

But the part that actually eats my time isn’t “what does this paper say?”

It’s figuring out why this paper matters compared to 40 others, which methods are basically variations of the same idea, where two subfields are making different assumptions, and whether there’s actually a research gap there or I’m just missing a paper.

That’s the part where most AI tools still fall apart for me.

I’ve tried the usual ChatGPT / Claude workflow, plus stuff like Elicit and citation-search tools. They’re useful for pieces of the processes ,but I still end up being the one manually connecting everything together. Recently I’ve also been testing mira ai science because it takes a more multi-step approach to the research question rather than just giving me another summary

Mixed feelings so far.

Sometimes the decomposition is genuinely useful and points me toward something I hadn’t considered. Other times it happily creates a sub-question that makes zero sense once you know the domain, so I definitely wouldn’t let it run unsupervised.

But it did make me realize that “research assistant” and “research agent” probably shouldn’t mean the same thing.

For me, an actual research agent would need to keep track of the original research goal while moving between literature, hypotheses, experimental choices and results — and explain why it’s taking each step.

Otherwise it’s basically still search & summarization with a nicer interface.

Curious where other researchers draw the line. What would an AI tool actually have to do before you’d call it a research agent rather than just a research assistant?