Hi
I have tried with ChatGPT agent. I asked it to go into Ebscohost databases and summarize top ranked papers on topic x.
During the login process, it would hand over control to the human. OpenAI claims to be not logging/screenshotting when the human takes over. Once the human hands over to the agent, the agent will continue it's task
I also managed to ask it to go into Scopus and do research on X. Again, it worked as expected, though time was up after just 30 minutes.
Thus far though, I don't think it is very good. It is extremely slow, cross domain errors, the agent can get confused when there are too many links and ends up just "googling" sometimes (e.g. Looking at your Primo).
So far, I don't think i have seen it use full-text from our databases either. Because it is so slow, I think it managed to only summarise maybe 10 articles from Scopus because the 30 minute time was up.
Oh and don't brother to try using Google Scholar via library link, it's totally blocked.
At this stage, I feel academic (not general) deep research tools are still better than just straight out using agent to search databases. That said SciSpace Agents seem to do very well.
In the long run, I expect a blend of techniques, for example there is little point doing a "live browse" of website x, when your search engine is setup with direct access via a vector database, which is much faster and more stable.
Still, I can see agents being called for one particular use case, download full-text via institutional entitlements.