TechCrunch ran a piece this week titled “Google packs Search and Gemini with new AI study tools,” and I read it while a friend was texting me screenshots of a Canvas page she couldn’t get any AI to look at. Guided learning in AI Mode, quizzes and flashcards generated from material you hand Gemini, step-by-step walkthroughs of problems. It all works. It works on Google’s pages, and on the open web, and on files you’ve already pulled down onto your own disk.

That last part is the whole story.

The part of studying that happens behind SSO

Think about where a semester actually lives. The syllabus is a PDF attached to a Canvas module. The assigned reading is a link that routes through your university’s EZproxy and lands on ScienceDirect with an institutional token attached. The lecture slides are in Blackboard under a week folder someone named “Week 7 (rev2 FINAL)”. Your own notes are in Notion. The discussion thread where the professor clarified what the essay prompt actually meant is in the LMS, three replies deep, and it will never be indexed by anything.

None of that is public. All of it requires you to be signed in as you, through your school’s identity provider, with a session that expires and gets renewed by a Duo push on your phone. Google’s study tools cannot see any of it, and I don’t think that’s an oversight they’ll patch. Search indexes the open web. Gemini reads what you upload. Neither of those is a mechanism for reaching a page that only exists because a SAML assertion said you were allowed to see it.

So the tutor Google just built is a tutor who reads the syllabus you photocopy for it but never gets a library card. Which is fine for “explain the chain rule” and useless for “what does this professor actually want in the response paper due Thursday.”

Upload is a workaround, not a feature

The suggested fix is always the same. Download the PDF, then upload the PDF.

Try it on a proxied journal article sometime. The link is a redirect chain, the viewer is a JS reader that doesn’t hand you a damn export button, and half the course materials in Canvas are HTML pages rather than files, so there’s nothing to download in the first place. Then you’re screenshotting. And once you’ve uploaded a licensed article to a third party, you’ve made a copy of something your library leased on your behalf, which is a decision worth making deliberately rather than by reflex.

The other door is OAuth

Grant a study app access to your Google Drive and your calendar. Now a token exists that reads your files whether you’re at your desk or asleep. I’d rather not.

Your tab is already inside

The alternative is boring and that’s why I like it. A browser agent in the side panel reads the page rendered in front of you, using the session you already have. No export, no upload, no token, no second login. If you can see the article, so can it.

We built Dassi around exactly that constraint, and the study cases were not the ones we planned for, but they’re the ones people keep writing in about. Open the proxied paper, ask for a summary of the methodology section. Sit on the Canvas assignment page and ask what the rubric is actually weighting. Have it pull every due date out of the module list into something you can paste into your calendar yourself. Cross-reference your Notion notes against the lecture slides in the other tab and flag what you never wrote down.

It runs on your key or your ChatGPT login, so the model reading your coursework is one you picked. This is the same gap we wrote about in Your AI Browser Agent Can’t See That You’re Logged In, and the same shape as Google’s Workspace “intern”, which shows up eagerly for Docs and Sheets and goes quiet everywhere else.

Where I’d be careful

Reading a page is not the same as being right about it. The agent will confidently summarize a paper it half-parsed, and on a badly-rendered PDF viewer it sometimes gets column order wrong, which matters a lot in a two-column journal layout. Check anything you’re going to cite. And your institution’s academic integrity policy has opinions about AI that a browser extension does not override.

Google will keep shipping study features, and they’ll keep being good, and they’ll keep stopping at the point where you have to type a password. The library got there first, honestly. It’s had a turnstile for two hundred years.