I used to spend more time managing my studying than studying. Notes scattered across three apps, flashcards half-made, highlights I never revisited. The fix was not more discipline — it was a pipeline. Today, almost everything between “I found something worth learning” and “it is scheduled for review” runs itself, and my active part takes about fifteen minutes a week.
Here is the whole system, honestly annotated, including the parts that broke.
Step 1: One inbox, zero decisions
Everything I might want to learn — PDFs, articles, lecture slides, my own messy notes — lands in a single folder in Notion (a plain folder on disk works too). The only rule: capturing costs one action. If saving something requires thinking, I will skip it, and so will you. No sorting, no tagging at this stage. Decisions at capture time are where systems go to die.
Step 2: The summarization prompt that runs itself
Once a week I drag the week’s accumulation into a Claude Project (ChatGPT with a custom GPT works identically) that carries this standing instruction:
For each document I upload, produce: (1) a 150-word summary in plain language; (2) the 5 claims or concepts worth remembering, each with a one-sentence explanation; (3) a list of anything the document assumes I already know. Output as markdown, one section per document.Fifteen documents process in one batch while I make coffee. The third output — what the document assumes — has quietly become the most valuable: it keeps a running list of my prerequisite gaps, which is exactly where my studying should go next.
Step 3: Flashcards without the typing
The summary’s “5 concepts” sections get pasted into a second prompt that outputs Anki-ready CSV: question, answer, and a tag for the source document. I import the file, and spaced repetition takes over from there. Card quality sits around 80% of hand-made — good enough to review, and anything shaky gets edited the first time I miss it. Editing a bad card after a failed review takes ten seconds and is itself a form of studying.
Step 4: Review is scheduled, not chosen
Anki decides what I see each day. On a good week this is 10–15 minutes; on a bad week I let the queue slide and it forgives me — that is what the algorithm is for. The one habit I protect: cards I answer wrong get a manual pass. If I cannot explain why I was wrong, the card was bad or the concept is, and I fix it on the spot.
The glue (and what broke)
I tried to automate the handoffs themselves with Zapier — auto-summarize on file upload, auto-import to Anki. It worked, technically, but it silently produced worse summaries: batch processing with one careful instruction beat per-file triggers on both quality and cost. So the honest version of this pipeline has exactly two manual moments: dropping files into the inbox, and clicking “generate” on a weekly prompt. Everything else is a saved prompt and an import button.
Where to start
Do not build the whole pipeline this weekend — that is RemNote-style over-engineering, and I have the abandoned database to prove it. Start with one document you genuinely need to learn. Summarize it with the prompt above, turn the five concepts into cards, and review them for two weeks. If the system proves itself on one document, it will survive ten. If you would rather see where all this is heading beyond personal productivity — classrooms, curricula, AI tutors — my next post covers the trends worth watching.
Read next: 5 EdTech Shifts That Will Actually Matter by 2030 (And 3 That Won’t)