Learning in the Age of Agentic AI
Oct 19, 2026

I wanted to write a note about how I use agentic AI for learning. First because it’s novel (to me at least). And second in the hope that someone else finds it useful.

I use OpenClaw – an agentic AI harness – that I connect to Anthropic’s Claude using metered intelligence via their API.

The process takes about an hour to set up (with help from AI), and now I’m able to chat with OpenClaw using Discord (WhatsApp, iMessage, etc, are also available). Using just my phone, I can write files to my computer, request files, move them, and so on.

First I’ll tell you about my reading process. I’m basically old-fashioned: I print things out, I underline important passages with a pen, I write marginalia notes, questions, and I make arguments as to whether I agree or disagree with the claims being made in the paper or book that I’m reading. In between, I take long walks to think about ideas, what they mean, how they might connect to other ideas, or how they fit into my broader worldview.

Once I make my initial pass, I add highlights and notes into the digital version of the paper on Zotero. But since human memory is not optimised for memorisation, this note-taking process tends to break down.

In 1885, the psychologist Ebbinghaus showed us there is a ‘forgetting curve’ - within a day or so after you learn something, your recall will drop significantly. This process continues unless something is done to reverse it, to ‘flatten’ the curve.

Later, spaced repetition was developed as a way of slowing down the forgetting curve. You review information at increasing intervals (e.g. 10 minutes after, then at 1 hour, then 1 day, then 1 week, and finally 1 month).

This is where agentic AI comes in. Once I read a paper, I drop it into my Discord chat with OpenClaw and ask it to quiz me on the paper using Anki-style questions. Here’s an example exchange:

OpenClaw:
“In De Anima II.5, Aristotle describes the move from having knowledge (first actuality) to exercising it (second actuality). Is this an alteration in the strict sense? Why does that matter for Sisko’s argument?”

Me:
“It is not an alteration in the strict sense. Aristotle claims it is not an alteration, or it is an alteration of a different kind to that of material alteration. It matters for Sisko’s argument, that material alteration is necessary for thought and perception, because it raises doubts about proposed claims in the literature that DA II shows that material alteration is not involved in perception.”

OpenClaw:
“Correct. You’ve got the main point: the move from first to second actuality isn’t an alteration in the strict sense (it’s more a perfecting or fulfilment). Your explanation of why it matters is also good. Some scholars take De Anima II to show that perception doesn’t involve material alteration, and Sisko wants to block that inference.

I’d tighten my own card to match your wording, because it was muddled. I’m marking it correct.”

And then on to the next question.

Using metered pricing, this process averages to about 10p per question. You could of course use an LLM through a web interface, but I personally like having the option of being able to point the agent to the PDF file in its sandbox directory.

In addition, my review questions and answers are saved automatically, and OpenClaw messages me at spaced intervals.

All in all, I’m very impressed with how well it works, and I’m excited to see how I can further leverage agentic AI to learn, while still retaining my good old-fashioned methods of working.