# Reading Alongside LLM Agents *February 2026* The iPad and the e-ink tablet I bought for reading a few years ago are gathering dust on my bedside table. Instead, half my reading happens in Claude Code. Not what I would have pictured a year ago. I've been a heavy [Readwise](https://readwise.io/) user for years, and a [Reader](https://readwise.io/read) user since it launched in 2022. Late last year my reading workflow shifted from apps and devices to conversations with LLM agents. Here's what that looks like. I drop a YouTube link into Claude. It saves the video to Reader, pulls the transcript, and we talk through it. I ask it to save the best parts as highlights, paraphrasing the content as needed. What used to be a juggling act across apps is one conversation. Synthesised highlights work especially well for verbal content - podcast guests and YouTubers don't always express ideas cleanly. The agent takes the raw, meandering way someone expressed a thought and reshapes it into a cleaner form. ![[readwise-claude-cowork.png]] Books are even better. I ask the agent to map out the structure, then stitch together just the parts I care about, even across chapters. I compare ideas across multiple sources in the same conversation. For topics that are hard to picture, I get it to generate mind maps in Excalidraw, sketch ASCII diagrams, or code up throwaway demos. Whatever form makes the concept click. None of this applies when the writing itself is the point. But the faster I get through everything else, the more time I have for the reading that actually rewards slowness. It also changes how I triage. Before when someone sends me a link, I copy it, switch to Reader, paste it, switch back. Multiple taps to save one article, and I still have to skim it later. Now I just forward the link to the agent and I'm in the process of setting up automated triage - the agent pulls my RSS feeds overnight, ranks by relevance to what I've been reading lately, and has a shortlist ready before I sit down with coffee and laptop. These workflows came up often enough that I packaged them into a [Readwise skill for Claude Code and Codex](https://github.com/ryanlyn/readwise-skill) - it handles core actions like saving highlights, pulling articles, searching highlights, getting feeds, and lets you compose them all within one conversation. I open-sourced it if you want to try something similar. I still feel like I've barely scratched the surface. There are a lot of questions that can be asked when the agent works across everything I've read. What do years of highlights say about a topic? Where are the gaps? How has my thinking shifted over time, traced through margin notes I've forgotten I made?