This week
A few months ago a lot of people started using an app called Warp to interact with Claude Code. Warp is a terminal app that replaces your computer's native one. Warp markets itself heavily on its AI features.
I resisted it for months even as I saw it being hyped everywhere I looked among Claude Code users. I tend to resist hype and dig my heels in even more.
Then I downloaded it. Within 10 minutes I was completely sold, and the funny part is that none of the things that won me over were the AI features. Warp markets itself as an AI coding platform. I'm ignoring all of that. I already have Claude Code.
I used to have my Claude Code window on one side of my screen, Obsidian on the other, constantly switching between them. That setup is gone. When I'm doing Claude Code work now, everything lives in Warp.
I made a video walkthrough of the four features that made the difference for me. Link below.
Watch: Warp for knowledge workers →
Read: Full companion guide on cleverest.me →
What I'm reading
Claude Dispatch and the power of interfaces → — Ethan Mollick / One Useful Thing. In the first paragraph on my website, I write "...for most teams I work with, AI still means copy-pasting between tabs and using someone else's prompts." Mollick expands on this type of friction faced by knowledge workers using AI. He argues that the interface is now the bottleneck to adoption. Claude Code is powerful but intimidating. Working all day with a chatbot is exhausting. What's next then? I think using the interfaces you already have (integrating AI in all the places you currently work) is what works best now. The future may likely hold more "interfaces on demand" (read more about that in the article).
I got curious how compaction works as a PM → — Tal Raviv / X (thread). A short but sweet thread written by a product manager who got curious about how Claude Code actually works under the hood. He dug into the files on his computer and watched what happens when a long conversation gets compressed.
Learning by getting it wrong on purpose → — Paul Kirschner / kirschnered.nl. This article is courtesy of my background in learning science. Did you know research shows that getting something wrong and then correcting yourself is one of the best methods for retaining information long term? This also works even if you know the right answer and deliberately get it wrong on the first pass. That's what I do with Claude Code all day. Try something, watch it fail, try again differently. Failing is the most useful part of the process for me.
Slow note
One day last week I dropped my daughter off at daycare and sat down to my computer at 7:45. Before I even finished my coffee, I was already juggling website updates, content planning, workshopping YouTube short ideas with Claude Code. By 10:30 I felt dizzy. I needed a screen break. I decided to crochet a rose.
You can knit with a machine, but no one has ever invented a machine that can crochet. It's too complex for automation.
I'm optimistic for the future of AI. But if it really does all go to hell, I imagine sitting in the middle of all the chaos, just crocheting.