How I’ve Been Using Claude Code to Learn
A small example from a recent learning loop around Git Worktrees
Back in 2016, just getting started in my software development career, I wanted to understand Git properly. So I read the Pro Git book. Cover to cover. Over multiple days. Not to memorize commands, but to understand how it actually worked.
For nearly a decade, that foundation served me well. I became the person on the team who could untangle merge conflicts, recover lost commits, and help engineers out of messy Git situations. It saved me countless hours over the years.
This week, I discovered git worktrees, and I had no idea what they were.
A Gap I Didn’t Know I Had
Lately, I’ve been following a handful of developers who are writing and experimenting in the open with Claude Code. They show up in my timeline not because I deliberately curated a list, but because I keep lingering on posts about AI-assisted development.
One day, I came across a thread of practical tips for getting more out of Claude Code from Boris Cherny, he is one of the creators of Claude code itself. One suggestion was about using “learning mode” to have Claude generate visual HTML explanations of concepts.
The other one mentioned using worktrees for programming in parallel.
I realised I’d never really encountered worktrees in practice. I tried to recall if they were covered in the Pro Git book. I couldn’t remember seeing them. A quick search confirmed why: git worktrees were added in Git 2.5, released in July 2015, just after the version of the book I had read.
No big mistake. Just a feature I’d never had a reason to reach for before.
Until now.
How I’ve Been Learning Lately
After reading that thread, I decided to combine two of Boris’s tips: learn about git worktrees using Claude Code’s learning mode.
Learning mode is one of Claude Code’s output styles. You enable it by typing /config and selecting “Learning” from the options. It changes how Claude responds: instead of just giving you answers, it explains the why, offers insights that connect to broader patterns, and sometimes asks you to write small pieces of code yourself so you’re not just passively reading.
I asked Claude to generate a visual HTML explanation of git worktrees. What came back was an interactive diagram showing how worktrees share the .git directory, how branches map to directories, and what happens when you add or remove worktrees.
I still believe in the deep-reading kind of learning. The Pro Git book was dense, comprehensive, and required long stretches of focus. Sitting with a well-written book, building a mental model slowly, establishing depth. Those things compound over time. Reading that book was a great help to my career and has saved me countless development hours, week after week.
But this was different. Immediate, visual, iterative. I could ask follow-up questions. Not better or worse. Just a different tool for a different moment.
I turned that exploration into a small Vercel app that explains worktrees visually. Because if I just learned this after 10 years with Git, maybe others will find it useful too.
The Pace of Now
There’s a strange mix of emotions that comes with being a software developer right now.
On one hand, I feel lucky. The tools available to us are extraordinary. I can talk to an AI that understands code, generate visual explanations on demand, and ship small interactive apps in an afternoon.
On the other hand, there’s a quiet unease. A sense that the ground is shifting. That I might not always be learning the right things. That the opportunity cost of choosing one path over another feels unusually high.
And yet, in evenings like this, when you stumble into something new using these tools ,it scratches that familiar itch. The simple satisfaction of learning something new, in deep detail, using AI tools of today.
So I’ll keep doing that. Notice something useful, learn about it, try it, and write it down.
For now, that’s enough.