Initial Thoughts on Claude Code

Why Claude Code?

Everyone seems to be going nuts over Claude Code these days. I've seen plenty of social media creators, ranging from experienced software engineers to (expert?) vibe-coders, swear by it and choose it over other LLM coding tools. But what really convinced me to give it a shot was seeing viral tweets from Boris Cherny, the creator of Claude Code, on his latest workflow with it. The workflow step that stuck out the most was that Boris runs about five instances of Claude Code in his terminal and between 5-10 Claude sessions on the web. This sounded absolutely ridiculous, but it did get me really curious. It got me thinking, 'Am I behind in terms of my usage of LLMs? Am I not maximizing the use of these tools?' Now, I've used other popular LLM tools such as Cursor and GitHub Copilot, but I mainly rely on them to help solve concise tasks or brainstorm ideas. I'd say I mostly use the 'Ask Mode' due to my lack of faith in the 'Agent Mode,' but I've only recently started using VS Code's 'Plan Mode,' which seems to improve feature planning and coding output with LLMs. Anyways, to scratch that curiosity itch, I decided to give Claude Code a trial run and purchased one month of the Claude Pro plan.

My Experience using Claude Code

I figured there was no better way to get a good feel for Claude Code than to build an actual project with it, which is this simple static website you're on. My goal was to get a good feel for Claude Code and see if I could crank this website out within a few hours. By default, Claude Code uses Opus 4.5 (or 4.6 now), which I actually haven't used in GitHub Copilot because it consumes your premium requests 3x faster. But hey, since Mr. Boris Cherny uses it, it's worth at least trying it out. Now, my first prompt was an ambitious one—I pasted a link to a website (www.baxate.com) and asked Opus to use it as a reference for this project. It took a while, and to nobody's surprise, the result was a very rough-looking skeleton of the reference website which could use a fair amount of touch-ups. But honestly, as a starting point, I can't say I was too disappointed. What really surprised me was when I checked my usage for the session—it was at 11% after just one prompt.

Anyways, I was also curious about trying out planning mode since Boris and many others typically start their session with it. It was pretty similar to VS Code's planning mode, and I found it to produce some pretty decent code. The planning-to-implementation workflow actually takes between 3 and 10 minutes to complete, which helps explain why Boris has multiple sessions running concurrently. However, this workflow burns through tokens like crazy, and I hit my Pro plan session usage limit within 3 or 4 prompts. So yeah, unfortunately I won't be running five concurrent Claude Code sessions like Boris unless I cough up some more money for the Max plan ($140 CAD/month!). The one pro tip I found after doing some TikTok research was to use Opus for planning and Sonnet for execution. I honestly did not notice a drop in code quality, and it did help reduce my usage, which I badly needed.

Final Thoughts

So all in all, I think I can see why everyone's hyping this tool up. I was able to build this website within a few hours while actually letting Claude write most of the code for me. I would, of course, fine-tune the plans and review the code, but it really did most of the heavy lifting. There's a lot of other neat stuff you can do, like creating skills, connecting to MCP servers, etc., which I haven't explored yet. The CLI tooling was also a lot more intuitive than I expected. Now, the biggest downside is simply burning through your usage too quickly. Seems like 3-5 planning mode sessions is the max you can get out of a session, which feels sparse. Running multiple Claude sessions in parallel just doesn't seem feasible, like I mentioned earlier. Feels like you can have two or three times more usage with tools like GitHub Copilot and Cursor. Either way, I'd still give it a leg up on GitHub Copilot and Cursor simply because the developer experience felt better. It's hard to explain, but the planning-to-implementation flow just felt sharper and easier to execute upon than in other LLM coding tools. I'm really not sure if it's because of the CLAUDE.md file that's easy to create (provides additional context to prompts), or because I mainly used Opus for planning, which I typically don't use in other tools. Regardless, I'll keep programming with Claude Code for the rest of the month and maybe even try to replicate some of the planning mode workflow at work with GitHub Copilot. It's very interesting to see that we're transitioning more and more to a review and edit workflow as these tools get better, but that's a topic for another day.