Reflections after using Claude Code for everything
I’m a frontend developer, and for two weeks, I decided to use Claude Code for literally every task at work. What started as a simple test quickly turned into a full-on experiment. I wanted to see what would happen if I fully committed to AI-assisted coding.
During those two weeks, I worked on all sorts of tasks. Some were purely technical: implementing new features, fixing bugs, planning projects, reviewing code. Others were more organizational: writing documentation, preparing presentations, setting up meetings, creating tickets in Jira.
No matter what it was, I tried to start with Claude Code.
(Quick clarification, because the world of AI tools is getting crowded: Claude Code is an AI tool that works directly in…
Reflections after using Claude Code for everything
I’m a frontend developer, and for two weeks, I decided to use Claude Code for literally every task at work. What started as a simple test quickly turned into a full-on experiment. I wanted to see what would happen if I fully committed to AI-assisted coding.
During those two weeks, I worked on all sorts of tasks. Some were purely technical: implementing new features, fixing bugs, planning projects, reviewing code. Others were more organizational: writing documentation, preparing presentations, setting up meetings, creating tickets in Jira.
No matter what it was, I tried to start with Claude Code.
(Quick clarification, because the world of AI tools is getting crowded: Claude Code is an AI tool that works directly in your terminal, similar to Copilot or Cursor, but chat-based and fully terminal-driven.)
My notes in short
- AI speeds up your work.
- AI slows you down.
- AI makes me lazy and stops me from thinking.
- AI helps me think on a higher level and learn faster.
- AI smooths my daily flow.
- AI breaks my focus and context.
Surprised? Yeah, me too.
In short, the whole experiment can be summarized in one phrase:
It depends.
Because it really does. It depends on what you use AI for, how you use it, and what you expect to get out of it.
Claude vs. the rest
Overall, Claude Code outperformed Cursor and Copilot for me. Even though they all run similar models (in this case, I used Sonnet 4.5 most of the time), Claude handled searches better, understood project context faster, and produced solutions that worked right away.
So my first impressions were genuinely positive.
But like any tool, it shines in some cases and completely misses in others.
When AI works great
Claude was incredibly helpful for repetitive coding tasks.
If I already knew what needed to be done and could describe the steps clearly, it sped up my work massively.
It was especially powerful when working on tasks similar to ones I had done before. I could define requirements, give good examples, and let it handle the rest. Then I’d just review and polish the code.
There was something very satisfying about watching it run, test things (I also experimented with Playwright MCP, so Claude could open the browser, test flows, and fix errors by itself), and deliver results while I focused on the bigger picture.
Another area where AI shined was refactoring.
I was in the middle of a big “goodbye-Redux” refactor (hands up if you’re doing the same 🙋♀️). I had a clear plan and migration steps. Claude was a lifesaver for repetitive parts like copying, renaming, and updating imports. It handled it faster and more reliably than I could.
That freed me up to focus on architecture and structure instead of fighting with missing helper files or broken imports.
And one more huge benefit: planning and brainstorming.
When I had multiple possible approaches, Claude helped me quickly compare them, build small PoCs, and test ideas. Having that feedback loop right inside my project was awesome. Plus, the documentation generated along the way made code reviews and follow-up work smoother.
When AI gets in the way
Of course, it wasn’t all sunshine and rainbows.
Because I decided to use Claude for everything, I started… missing coding.
Sometimes I felt like I wasn’t really thinking anymore, just typing prompts. I even caught myself scrolling Slack or reading random stuff while waiting for the AI to finish its job. And that completely killed my focus.
Working with AI does make context-switching easier.
I can keep two or three threads running in parallel (as long as I don’t get lost in my Git worktrees 😅).
But once I let my mind wander outside of the task, answering messages or reading something, coming back is hard.
I’ll probably dedicate a separate post to the topic of “context when coding with AI,” because it’s a big one.
The scary part? Sometimes I caught myself delegating tiny things, like asking Claude to change a color or font size, just because I didn’t feel like doing it manually. It would’ve taken seconds.
That’s when I realized: I don’t want this to become my normal.
I don’t want to show up, type a few prompts, wait for AI to spit out the result, and call it a day.
Sure, I still reviewed everything, made sure I understood the code, and didn’t commit anything blindly.
But sometimes I didn’t feel as connected to what I built.
The “is it worth it?” dilemma
Another big question that came up, one many devs probably face, is:
When is writing a prompt actually worth the effort?
Sometimes I already know exactly how to solve a problem.
But since I promised myself to do everything with Claude, I had to stop and write a full prompt, add examples, and provide context, and that took time.
Was it worth it? I’m still not sure.
If I already know how to do something, maybe it’s faster to just do it myself.
On the other hand, if I can hand it off to AI and move on to something else, maybe that’s still a win.
But do I really need to move on?
That’s where the line starts to blur.
One advantage here is that writing a good prompt helps organize your thoughts.
Often, when I started describing a problem, I realized I hadn’t considered something: “Oh wait, we didn’t think about X!”
I might’ve noticed it later while implementing, but AI forced me to spot it earlier. That’s a win.
Setup fatigue and the automation trap
Another challenge: setup time.
Because it was a new tool, I was learning on the go, testing sub-agents, slash commands, and automation workflows.
Some of them worked beautifully. Others? Total flops.
I built two helpful sub-agents and two useless ones.
Finding the balance between “should I automate this?” and “should I just do it manually?” was tough, especially while handling real project work.
I couldn’t disappear for a week saying, “Sorry, I’m configuring AI agents.”
And with how fast the AI landscape evolves, sometimes investing time in setups feels pointless, because by the time you finish, a new and better version is already out.
Final thoughts
Whew, that was a long one.
But maybe that’s fitting, because my feelings after this experiment are just as mixed.
Claude Code has become my main coding companion, but not for every task.
I don’t think there’s a universal rule like use AI for this, but never for that.
Everyone needs to find their own balance.
For example, I love writing tests, so I don’t delegate them to AI.
But when it comes to setting up folder structures, I gladly hand that off.
Honestly, I enjoy watching the automation unfold.
Still, the bigger question remains:
How much should I give away to AI, and what should I keep for myself?
When does “AI writes my code” turn into “I’ve forgotten how to code”?
In the past, we talked about tutorial hell, where you can follow examples but can’t build from scratch.
Now we might be entering AI agents hell, where the code looks fine and feels like something you’d write, but would you, really?
On the other hand, is writing from scratch even the true test of skill anymore?
Knowing docs by heart doesn’t make you a great developer.
Even AI knows the docs and still gets lost in real projects.
So maybe the question isn’t whether to use AI, but how consciously we do it.
Does offloading code generation free us up to focus on what really matters, or is it just an illusion of progress?
💬 What do you think?
Have you tried a similar experiment yourself?
Or is there something you’d like to ask about the experience?
If you enjoyed this post, follow me here on dev.to