This is basically my experience with it. I thought it'd be great for writing tests, but every single time, no matter how much coaxing, i end up rewriting the whole thing myself anyway. Asking it for help debugging has not yet yielded good results for me.

For extremely simple stuff, it can be useful. I'll have it parse a command's output into JSON or CSV when I'm too lazy to do it myself, or scaffold an empty new project (but like, how often am i doing that?). I've also found it good at porting simple code from like python to JavaScript or typescript to go.

But the negative experiences really far outweigh the good, for me.

Have you tried context7 MCP? For things that are not mainstream (like Javascript, Typescript popularity), LLM might struggle. I usually have better result with using something like context7 where it can pull up more relevant, up to date examples.

I only use 2 MCP servers, and those are context7 and perplexity. For things like updated docs, I have it ask context7. For the more difficult technical tasks where I think it's going to stumble, I'll instruct Claude Code to ask perplexity and that usually resolves it. Or at least it'll surface up to me in our conversation so that we both are learning something new at that point.

For some new stuff I'm working on, I use Rails 8. I also use Railway for my host, which isn't as widely-used as a service like Heroku, for example. Rails 8 was just released in November, so there's very little training data available. And it takes time for people to upgrade, gems to catch up, conversations to bubble up, etc. Operating without these two MCP servers usually caused Claude Code to repeatedly stumble over itself on more complex or nuanced tasks. It was good at setting up the initial app, but when I started getting into things like Turbo/Stimulus, and especially for parts of the UI that conditionally show, it really struggled.

It's a lot better now - it's not perfect, but it's significantly better than relying solely on its training data or searching the web.

I've only used Claude Code for like 4 weeks, but I'm learning a lot. It feels less like I'm an IC doing this work, and my new job is (1) product manager that writes out clear PRDs and works with Claude Code to build it, (2) PR reviewer that looks at the results and provides a lot of guidance, (3) tester. I allocate my time 50%/20%/30% respectively.

Thanks, I’ll check out Perplexity. We seem to be using a similar stack. I’m also on Rails 8 with Stimulus, Hotwire, esbuild, and Tailwind.

Playwright MCP has been a big help for frontend work. It gives the agent faster feedback when debugging UI issues. It handles responsive design too, so you can test both desktop and mobile views. Not sure if you know this, but Claude Code also works with screenshots. In some cases, I provide a few screenshots and the agent uses Playwright to verify that the output is nearly pixel perfect. It has been invaluable for me and is definitely worth a try if you have not already.

I didn’t realize screenshots worked until a few days in, that was a great discovery. And recently learned you can directly paste in copied screenshots using ctrl+v (instead of cmd+v on a Mac).