raf.dev
React Native developer, I like to make my own products
652 posts
985 followers
492 following
Regular Contributor
Active Commenter
comment in response to
post
Wow that's amazing! I love the video of how it was found
comment in response to
post
This one is better
comment in response to
post
i made a chart
comment in response to
post
I recently did something similar, open-meteo doesn't require an API key: open-meteo.com
comment in response to
post
This article has a lot of best practices, which have helped me out. I think it's really important to dig into the full capabilities of the tools you use.
www.anthropic.com/engineering/...
comment in response to
post
Custom slash commands dramatically improve the developer experience. You could maintain a document with useful prompts, but accessing them via /commands makes you more invested in creating reusable workflows. It's like using a prompt as a function.
comment in response to
post
You can still use images in the context of your conversation with Claude Code. I thought I'd lose the ability to use images switching to a terminal tool. But you can copy and paste images into Claude Code just like with Cursor.
comment in response to
post
Sub-agents solve the biggest friction with AI coding tools. The main friction is running out of context. Sub-agents let you parallelize work and save context since each has its own context. Just tell Claude to use sub-agents in your prompt.
comment in response to
post
Claude Code's planning approach is much better than Cursor. The way it responds in chat gives me confidence that it's breaking down work into tasks properly. Instead of jumping straight to implementation, it reads files first, thinks through the problem, and makes a plan.
comment in response to
post
Oh yeah, making a plan is great. You can ask the agent to keep improving and improving the plan, and it will give it some really nice rails to execute well on. I'm glad it was helpful!
comment in response to
post
lol, "should I cancel the DJ out of respect?"
comment in response to
post
Ehhh, I like grouping tests in describe blocks because it's easy to do manipulation like xdescribe or describe.only
Also with AI writing the tests, the unit tests are going to become kind of a black box that is a lot more thorough than what we can write, and we won't pay that much attention to them
comment in response to
post
That's reasonable. Maybe the avant-garde will always exist. I think there will always be people who push the boundaries of what's available to them, but I'm worried we'll be so stratified that newer practices don't get democratized.
comment in response to
post
You shall have
comment in response to
post
That's where being an Android user really helps. It's the least pain in the butt developer experience
comment in response to
post
Put it on the backlog and prioritize it I guess. It depends on the values of the team. I think products are hurt more by what's broken than they benefit from adding new features. Once you lose user trust it's hard to get it back
comment in response to
post
I use Cursor and I find it very useful. People also talk about Windsurf but I haven't tried it out.
comment in response to
post
Part of being a developer is staying up to date with language features, libraries, etc. It's the same for AI tools. Cursor and Windsurf are the most commonly talked about AI editors, and they give pretty good results.
comment in response to
post
The problem is that you're using GitHub Copilot. There are way better tools out there.
comment in response to
post
You can't trust a Samsung keyboard
comment in response to
post
I'll be there 👋
comment in response to
post
thanks! I've been trying to go for a terminal aesthetic but I haven't been great at implementing something that just hits it for me yet.
comment in response to
post