galenward.bsky.social
63 posts
20 followers
80 following
Getting Started
Active Commenter
comment in response to
post
Thanks for the heads up! I'm on bright mode 24/7, so never would have noticed.
comment in response to
post
It’s dark mode. I’ll fix it tomorrow. Or Claude code will at least.
comment in response to
post
You may have it system-wide?
Either way, I should be able to reproduce.
comment in response to
post
What on earth?!? I’ll look into it. I take it you’re on dark mode.
comment in response to
post
The secure stuff probably doesn’t even work when you’re in Russia!
Also lacks emojis.
comment in response to
post
> use some nice icons for those buttons?
comment in response to
post
> The spacing between the entries like "core strengths..." and "Select the top 3-5..." is weird. Fix it. There shouldn't be buttons for "import from URL and enter manually.” In fact, just get rid of the URL import for now and make it manual always.
comment in response to
post
> Code review time. go through, do things like turn buttons into react components (think about what else falls into this bucket), look for code that could be simplified, and think about how to make it easier to update.
comment in response to
post
> Can you make name not required? (The email can just say "Hi," - the prompt already covers this). Also, can you update the "Send via Email section - we don't need all that junk. Just put two buttons: "Open in Email App" and "Copy"
comment in response to
post
But if all new code is generated by LLMs, you won’t have many examples in the dataset for the model!
comment in response to
post
This is why they have legislative aides! To answer the dumb questions and prepare them to ask hard questions!
comment in response to
post
Some in the comments say this specific video is old and unrelated. I can't confirm.
Protests are happening though: www.nytimes.com/video/world/...
comment in response to
post
New footage: www.nytimes.com/video/world/...
comment in response to
post
Claude Code will at times copy and paste, inline and flail like a junior engineer, but you can always tell it to review and apply best practices for an upgrade.
comment in response to
post
Catch 22: there definitely aren’t high-quality examples in my code base unless Claude Code wrote them.
comment in response to
post
🔥
comment in response to
post
Ultimate challenge: neg New Yorkers into voting for a YIMBY mayor.
comment in response to
post
Schumer had options short of blocking the bill, many of which would have made his base feel like he was fighting and angry.
He left everyone feeling like he just rolled over.
comment in response to
post
TBH, this is the only way to AI code.
Everything needs to be old, have lots of writing and examples on the web, and to have been boring by mid-2023.
comment in response to
post
I’ve never written a JavaScript app and built one from scratch, deployed on Vercel
comment in response to
post
It does vacillate between highly capable and script kiddie when writing code, but you can always say "best practices time: clean up the code and follow best practices" and it'll do better.
comment in response to
post
I'd suggest Claude Code is more human - we all make mistakes as we code, but we have systems for catching them. Claude Code does a pretty good job of self-correcting.
comment in response to
post
"don't start NPM server and hang every time you make a small change!"
comment in response to
post
Yeah - I have 3-4 markdown files that give Claude context. 1 is the tools we're using, 2 is the big picture goal, 3 is rules. I ask it to update the rules because it's better at organizing them than I am.
comment in response to
post
As part of my rules, I instruct it to keep a concise log of what it has done and why. That way you don't have to read all of the updates.
comment in response to
post
I love the ability to freestyle with it like this
comment in response to
post
Not my experience at all after extensive Claude Code use. What language are you programming in?
comment in response to
post
What does Cursor bring to the party in this context? Isn't it redundant? Or are you somehow still using Cursor for specific prompts?
(I run Claude Code in VSCode)
comment in response to
post
This the way to persuade Canadians in 2025.
comment in response to
post
You can be concise. “build and git commit” is sufficient to get a build, fix errors in the build, and create a git commit with a well written commit message.
comment in response to
post
18. It’s early days, so you’re still pushing buttons and pulling levers that Claude Code will eventually do on its own.
comment in response to
post
17. Ask it to save /compact to a file every time. It has a really helpful “logical next steps” section.
comment in response to
post
16. It’s early days, so nearly every day it gets an update that makes it meaningfully better. No longer do I have to type /compact to help it manage its context window.
comment in response to
post
15. Don't overdo it with context and rules: you need to give Claude Code big picture, what part of the big picture we’re working on now, and what we’re specifically doing right now. If overload Claude with a million rules and context, you make it dumber.
comment in response to
post
14. “Fix the text” or “make the UX snappier” is a better way to get creative solutions than “make that into an accordion.”
comment in response to
post
13. Find the right level of abstraction. I gave it a markdown describing a project step-by-step and asked it to “start with the header and footer” and it just went ahead and built out the 3 page app. Amazing!
comment in response to
post
12. Use your favorite LLM to clearly describe how your app will work and each page of the app. It doesn’t have to be perfect - Claude Code will improvise if you start a list and say “plus whatever else makes sense.” Put your description of how the app will work into a markdown file.
comment in response to
post
11. It isn’t great at writing microcopy. The context of coding seems to wash out its writing ability, so all the content is highly descriptive and flat. A text input for name does not need a verbose reason for why you’re inputing it.
comment in response to
post
10. Treat Claude Code like a team member. Your team members can make lots of decisions if they understand the big picture goals. Give it a 1-2 page markdown file describing the north star and how what we’re building right now fits into it.
comment in response to
post
9. Being less descriptive is sometimes better: “Refactor the codebase using best practices. Make components and do other things that improve the codebase like that.” is better than telling it to do specific things.
comment in response to
post
8. It will also refactor, make code more maintainable, and follow best practices if you tell it to do so.
comment in response to
post
7. It will brute force things and write unmaintainable code.
comment in response to
post
6. Just paste any further errors into it. You don’t need to type anything.
comment in response to
post
5. Compiled languages and linters let Claude Code see its errors and let you skip pasting errors in.