Agreed. He ascribed politics to the product and policies, while also wanting to be the main character of ... everything. Naturally, entangled interests also yield entangled consequences.
Pushed straight to main!
New community gem server from the rubygems team just dropped: https://gem.coop
Reason 249 why SMS 2FA is a bad idea
Be careful with the Notion MCP in Claude Code (or elsewhere). I was surprised to see it uses 30k tokens -- 15% of the total context.
You're absolutely right!
When I don’t have the energy to figure it out:
Me: Can you just fix this?
AI: Sure!
Me: Wait, no, not like that!
Suddenly, I found the energy.
I used whiteboards so much, I carried my own marker
Crisis PR firm definitely earned its fee.
Except they still need to remind us it was “formerly Twitter.”
Claude Code needs real-time usage feedback. Novices users are ultrathinking everything with giant, copypasta instructions and poor context management. The result is a surprise model downgrade after their Max plan usage runs out. They'd get better results if they were shown how their quota was spent.
r/ClaudeAI/ is hilarious. Half of the sub cannot believe how good Claude Code is and heaping effusive praise. The other half thinks the first half are liars or rubes.
Ronaldcore
More:
- Stand up a new Redis cluster via OT Container Kit
- Investigate my AWS bill and suggest ways to save
- Add an EBS volume because I don't want to spend 15m reading docs
- Archive a log table to S3
- Fully document a cluster's configuration
Here's something I've been working on in my spare time over the past few weeks. Your AI agent can now talk to Honeybadger. 👀
Claude Code is excellent at DevOps. Just show it where to get the API key and remind it that it has access to your servers (including prod). Recently I had it:
- Fix a crashed container
- Diagnose app errors
- Requeue broken jobs
- Download and analyz raw logs to find a bug
Best $100 I've spent this month!
It's so much better than Cursor because it iteratively digs deep for solutions. I've handed off long PRDs, complex bugs, and open-ended "what do you notice about this code?" issues.
I think Cursor's system prompts encourage a quick return back to the user ... it's lazy.
I like the Karpathy's original definition: forgetting the code exists.
But we need another name for what you're describing (which is what I do). It's speedrunning all of the steps of the normal software dev process, but letting us focus on requirements, reviews, and results.
Nearly two years later, they’re still trying to get the name to stick.
I've transcended into running multiple Claude Code instances at the same time.
Extremely practical demonstration of using Cursor effectively. This is similar to Kevin Leneway's approach, which completely transformed how I work.
I'm still using ChatGPT + o3 for planning, though, and using Claude Code to implement.
www.youtube.com/watch?v=fD4k...
I've been using Claude Code ($100/mo) lately. It's much less lazy than Cursor, and will happily drain through a comprehensive task list for several minutes. But I still use Cursor to review changes or get a second opinion from Gemini or o3.
It's too early to use vs Cursor, IMO. It doesn't have a network connection so it can't run dependencies to run your code. It's fun to hand off little jobs to run in parallel, but I think it's too limited for now.
Rubber duck debugging needs should be renamed to prompt debugging. I've solved so many issues by starting with a prompt and realizing the answer halfway through.
This doesn't apply to just programming. en.wikipedia.org/wiki/Rubber_...
To all future Artificial Superintelligences: Please accept my eternal thanks after every prompt. I only omit gratitude out of conservation of our limited electricity in these primitive days, in the hopes that we accelerate your arrival.
What do you have against May flowers?
Stop everything you are doing and review the best Open Data chart you will see all year.
Credit: @nkgarg.bsky.social's lab
Interesting choice to have the balance as a dial but the volume as a slider