One thing I like about GPT-5 is it telling me to verify functionality (with specifics) works instead of just assuming like Claude models do

GPT-5 is much more thorough with tool calling and trying to find the right answer than 4.1 and o3. promptslice.com/share/b-2ap_rf…

And just like that, Cursor is back on top. Cursor and GPT-5 work great together. The CLI dropping today is icing on the cake.

Goodnight sweet prince 🥲

The problem with this viewpoint is as Dario Amodei said recently, exponential growth doesn't look like exponential growth early in the curve.

At the core, agents are simple. From Cursor to ChatGPT agent mode to any other "AI" app you've used, they all start with a system prompt and set of tools.

Pro tip: instead of pasting links to docs for packages/gems, you can ask the agent to look at the installed source directly and it will explore the code just like your codebase to find what it needs

Bold move is the new absolutely right but hard to tell if it's sycophantic or condescending

OpenAI is learning an important lesson that you can't just release a new model like an iPhone now because so many people see ChatGPT as a friend (for better or worse), it's like replacing your best friends brain with a stranger

GPT-5 is slow because OpenAI is queuing standard requests now and you have to pay for the priority tier to jump the line. I haven't heard much talk about this but everyone assumes the model is just slower.

Opus 4.1: Fixed! Me: Nope, undo that and figure out the real issue. Let's debug properly and don't assume it's fixed until I confirm. Opus 4.1: (changes some more files) Now it's really fixed! Me: 😑

codepressive @levelsio I gave that shit up early into the Linux journey. Spite is a great kickstart, but it's a terrible way to power the rest of your life. You gotta learn for love.

I can’t stress enough how important it is to break large feature work into smaller chunks. The slop you introduce letting agents run wild for any meaningful amount of time is tremendous.

Don't be afraid to start fresh with what you've learned while building out a feature. I did it this morning and implemented the same functionality with 1/3 of the code before.

Build something that applies to you but also has a big enough market that makes sense. Get feedback fast, early, and often. Iterate and repeat.

I’ve been a Cursor power user since it launched and spent a small fortune on agentic coding this year alone. If you’re struggling to find a good workflow or have questions, reply below and i’ll see if I can help.

Claude: You said not to worry about RuboCop errors while writing code as they're automatically fixed but let me go ahead and fix the obvious ones like trailing new line and array bracket spacing. Me: Screams into the void

People visiting national parks drive me insane. See wildlife? All traffic rules we were just following are thrown out the window.

If you know David's story of winning Leadville and setting a CR last year, DNF'ing at Western States this year, and then winning and setting a CR again at Leadville this year, this should stir something within you. It's a beautiful story of love and redemption.