Is anyone else obsessed with TLA+? I don't even need it but it such much fun

If the AI race was F1 then Google is Ferrari

I don't know what it is, but I just enjoy Claude Code TUI experience so much. It's slow, buggy and codex is faster and more precise at times. But I always have more fun with Claude Code

Goddamit Anthropic, down again :(

I'll give you this one for free. Instead of porting the entire codebase from the current state. Port from the very first commit to the language/architecture you want. Then have a agent integrating into trunk. Thank me later

Okay fine, I'm not mad about the downtime anymore

When claude starts moving fast I know that it's not Opus 4.6 anymore. All it takes is the /model and voila... Sonnet!

Giving Codex/Claude a skill that makes it reflect on gaps in the though process that cause unnecessary back and forth it's such a life hack. In this cause codex pointed out the right thing, it assumed the code worked, rather than looking at what actually was rendered.

I've been building on Chris Tate idea of the agent-browser BUT for iOS native development. It's been a long fight to tap into apple's obscure APIs but I've hit the first milestone, extract ANY app's hierarchy directly from the system to inspect + give feedback to the agent. I'm

Since claude got the 1 M context window it got so good. It's able to do objdumps, read the assembly, make changes to the contract, and continue for an hour until it finds the bug. I tried codex and the 200k context window now feels tiny, it just can't compete

My prediction is that Anthropic and OpenAI will have a subscription tier for autonomous workflows. Bookmark this.

Codex Plugin for iOS is incredibly good

I've managed to get the codex security experience in 2 skills: npx skills add pproenca/dot-skills --skill threat-model npx skills add pproenca/dot-skills --skill threat-patch It's incredibly good at finding vulnerabilities and classifying them!

+20 minutes wait for a API response Claude :(

Codex is also becoming really good at staying on track during long running complex projects. It figured out how to check whether long running process was blocked or not by running ‘sample’ and other tools

Clearly OpenAI is going to release a new model soon. Either I am reading too much into the differences I see in Codex's coding skills. Or I am right about it. Let' see

Skill maintained by Pedro Proença. Thanks Pedro!

Codex 5.5 is really good at execution, but Opus 4.7 is a beast at producing skills that make Codex even better at specialist tasks. Codex just can’t achieve that