Been playing with ArtCraft apps heavily the past week. just realized a huge reason why these apps are likely to WIN over closed software: The CLI/MCP native design. It enables a training data flywheel thats hard to get on closed software...
Most agent skills never get tested on a fourth run.
My rule: build the skill from three real runs of the work. Pull out the decisions. Write the output contract before the steps.
Then give it fresh work with no hints. If it only works when you steer it, it's not reusable yet.
Fable 5 is the GOAT, but using it for a hotfix feels like waiting for Gandalf to finish thinking while the building burns.
GPT-5.6 Sol in Codex fast mode gets moving immediately. That speed keeps you inside the problem when every minute matters.
@thsottiaux removing the 5hr cap = raw power 🤓 I used my whole cap in one day. Finished porting the OpenAI Privacy Filter (1.5B MoE model) to run natively on Apple Neural Engine (ANE)... no GPU, just neural compute. Pretty cool.
I built YOLO26-ANE for the webAI YOLO26 MLX Build Challenge. I asked: what happens if YOLO26 runs on Apple Neural Engine instead of MLX?
On my M4 Pro:
webAI MLX/Metal: 155 FPS
CoreML/ANE: ~448 FPS
ANE sees.
GPU thinks.
#YOLOMLX@webAI
This weekend's @aitxcommunity / @OpenAI Codex hackathon is producing goldddd. Codex + 5.5 seriously cooks, big shout out for letting us hackers push it to the limit. I'm putting in 2 submissions for tomorrow's hack fair, and filing a patent. 👨🏼🔬
@AmericanAir has us sitting on a 90 degree plane with no A/C for 30 minutes prior to take offc and no A/C for the flight?! Absolute shit show flying American this holiday, 24hr delays on the first leg and delays & dysfunctional planes on the return trip. The worst.
A high-resolution, stylized photoshoot of a Lamborghini made entirely out of linguine pasta. Set it in a sleek, moody environment with a dark background to highlight the golden-yellow texture of the linguine. The name “LAMBORGUINI” should be visibly labeled on the side.
🏁 𝑺𝑷𝑬𝑬𝑫 𝑾𝑰𝑵𝑺: Recently I prototyped 30+ complete versions of my platform's landing page and app UI in under an hour. I'm using Lovable differently than most people.
��
And because of my downstream AI workflows, I was able to scrape my own webpage builds and craft a meta-model of positioning & messaging based on different ICPs that will improve my next round of platform iterations built with AI.