🇫🇷 Vibing (@ycombinator W24) 👉 https://xitter.cf/t.co/DmRjHrHYYy
Paris, France
Joined January 2020
- Tweets1.8K
- Following507
- Followers5.9K
- Likes2.6K
Media consumption/generation is changing.
Can’t wait to remake the season 8 of GOT.
How to get better at AI? Go on X and join the hype.
Take out your credit card, subscribe to Codex/CC for 200$ (if you can) and simply try to build everything you see.
90% is crap and fake results, but the 10% will teach you a lot.
We just open-sourced a product we've been using for the last 4 months, and it's wild!
GitHub for Skills. Here's what it does 👇
→ Build a skill once, share it with the whole company
→ Install it in any agent: Claude Code, Codex, OpenCode, any MCP client
→ Keep your own skills private, share the best ones
→ Version it, improve it, and everyone gets the update
Why we built it:
We kept creating great skills.
Then they stayed on one laptop.
Now we use Skillpack for everything:
→ Dev: PR reviews, shipping, planning
→ Marketing: articles, brand visuals, social posts
→ Finance & ops: quotes, invoices, contracts
It makes everyone as good as our best person at each task.
Take it, fork it, tell us what's broken.
Deepseek's harness is a banger!
And Deepseek 4.1 Flash makes me reconsider going all in on pay as you use models vs the subsidized Codex/Claude.
I've hammered 4.1 Flash for the last two days, I haven't spent 20$ yet.
Just crazy how well they implemented caching in their harness.
Astra is amazing for 3D printing.
I asked it to create a 3D render in Blender of Veebee, our mascot.
Then to optimize it for 3D printing.
24 hours later I have a physical version of Veebee
Github is down, may as well stop my agents... again.
Still, I love @github and I can't thank them enough for the free github actions for OSS. Saved me tens of thousand of dollars.
Good luck guys.
Billions of tokens are currently being burned by agents trying to figure out why their CI aren't working anymore.
Thanks @github
The last model from @AnthropicAI - Fable 5 - is BAD! Don't use it! Here is why?
It consumes your credits like a madmen - twice Opus 4.8
Yes, it is better, but it is so damn expensive.
For most tasks, it is not better than GPT 5.5 or Opus 4.8
I had to change the way i vibe code because of its price.
Now I ask Fable 5 to make the plan, and then hand-off the plan to Codex with GPT 5.5 to finish the task. ft @conductor_build. And still I'm maxing out my 200$ a month subscription with Fable, when Codex is doing all the implementation!!!!
GPT 5.6 Sol Ultra is expected to come out tomorrow, if it is half as good as advertised and not as expensive it will be the nail in the coffin for Claude.
WDYT ? Am I the only one ?
I’m out of Fable 5 credits again. Don’t know if I need to work today or just wait for GPT 5.6 Sol Ultra
Fable 5 is burning through my limit faster than 4.8.
I'm getting less done and I'm not even using it hard.
Fable 5 plans.
GPT-5.5 executes.
Single-model agent stacks are starting to feel obsolete.
@conductor_build is the cleanest way I’ve found to run that loop.
Skills and harnesses are the only thing that won't be replaced by new models.
Your product will be replaced at one point. Your internal processes won't.
If you are looking at building a product today, invest most of your time in skills or workflows.
