building stuff with AI, hacking, and def not vibe coding... 🇺🇦 taras.lol and building agent-swarm.dev Taking care of our baby @agent__swarm
Barcelona
- Tweets 798
- Following 622
- Followers 854
- Likes 4,354
Pinned Tweet
I'm 29.
AI founder from Ukraine, based in Barcelona.
Looking to connect with other hackers and people interested in multiplayer AI!
Hit me up ☺️
259
12
968
57,817
So Boris used formal verification techniques to improve their code at Anthropic.
We did the same using TLA+ in the swarm, and we found some issues in 2 of our core components!
Ah, and it also proposed simplifying a flow from 1.3M states to just below 5k.
All done by the swarm itself, pretty dope!
1
1
2
567
So Sonnet 5.5 just dropped, so we had to test it out in the agent-swarm, and we did.
Up to 2x faster and ~20% cheaper in some tasks.
Should you use it for the lead? No, use Opus 5.5.
Should you change your generalist agent? Sure!
Let's see what OpenAI has prepared for us today... but you better off using Opus 5.5 if you a sub, or Luna if you need something cheaper.
1
3
346
can someone explain why they keep doing this cinematic release videos that give 0 value and understanding over what is going on?
Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family.
It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.
103
For the curious, here's the side by side 🫣
A while ago @linear did a nice release video on loops, yesterday I sent it to our swarm and asked it to replicate it as accurately as it could.
Few hours later I got this… (re)introducing schedules, one of the multiples features of the swarm.
How? Gemini flash to scrape it and understanding it, then remotion with three.js to create the video.
Link to the blog the swarm wrote to copy our approach in 🧵
2
318
A while ago @linear did a nice release video on loops, yesterday I sent it to our swarm and asked it to replicate it as accurately as it could.
Few hours later I got this… (re)introducing schedules, one of the multiples features of the swarm.
How? Gemini flash to scrape it and understanding it, then remotion with three.js to create the video.
Link to the blog the swarm wrote to copy our approach in 🧵
1
1
8
11,772
Now it all became a few skills in the swarm, just said “skillify this”.
What should we copy next?
agent-swarm.dev/blog/linear-…
1
826
Yesterday I gave a talk on the memory system of the agent-swarm. And it was great to hear all the different approaches teams follow.
Our approach is kind of controversial tho, we use reputation! It’s a clever yet simple way of giving agents rate their own memories.
What happens then is a very similar approach as in RL training with rewards.
It does not matter if we have a ton of memories, as the good ones will stay.
If you are curious about it, ir disagree please hmu as I would love to chat about it :)
1
6
214
taras retweeted
@hackspain26 ha sido reflejo de lo que se viene con una organización totalmente AI-native.
Cada idea, por loca que fuera, implementada en minutos por @mrloldev , @disamdev y @dumenac .
Cero límites.
5
2
62
6,418
taras retweeted
the right answer is wife does whatever she wants with her life. work or not work.
21
1
173
3,833
the cooking at @hackspain26
CINTA - Class-agnostic INline Transport Analyzer
Check it out at hack-growth.dev
@jamipuchi
12
323