sebastian

founding eng. building the agentic OS for venture capital @

[sebastian @sebastianmxnt]

@Fremond_ peak

[sebastian @sebastianmxnt]

@andyyy best i've had is the Eagle's DNA in cambdrige @ the eagle

[sebastian @sebastianmxnt]

@viticci @ChatGPT too late already in grokbot

[sebastian @sebastianmxnt]

@travisk @JensenHuang

[sebastian @sebastianmxnt]

you're fucking kidding

[NVIDIA Newsroom @nvidianewsroom]

NVIDIA announces CEO transition. Read the release: https://t.co/MTTUvFHCJA

[sebastian @sebastianmxnt]

@catdaly what restaurant is this? looks great!

[sebastian @sebastianmxnt]

@I20Eric this is a vibe

[sebastian @sebastianmxnt]

i want more software like this. quirky, fun, completely useless.

[The Weather Channel @weatherchannel]

https://t.co/f3IHjpeyfo

[sebastian @sebastianmxnt]

@Currentreport1 https://t.co/YAhXQHf6be

[sebastian @sebastianmxnt]

I really like how you pointed out that serial time is the priority and not a secondary thought in between agent work. This is my key takeaway from your article. You can design a lot upfront let the agent go. Then, dedicate your best hours to ensure that the agent output is aligned with your expectation.

Re cognitive debt - I've been saying the same for a while. I have colleagues who can run 4+ agents and keep up with them and still produce code that they intuitively understand. I personally can manage with 2 agents and some background tasks.

Part of my architecture is also having a hypothesis of what the output should be and what modules are being changed. This helps me keep an up to date understand of the system.

[sebastian @sebastianmxnt]

humanity doesn't have enough compute.

[sebastian @sebastianmxnt]

@gdb the bottlenecks here will end up being manufacturer tolerances

[sebastian @sebastianmxnt]

@RobertRamspcxi

[sebastian @sebastianmxnt]

@geoffreylitt organic code? anyone?

[sebastian @sebastianmxnt]

i set up my fiancee with claude and they are vibe coding our wedding website. life is good

[sebastian @sebastianmxnt]

@santiviquez earthshot?

[sebastian @sebastianmxnt]

this is freaking insane. the applications of this will kill custom skills, system prompts, memories etc. you can have a second brain programmed with your experience and knowledge. deployment will be a scaling problem but each user can have their own LLM weight-programmed to fit their specific needs.

[Rikiya Takehi @rikiyatakehi]

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq

[sebastian @sebastianmxnt]

*insert jlaw "what do you mean" meme*

[Rikiya Takehi @rikiyatakehi]

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq