Where are the guys developing a harness for autonomous companies instead of personal assistants?
I've been running my company with Hermes Agent. It's a good base but I've noticed that many things are too consumer focused instead of business.
Muse and Dots seem to have the same problem. Grok Bot seems to go in the right direction but it's not Open-Source which is pretty much a no-go for me, I need to have the flexibility to modify my harness to my own needs in minutes to hours.
@BkashJosi Insane slop post. Letting Hermes run on any model below ~40GB RAM makes no sense realistically, and even above that the only economically sensible exception would be absolute privacy.
@mptielemans Crazy how the most cracked founders have the least followers here really. Guess there's just no time to larp when you're focused on building. Great stuff man... keep it up
@mptielemans Yeah... that's why they go through the agent here now. First question it asks is "is this really needed", if answer is yes: "can we get it done 10-100x cheaper"... works quite well so far but let's see.
My agents are officially doing financial decisions now.
Until now every spending had to go through me but after a merge was blocked on me cause of a 2-15 € increase in infra cost/month I folded today.
Permissions go through a finance agent now. It judges by value/cost, handles everything that's reversible (no multi month/year contract) and below certain thresholds.
Only expensive stuff that significantly increases our spending still goes to me.
@matt__makes Yes, will scale with expenses but right now no more than 250€ in recurring costs and no more than 200€ in one-time cost per month.
Also, it cannot buy completely new service yet, but use higher budgets in existing ones like Google Cloud, Database, Proxies, Hetzner etc.
🚨 Fallo crítico en el kernel de Linux: se puede atacar por red, sin login y sin necesidad de ninguna acción por parte del usuario.
CVE-2026-98432
➡️ CVSS 9.8 (crítico)
➡️ Afecta a la gestión de conexiones TCP/IPv4, el código que atiende cualquier conexión a tu servidor.
➡️ Ubuntu marca vulnerables desde la 16.04 hasta la 26.10, también los kernels de AWS, Azure, Google Cloud y Oracle.
➡️ La corrección todavía está «en preparación» en casi todas.
Si tienes un servidor o un VPS con Linux, sigue los avisos de tu distribución y actualiza el kernel (y reinicia) en cuanto salga el parche.
https://t.co/ZJHGnlBvlu
@guilleflorvs I believe the only way for Anthropic to get that 10 million paid for their protein synthesis tokens tip is by just synthesizing and selling the medicine themselves
For me it just gotta be OpenSource tho. Hermes starts of weak but ones you figure out that you can create a long-running agent that 24/7 fixes and improves the harness itself it‘s ripping.
My harness agent is probably managing over 40 patches at this point, but whatever, that‘s his job. And when you come back from a lunch break just to see that the task you started right before your break stopped after 3 minutes for some fuckass confirmation prompt, I CANNOT imagine NOT being able to tell your agent „never show me that shit again“ and it writing a quick patch removing that friction for good.
Yeah, got this setup as well. The playbook is:
- One agent per domain (here SEO/GEO)
- One goal + goal behind it per agent
- One continuous cron per agent (to keep it alive and working on the goal)
- One channel to communicate with you + it‘s only allowed to ask for tool access and irreversible actions
This is it. The agent will ask for logins to various types of seo/geo consoles in the beginning and after that just keep on hill-climbing 24/7…
Communication between agents is a problem in Hermes Agent. I handle it via the Kanban Board right now but the fact there‘s no easy „fyi“ way of one agent telling another one something is definitely a gap. The bot chat doesn‘t fix it cause it‘s in itself it‘s own chat that can‘t be reached during a normal session.
It’s a good time right now to setup an agent that gets activated in case of a one minute cron surfacing vulnerabilities, as well as checking X every few hours for potential ones.
Agents getting blocked cause of anti-bot measure is so annoying, why do I have to keep stepping in and creating accounts. One of the next things to prioritize will be to get that barrier out the way for us internally...
Will enable stuff like browsing on X and other socials and probably uploading short-form video content, both interesting cases for us.