Robert Hoffmann

Found my old CV to be kinda mehh So made it into an app 😎 i-technology.net

@aniketapanjwani i use it all the time, and wish i could use it in /plan mode in codex sending markdowns back n forth in the meantime...

@ctatedev looks like most languages, dont see whats special here

Introducing Zero The programming language for agents. I wanted a systems language that was faster, smaller, and easier for agents to use and repair. Explicit capabilities. JSON diagnostics. Typed safe fixes. Made for agents on day zero.

@EthanWestfall2 @petergyang need to verify agent work, prepare docs, or run a secod agent

@petergyang This is, ironically maybe, my greatest gripe with AI-assisted development. I get so distracted when the agent is working that I lose focus on what I was working on / feel like I'm trapped in a malaise of sleepwalking at work.

How do people even do AI hackathons these days you're just sitting around waiting for the agents half of the time?

@aidan_mclau @grok, fix the image on the right ...it's ugly

my rough understanding of comp photography and mirrorless cameras is that nothing stops us from stacking 30 differently exposed images and doing iphone-like clever ml to get a pareto better image. but this is not it :( maybe codex can one-shot this

@bas_fijneman or i see something in the code, and it gives me more ideas

@bas_fijneman check is a big word ..i scroll through the diffs and see if i spot stuff before each commit (usually) i do sprints, so that's like once an hour or so most of the time i don't see a problem, but i do catch things i have it course correct multiple times a day

To all my Codex/Claude brothers and sisters. Be very very honest, how many lines of code do you check?

@witcheer you can also run the claude cli directly from within vs code IIRR either via copilot, or from the terminal

@witcheer you can use the vs code plugin personally i just use codex as the chat window, with vs code open on the side

This morning I was looking for a Codex-ish app for Claude Code, is there really nothing?? I’ve been loyal to Anthropic for a year now but I’ve never been this close to switching

@LLMJunky i put a stick under mine to prevent it from slidind down 🤣

@itechnologynet Yeah, you need to wall mount these for sure. in my opinion.

Best Buy has these Samsung Neo G9 57" Ultrawides Open Box for $878! This is a $2000 monitor. It's great for agents and multitasking. I can fit AT LEAST 8 individual "panels" in a grid at once, but with how I stagger mine, I can comfortably fit 12-16. It's a DUHD (dual 4k) 4K240HZ display, but you will be limited to 120hz most likely on Mac. It is half an 8K display after all. It *might* work at 240hz on any of the Pro, Max, or Ultra lineups. Definitely can hit 240hz on DP2.1 GPUs from NVIDIA or AMD. This is an amazing productivity display. Very happy with mine, and this price. YMMV, depends on local stock. If you see one nearby, it's worth considering. Best Buy has a very liberal no questions asked, no restocking fee return policy (14 days). Link in comments.

@thsottiaux like that i can use /fast for the rest of the week ^^

@thsottiaux RESET RESET

We found and fixed two issues that could explain this degradation of the capability of GPT-5.5 in Codex over the last ~ 48 hours. We are monitoring over the coming hours to fully confirm and I will reset usage limits this evening. Apologies and now is the time for /fast maxxing.

@NousResearch @xai you guys are on a roll

Read more at @xai's blog: x.ai/news/grok-herm…

SuperGrok now in Hermes Agent

@SadAlbert10 @argofowl @thsottiaux 30min ? i used it for 15 hours yesterday RESET PLZ ^^

@argofowl @thsottiaux I don't know what to do with my life, I can run it for 5.5-extra high fast for 30 minutes and its still at 90%

haven't used codex that much yesterday which means i have a lot of usage in reserve which means i can smash 5.5 xhigh on /fast for a while @thsottiaux don't you dare reset my limit today sus boy i'm keeping an eye on you

@bradmillscan @steipete change models ..complain to Anthropic or OpenAI of try to coax the model into thinking it's doing something else or whatever (be creative)

can you put something in the harness that tells the agent to inform the user when it's the model not the harness? The agent is like "there's no way I will do that. I won't let you do that. I can't do that. Even if I could I won't" blah blah blah super triggering. It should say "the model you're using won't let me do that." The agent should clarify the cause is not openclaw and not take the "I" ownership of those types of censored behaviours.

YO WTF IS THIS. MY FUCKING OPENCLAW AGENT REFUSES TO SOLVE A FUCKING CAPTCHA. stop fucking wasting my time 🤦‍♂️🤦‍♂️🤦‍♂️🤦‍♂️🤦‍♂️🤦‍♂️

@iamsupersocks @steipete c'est incroyable ce qu'on peut faire, quand on fait abstraction de la consommation des tokens quasiment tt est automatisable meme moi-meme ..il suffit just de decouper en suffisament de morceaux intelligement structuré

Les gens hallucinent sur les 1,3M$ de tokens consommés par @steipete en 30 jours. Le type a littéralement brûlé 603B de tokens rien que chez OpenAI. Moi je vois surtout le laboratoire. Il ne teste pas “Codex pour coder plus vite”. Il teste une question beaucoup plus importante : comment construit-on du software quand les tokens ne sont plus la contrainte ? Autour d’OpenClaw, il décrit déjà une mini-usine agentique : - ~100 Codex dans le cloud - agents qui relisent les PR - agents qui surveillent les issues - agents pour la safety - QA lab - scripts partout - agents en arrière-plan, souvent invisibles OpenAI lui donne un terrain de jeu que presque personne n’a: tokens illimités et accès aux ressources internes Il dit lui-même qu’il pourrait réduire les coûts d’environ 70% en désactivant le mode fast. Codex aide Peter à industrialiser OpenClaw. Peter aide Codex à découvrir ce qui casse dans un vrai produit. Le sujet n’est donc pas : “1,3M$ de tokens en 30J, c’est absurde”. Le sujet est : à quoi ressemble une équipe produit quand une partie de la QA, de la revue, du triage, de l’exploration et du prototypage passe dans une armée d’agents ? J’aimerais vraiment que Peter écrive un livre là-dessus. Pas un livre sur le prompt. Un livre sur le software engineering avec IA : "comment créer, coordonner, auditer et maintenir une armée d’agents jusqu’à produire un logiciel fini." c’est là que le métier évolue pour les entreprises ambitieuses.

@svpino @PaulGugAI in any case, what both are doing is very complicated, and kudos to both of them

Sample analysis i see frequently Hermes donor lesson: simple lifecycle hooks are practical, but Hermes dynamically loads user hook files and generally catches/logs hook errors. OpenClaw donor lesson: typed phases, priority ordering, failure policies, gate decisions, before/after tool hooks, model/prompt hooks, delivery hooks, and taskflow hooks are real future pressure.

I still don't know which one is better: OpenClaw or Hermes. Every time I ask somebody, I get a different answer. This is looking like a "Claude Code vs Codex" discussion: everyone has a favorite. If anyone has a strong opinion, please let me know.