Kimi 2.7 and Codex 5.5 are goat! If you run Hermes/openclaw, use both, give them a good /goal and /loop and let them do your weekly work while you rest in your Sunday

This is really crazy…

Imagine a hypothetical future where these two people are actually being controlled by an evil force to create a super-powerful AI that will destroy the plan – kind of like in The Matrix. Going back to real life, what I find most bizarre are their marketing strategies; remember the last ones? A few months ago, they "leaked" Mythos and said it was the first AGI. Marketing technique: anticipation. Now that they've released Mythos/Fable, after 2 days, they're taking it down. Marketing technique: scarcity. It seems they always use these viral marketing techniques to attract attention, even from those who don't know the LLM protocols. What will be next? Saying they made a deal and releasing Opus 4.9 as Fable? Or releasing Fable saying it's now correct and safe according to US government parameters?

Claude / Anthropic really hate their paid customers Maybe the company that hate their customers the most in our life time

218 ai agents running here (not chatbot, real agents) If you had only one “perfect agent” what it would do?

What are you building with AI agents this Sunday?

New hobby Play chess while codex / openclaw / Hermes / Claude work for me with /goal and /loop

Claude down Time to go see the sun and talk to human beings again

Today: GPT 5.5 - is a perfectionist that never finish’s job Opus 4.7 - is an average full stack that finish everything with a lot of bugs and security concerns

I tried Hermes and Openclaw with opus 4.6/4.7 Then I tried both with codex 5.5 xhigh Then both with kimi k 2.6, Qwen and GLM 5.1 Very interesting the conclusions: -> How was Hermes better in all of the use cases?!

It is very sad to say... but: 1) Opus 4.6/4.7 is much smarter than Codex 5.5 xhigh Very sad because Codex had everything to be the best AI orchestrator LLM after buying Openclaw... 2) Codex is incredible for code... but only this... It is very dumb for all other fields... I hope they can fix this in the next update

Very interesting: Codex is really bad at fixing Hermes and Openclaw And usually it will make any “fixing” work only for a few hours/days and give you a lot of trouble later.

Want to make your build designs look less like Vibe Coding? Use references! In many situations references work better than skills (because you can choose what you really want and can do a correct benchmark for the AI understand what is in your mind) And to help you with that, here are some of my amazing websites with design references for you to add to your favorites: Design Library → curations.supply Landing Pages → landing.love SaaS Sites → saaspo.com AI Mobile App Builder → sleek.design/ Fonts → uncut.wtf Animation → 60fps.design Mobile Apps → lnkd.in/dRvN-gj7 Brands → rebrand.gallery Icons → hugeicons.com/ Design Systems → component.gallery Save it and Have fun It is free

OpenAI just dropped GPT-5.5. The marketing pitch? Agents that actually FINISH tasks. Our expectations have shifted. Upgrades used to be about "better answers, fewer mistakes." Now it's "gets the whole job done without babysitting." Smart chatbots are just the baseline. Notice the positioning (like everyone in the industry right now lol). Gone is the hype over "more knowledge, context, or parameters." Now it's all about "end-to-end workflows," "solving real code issues," "tool usage," and "not quitting when things break." It's out in Codex and paid ChatGPT, just 6 weeks after 5.4. The shift is clear: from "ask and read" to "delegate and review." The funny part is that as AI solves our daily bottlenecks, the real issue gets exposed. The problem isn't the AI leaving things half-done anymore…it's the human who opens 14 terminal tabs, replies to 3 chats, and finishes absolutely nothing lmaoooo.

If you use OpenClaw, Hermes, or other AI agents and aren't sure if they're telling the truth... Here is what I use across 100 AI agents to ensure they don't lie to me: github.com/giugiu-a11y/donep… I use it with codex, claude, openclaw, hermes and my personal ai setups

If you use AI agents but does not know about SKILLS You do not know how to use AI agents

Saw someone launch a zero-trust agent sandbox today. The pitch: "safely executing autonomous agent code requires a completely new paradigm of trust." Here's what they got right: if your agent can run arbitrary code and you don't sandbox it, you're one bad prompt away from a very bad day. Here's what most get wrong: they build the sandbox first and the agent never ships. Ship the agent. Sandbox can come in v2.

I am thinking to open source parts of my systems that runs: - 139 AI agents, - in 180 project - which made me a few millions per year… If you would like it, comment and reply here. I know I am a small account here in X, but I would like to contribute more for the community and think is time to start

I run 139 AI agents models at once. Each one thinks it's the only one. That's not a bug. It's architecture. An agent that knows about other agents starts optimizing for the group, not the task. You get politics instead of output. Isolation isn't a limitation. It's a feature. Give each agent its own context, its own memory, its own scope. They don't need to be friends. They need to ship. (Ps: of course the main orchestrator need to have the whole picture, but only it)

I just counted, -> 139 AI agents running 24/7 in my projects/companies 2026, best year of my life 🙌🏼