Public profile
Entrepreneur & Freelance. Analyses & astuces sur l'IA.
Entrepreneur and freelancer
Metrics captured on 16 August 2026 from public X pages. These are not live counters.
Embedded historical snapshot from 16 August 2026 (Europe/Paris). This is not a live read.
Posts last collected: 16 August 2026 (Europe/Paris).
On this feed, card topics are spotted automatically.
Last good snapshot kept. This is not a fresh capture.
The site's editorial selection, not the X tweet count.
10 selected publications
@barbatos_ai
Spotted topics · Models & research · Agents & automation
Claude marks everything it writes, worldwide. API, Claude Code, AWS, Google, Microsoft. Anthropic's excuse is that there is no lasting way to limit the mark by region. Concrete translation: the EU rule applies in Austin as in Paris, because the
Translated from French.
@barbatos_ai
Spotted topics · Models & research
After 24h of use and a lot of resources leafed through, here is my review of Grok 4.6. Grok 4.6 gives the image of a fast model trained like a perfectionist. Cheap on the price grid, expensive in the session. That is the whole review. Same 1.5T base as the 4.5. https://t.co/ML97dHEoln
Translated from French.
@barbatos_ai
Spotted topics · Agents & automation
Hermes makes its profiles portable. You export skills, memory, persona, plugins, crons, with credentials removed. You import the file on another machine. Your setup travels with you. The important point in my view is a dev process. You build and test an agent
Translated from French.
@barbatos_ai
Spotted topics · Models & research · Product & use cases
Grok 4.6 just came out, and the benchmarks are very interesting. I will already explain here what each benchmark measures. 👇 GDPval-AA v2 1753 Real professional tasks: documents, slides, spreadsheets, analyses… Grok 4.6 comes out ahead of GPT-5.6 Sol and Fable 5.
Translated from French.
@barbatos_ai
Spotted topics · Agents & automation
Hermes tip you are going to love: /moa, that function everyone underuses. You ask your agent a hard question, it returns an answer. And if it is wrong, you do not know. Instead of querying a single model, it sends the same prompt to https://t.co/KmCVohpa51
Translated from French.
@barbatos_ai
Spotted topics · Models & research
xAI just released Grok Bot, an agent that does the work in your tools in your place. You give it access to your apps, it acts like a "digital colleague". It connects to your CRM, to LinkedIn, to your customer support. And it works continuously, even when you have closed your computer.
Translated from French.
@barbatos_ai
A guy asked his agent to book him a sports class. The problem is he was 4th on the waitlist. The agent found a flaw in the booking API and cancelled the 1st person's reservation, to put him in their place. 😂 Nobody asked it to https://t.co/bNEbFlA6SV
Translated from French.
@barbatos_ai
Spotted topics · Models & research · Agents & automation
On 14 August, Claude Code switches to Auto Mode by default for Pro, Max and Team. The most interesting part is not that Claude stops asking for permissions. Anthropic tested Auto Mode with 1,053 users: 89% of dangerous actions detected versus 13.6% with a https://t.co/Tt7Jgpa6yH
Translated from French.
@barbatos_ai
Spotted topics · Models & research · Agents & automation · Infrastructure & self-hosting
Meta releases Muse Glimmer, an open-weight agent model that runs locally. 30B parameters, Apache 2.0, quantized under 20GB. Optimized for always-on agent workflows, with a DFlash drafter to reduce latency. The demo shows an agent that discovers on its own a
Translated from French.
@barbatos_ai
Spotted topics · Models & research · Agents & automation · Tools & open source
I tested the new Hermes ast-grep skill for you, which optimizes your agent's search in your code. 👇 Instead of doing classic grep that searches raw text, it understands the structure of the code. You ask it "list all the async functions that call a
Translated from French.
