I moved all my dev work to a Hetzner VPS and run it through T3 Code. Should've done this months ago. My MacBook used to be the "server" so I had to keep it awake nonstop with caffeinate. Now everything runs on the VPS. I can start a task in T3, check on it from my phone and reply to the agents running from wherever I am. My Hermes Agent can also start new threads in T3 Code and manage them for me at anytime. I just love this workflow.

Nous released their own benchmark for models running inside Hermes Agent. Current ranking: 🥇 Claude Opus 5.5 🥈 GPT-6 Astra 🥉 Claude Sonnet 5.5 This matches my experience. I've completely switched to Opus 5.5 and it's been so good. It's incredibly fast and it handles facts and small details much better than Astra, and also it seems like I have unlimited usage... portal.nousresearch.com/benc…

Astro + Cloudflare is all you need.

Had to try something out with @dhh's @OmarchyLinux

Everyone is talking about how good Claude Opus 5.5 is at motion design. So I tested it myself, and I get it now 🤯 Here's an example I made for Grok @Bot. Everything on this video was one-shotted and generated with code, including sound. It took Opus ~30 min with fast mode off. Prompt used: "Make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. Go all out. Your goal: promote Grok Bot."

I used Opus 5.5 heavily all day yesterday and barely made a dent in usage. Really impressed. Claude is still undefeated for visual design.

So this is where we stand... Forked HermesOne, removed all useless features, and now I have a clean app to talk to my Hermes Agent.

I don't why people are hating on T3 Code... I've been using it quite a lot lately (even switched to the Nightly version) and I really enjoy it.

The personal software era is here. I've lost count of how many tools I've built with AI for exactly 1 user: me. Tools for work. Tools for my personal life. Each one built around how I actually work, without features I'll never use. No monthly subscription for every tool. If something is missing, I can add it instead of waiting for someone else's roadmap. I'm not going to rebuild every app I pay for. Some are genuinely worth the money and hard to replicate... But the cost of building software is dropping really fast. Anyone can now build pretty much anything.

Claude: "I estimate this will take 1-2 weeks"

This is what an inbox should look like.

You'd be surprised how easy it is now with AI to rebuild a WordPress site from scratch with Next.js and Tailwind, or whatever stack you prefer. Give Claude Code or Codex the URL and clear instructions on the structure and components. SSH access helps it inspect the existing theme files directly, but it can still get them without it anyway. Let Opus/Fable 5 handle the implementation then have GPT-5.6 Sol review the code and run a full maintainability, security, and performance audit. You end up with a fully static, version-controlled site that runs on a tiny VPS without WordPress and score 100 on Lighthouse without much effort. Set up Dependabot to open PRs for vulnerable dependencies. Have Hermes or your favorite AI agent review them and merge the safe fixes. Need to update a section? Add a new blog post? Send it to your agent, have it open a PR, review, merge. I've been using WordPress for years and still do every single day as we maintain 200+ WordPress sites at our agency, but this workflow is hard to beat... We live in a new era.

I guess the benefit of being Google is that you can make Gemini rank #1 for "AI" searches.

We're getting closer to 2027, which means Covid happened almost 7 years ago. Where did time go? I'd like a refund.

Grok Bot is really good... Like really good. As a heavy Hermes Agent user, I'm impressed. You install it and can immediately start building your own agents, give them access to tools, and let your bot create other bots and orchestrate them. It also seems to include a computer with 120 GB of storage, browser access, etc. All your agents work in that shared environment. There's no configuration or setting up this and that... It just works out of the box. If Hermes Agent had this kind of app experience and onboarding, it would be golden. I'd probably switch to Grok Bot, but not being able to choose the model or truly own my data is a dealbreaker for me. I still prefer my Hermes setup. I'll keep testing it though. They really cooked on this one.

One instruction I'm testing with coding agents is to follow the KISS methodology: - Strictly follow KISS (Keep It Simple, Stupid) - Choose the simplest solution that fully solves the problem - Avoid premature abstractions - Don't handle every theoretical edge case - When in doubt, ask yourself: "Is this as simple as it can be?" Some AI models tend to over-engineer tasks way too much and so far this seems to help.

Decided to close my laptop for a break, then Grok 4.6 and DeepSeek V4 Pro dropped right after. Wen break?

All I need for Hermes Agent to achieve perfection is an app like Codex that works and syncs beautifully well across desktop and mobile.

Interesting: Claude models released from August 2 onward will now embed a hidden watermark in the text they generate. I didn't even know this was thing. The change comes from the EU AI Act, but Anthropic is rolling it out worldwide across Claude, API, Claude Code, Cowork and other products. You won't see anything unusual in the output, the watermark is built into the wording itself, stays with the text when copied, and might survive minor edits. They also plan to release a tool that lets others detect whether text contains the watermark. That means even if you write something yourself and only ask Claude to fix the grammar, the edited version could still end up watermarked.