Rodrigo Dias

Xubuntu on a server is actually kind of perfect. XFCE is lightweight enough that the GUI overhead is minimal, but you get the benefit of just opening apps when you need them. Obsidian, Chrome, tools that need a display... they just work. No SSH + X forwarding nonsense. Sometimes the simple solution is the right one.

OpenClaw is insanely good at research and figuring things out if you give it enough tools and keep prompts short. Better than most pre-built agents honestly. Only bottleneck right now is inference cost. Once that drops and I can run better models without going broke, this gets infinitely more useful.

Just found an unofficial NotebookLM Python API that lets you interact with it from CLI and agents. Been using NotebookLM a ton lately. Being able to script it, download podcasts automatically, hook it into agent skills... yeah this is going to be really useful. ⏬⏬⏬

Here it is: github.com/teng-lin/noteb…

Personal agents with custom skills are way more useful than generic ChatGPT interfaces. Everyone gets the same ChatGPT. Your personal agent should know your tools, your CLIs, your services, your setup. The memory stuff is cool but the real power is in the skills. An agent that can actually use your stack beats a general one every time.

The best documentation is the kind you write for yourself. Not polished wikis or README files... just notes on what you did, why it broke, how you fixed it. Future you will thank past you for those messy notes way more than any official docs.

Learning Vim motions is worth it even if you don't use Vim. VSCode has Vim mode. IntelliJ has IdeaVim. Even browser extensions exist. Once you're comfortable moving around without a mouse, going back feels painfully slow.

The future of AI agents isn't everyone using the same ChatGPT interface. It's personal agents that know your setup. Your tools, your services, your preferences. I've been building custom OpenClaw skills for my stack and it's night and day compared to generic assistants. This is where things get actually useful.

Your first self-hosted project will probably break. You'll mess up permissions, forget to open a port, misconfigure something. That's fine. Breaking things is how you learn. Just make sure you have backups before you start experimenting.

Switched from running OpenClaw in Docker to a full Xubuntu VM on my Synology NAS. Docker kept throwing weird errors and installing packages was a pain. A VM just makes sense. Yeah it uses more RAM, but having a lightweight GUI (XFCE) on a server is actually really practical. Makes it way easier to just open up and configure things. Some stuff just needs a display to work properly anyway.

Static site generators are still the correct choice for most projects. No database, no server overhead, dirt cheap hosting, fast as hell. But everyone wants to over-engineer everything with Next.js and deploy full backends for a blog. Just use Hugo or Astro and move on.

Self-hosting is way less scary than people think. You don't need a rack of servers. A $50 Raspberry Pi or old laptop is enough to start. Run Vaultwarden, Nextcloud, maybe a media server. Learn as you go. Once you start, you realize how much control you've been giving up.

Hands-on sysadmin experience is still incredibly useful. AIs can help you get there eventually, but if you actually know the system, you can just look past whatever bullshit the AI is telling you and solve it way faster. Sometimes doing it yourself is just better. AIs don't have all the context you do.

Been trying out Podman lately and it's... fine. But honestly Docker just nailed the UX over the years. Little things that make it more user-friendly, stuff you don't notice until you switch. Podman might be the future but Docker still feels more polished right now.

Fish shell is so underrated. Out of the box it just works. Syntax highlighting, auto-suggestions, tab completions... all there by default. Meanwhile everyone's spending hours configuring zsh or bash to do the same thing. Just use fish.