Hey there!

Summary of this email:

• Tailwind CSS: Architecture and best practices course 100% published
• How Clawdbot almost wrecked my house 😅

⏱️ Estimated reading time: about a minute.

⛵ Tailwind CSS: Architecture and best practices course 100% published

In the AI era we're living through, Tailwind CSS has become a foundational element: If you ask any agent to build you a website without specifying anything, it's going to use Tailwind.

But there's a difference between starting to style and having styles that last over time and allow changes without breaking other things. That difference is architecture.

Just as we know that having a good foundation in code lets everything scale, the same applies to styles. It happened with CSS, which is why patterns like BEM and Atomic were invented, and it happens with Tailwind.

That's why we've created a course on how to apply these best practices to build a solid architecture with Tailwind. Now 100% published on Codely Pro Premium. 😊

😅 How Clawdbot almost wrecked my house

If you opened Twitter, Reddit, TikTok, or any social media this weekend, your feed was probably flooded with content about Clawdbot.

What is Clawdbot?

Clawdbot, now renamed to Moltbot since the name was too similar to Claude and Anthropic wasn't happy about it, is "your personal AI assistant".

You install it. At that point you choose which agent you want to use: Claude, Codex, another via API... and which medium you want to communicate with it through: Telegram, WhatsApp... Also at that point you give it access to control your computer.

That way you can ask it via Telegram to delete all files from your downloads folder or to send an email.

It has a huge advantage: it "learns".

In simplified terms, every X minutes it analyzes the conversations that have taken place and saves the learnings to json files, so it can be more fine-tuned in future interactions.

But it also has a huge disadvantage: security. You're giving total control of your PC to an agent that can really mess things up.

In my case, because I can't resist seeing something and not installing it (😅), I said "let's play with it". That said, I installed it on a mini PC I have, so I wouldn't risk corrupting my Mac and not being able to work the following Monday.

On that mini PC I have a Proxmox setup with Home Assistant, Z2MQTT, AdGuard... my plan was to control the house from Telegram.

But I didn't create a container to install it inside Proxmox -- I installed it alongside it.

You can probably imagine what comes next, but I'll tell you anyway.

I asked it "optimize Proxmox to consume less resources". And boy, I shouldn't have asked that!

After the bot spent a while modifying configuration files, the mini PC restarted. A restart consists of a shutdown followed by a power-on. As for shutting down, the mini PC did shut down, but then it wouldn't turn back on. It... wouldn't... turn on... 😞

If you're not familiar with Home Assistant, imagine Apple Home, Alexa, or Google Home. Imagine all of that getting wiped. You have to re-pair all the devices (some behind light switch panels), name them, create all the automations, dashboards, install the plugins...

After years of having it and gradually tweaking it, I estimated I'd need to spend at least 30 to 40 hours to get everything back to how it was. FORTY HOURS. A full work week.

Hours that I could be spending on something else (sleeping?), now I have to spend setting up something I already had and all because I wanted to play with a shiny new toy.

Things could have gone much worse, but luckily I have backups and I ran the experiment knowing I had that safety net, otherwise I wouldn't have done it. That said, I had never actually tested restoring a container backup until this moment and fortunately everything worked fine.

Moltbot is a very useful and interesting tool, the idea of having a personal assistant is where all AI companies are heading.

We'll probably see hardware (AirPods and similar) focused on this, but in the early stage we're in, if we want to play with these tools it's very important to control what we give access to and always have safety nets like backups.

But this was the hype of the week, the previous one was the Ralph loop, before that Skills, before that Claude Code for everything... I miss the days when every month a new JS framework came out that was going to change everything!

Sometimes arriving late is better than arriving early. With most AI topics, this is the case. If you arrive late, you give time for best practices to develop for using these kinds of tools:

• What permissions and where to give Moltbot access.
• Why it's better to use the Ralph loop from bash than using the Claude Code plugin.
• When to use Skills vs when to use rules.

But for that, there have to be people who arrive early and bump into all these issues. That's what we enjoy doing at Codely. We're a bit masochistic, yes.

That's why we recommend watching Café con Codely every Friday at 9 CET, where we share the week's news and how we used them (and messed up 😅). Live on our YouTube and Twitch.

And since you've made it to this part of the newsletter, here's the joke of the week, which I know you were waiting for:

> Where does an API go to eat? To the RESTaurant! 😂 😂 😂

Cheers!