Vibe Coding and Cooking

Welcome back! Dear reader, I am tired this last week! I definitely hit a lull with doing AI related things. I want to take a moment and step back. This experiment is about two things. 1. Can AI replace our jobs with today's capabilities? 2. (This is the big existential question) Who am I? Who am I as a person, as an individual? Most of us I would say define ourselves by our jobs or our accomplishments in those jobs. "I am Joe, Senior lead architect with 20 years experience in my field and I have done X Y Z and I hold blah blah blah certificates and I have 5 recommendations on LinkedIn from previous employers." And I am not dogging on anyone who says those things or defines themselves that way. I more often than not in a situation where I am meeting someone will get asked "Tell me about yourself!" and I immediately come back with what I do for work. This isn't a bad answer or incorrect in some way because, part of who I am is what I do for work.

But I am also so much more than a Senior Systems Engineer/Architect. I am a father, brother, son, gardener, tabletop wanna be game player, believer, lover, gamer, Star Wars fan, lover of all things SciFi, and so on and so forth So last week I spent time doing those things instead of being an AI Builder of slop. If you don't know, we now call a lot of what has come out on GitHub or applications in the last few months "AI Slop" because well, it's very poorly put together. I had mentioned in my first post an add-on to Claude Code called TELOS that I use as part of Private AI Infrastructure from Daniel Miessler that I do plan on eventually writing some blog posts on, but these take that concept of who YOU are as an individual and making AI work around that. Who am I is part of my ABOUTME.md file in TELOS. So my AI knows, the goal is to have a completely autonomous home lab environment that eventually is self healing and self updating to keep me safe so I can do what is most important to me, which for this pirate is spending time with his family on great adventures! 🏴‍☠️

Okay! With that said, let's talk about what happened next in the home lab! My goal here is to catch you up to where I am today. Most of what I am about to talk about happened weeks ago, and much like AI, my context has since filled up and I am only remembering high level stuff! 🧠

So the first weekend was pretty mind blowing. The document repository I had Claude build was already full of great information, diagrams of what we had in place, and how things were configured. So with core infrastructure in place, I could start on my end user apps. Now these aren't really the exciting part because they're apps like Immich, Paperless, SearXNG... things you typically hear about from home labbers on YouTube. Why isn't this that exciting or that impressive? Well because it's all prebuilt apps with mostly great documentation on how they work. Most of them have years of dedication and code behind them and wonderful communities of people who have poured their heart and souls into them. So it's pretty easy for Claude Code or Gemini or Perplexity to build this for you because all it has to do is follow documentation! Hey look at that, much like a human would do! Of course much faster because well, LLMs read as fast as their GPUs will let them. ⚡

So for my friends reading this who are not "technical" or don't work in technology, you can do this! There is a world of open-source technology for you to be able to take advantage of. Keeping your data as your data, and let it live in your house! Now I very strongly recommend sticking with stuff built by humans for the time being. And anything that at this point has been under development longer than 4 months is probably done by humans. Remember my earlier comment about AI slop? Yeah, there is a lot of it out there. And a lot of it is very insecure. But let's take an app like Immich, what is it? It's a self hosted photo backup library you can point your cell phone at! 📱 What does that mean? Your phone can send photos nightly that you have taken to a computer running on YOUR network! This removes the crutch of Apple iCloud backups or Google Photos backups. Are you paying $10, $20, $30+ dollars a month because you have years of photos stored in the cloud or have you even lost photos because you didn't want to pay money and so your old photos are getting wiped out? Host it yourself! (Here comes prepper Blackbeard) What if one day the internet went down for a couple hours, days, weeks, months... forever? All the memories those photos represent are just gone. Now you might be thinking "Blackbeard, that is extremely unlikely!" No, it's not. Most of the infrastructure our internet uses is from the '80s!!!! The internet we know today runs on a backbone that was never designed to do what we are doing today. Should we talk about the power grid?!?! My brothers and sisters in Texas know what I am talking about! Can I hear from my SoCal people?! Rolling brownouts for the last 3 or 4 years during the summer... MUST... COME... UP... FOR... AIR!!!! 😤

These rabbit holes are everywhere... HAPPY EASTER!!! 🐰 Okay, we were talking to my non-technical friends about... Immich... right. So if you're thinking "There's still no way I can do this" just talk to an AI LLM about what you want to do. Here, I'll write the prompt for you:

"I want to host my own instance of Immich. I have about 5000 photos from 5 generations of iPhones. I have a budget of $800 to go out and spend on a desktop that I will plug in to my router so I can start backing up my photos to Immich. Here's what I am thinking for a plan

1. Research the best desktop I can buy for $800 dollars
2. Buy the desktop
3. Install Immich
4. Install the Immich app on my phone

I'm not a technical person, so I am not 100% sure what I am looking for or what kind of desktop is compatible with Immich. Can you help me understand better what I need to do for each step and provide me guidance along the way?"

I just dumped that into ChatGPT and nothing it said was wrong! It gave me 3 options for desktops (good, better, best), I even said "My family is mostly Mac users" and it gave me options for Macs I can buy. Then you just go through step by step what you are doing and it will guide you! 🗺️

Now what does this tell you? If you are a technically inclined person, you already know what you need and what you want so you just have an agentic coder do the deployment for you. If you feel like you are not a technical person, you chat with an LLM and tell it what you want and where you are at in the process. But what does that mean?

Well friends, it means the knowledge gap is shrinking. No longer do you need to be an expert in a field or educated in a subject to know how to do something. For years we have put what we have learned in text. Whether it was a book, blog, white paper, research paper, news article, whatever. We have put our knowledge into words on paper or digitally and guess what, AI LLMs have been trained on it! That information is how chat LLMs know how to tell you to install Immich on your own hardware. 💡

So armed knowing that I don't need all the knowledge to build something, I started building! Now I may have mentioned this already but let's dive into another Claude Code built application. I had been seeing online people with dashboards for their agentic coders. At this point OpenClaw was spreading like wildfire. YouTube and every social media platform I am on was showing me people running OpenClaw and these dashboards of agents doing stuff. I then saw a YouTube video from NetworkChuck with his Claude Code monitor(1). So what did I do? I took a screenshot of it of course! I then wrote my prompt to tell Claude Code I wanted a web page to be able to monitor what was happening with Claude Code, hosted from my remote server and to build it as a Python web server like the Doc Site we had built. I gave it the screenshot I took (Ya, I stole the idea... and... I'm a pirate! Remember) and it started building. And in probably 10 minutes, I had my Claude Code monitor. 🏴‍☠️ Now I did some tweaks and added a couple things like the Ollama monitor to monitor my at home LLMs, but for the most part I had what I wanted in no time at all. So if you want your own, guess what you can do, download my screenshot and tell Claude Code to build it for you.

Then I decided, why stop there! I also use SaltStack Community Edition in my home lab but it has no GUI in the community edition. So I built one! I'm not going to show you that one, there's sensitive data there and it's also not that exciting, especially if you think SaltStack is a seasoning for food. About a decade ago technology went through this really weird kitchen phase. We basically started naming stuff after what happened in a kitchen. Like Salt, Chef, this avocado VPN thing... it was a thing. Chef had recipes in cookbooks for it... You think what we are doing now is confusing haha, you should have seen us 10 years ago! 😂

At this point I now had 6 applications I had built that are my own. They do exactly what I want, how I need them to do it, and they look like my apps. I'm not paying anyone to host them or paying subscriptions to someone who has built half of what I want or extra features I'll never use. They are my own. Of course, now I am the one that has to maintain them but I'm okay with that, because technically I just ask Claude Code to update them! 🛠️

Let's wrap this up. First, saying you are not technical or lack the knowledge in something is no longer a blocker. If you don't like reading like me, check out Google's NotebookLM, it takes text based information and creates podcasts for you! It's amazing and a great way to learn on the go. Use AI to try and do something you never thought you'd be able to do. Second, to my vibe coders, you are doing amazing work! Keep going! We now live in a world where anyone can build applications for whatever niche you are in or apps specific to your business, and it's wonderful! BUT, I have read quite a few articles and heard quite a few stories of applications that have been vibe coded and how extremely vulnerable they are (2). Even people losing tons of money because malicious folks were able to give themselves unlimited tokens in AI applications and run like crazy with it. You MUST think security, you HAVE to make security part of your pipeline. If you don't, people will find your vulnerabilities and use them against you. It's not a question of if, but when. And guess what, it'll be soon. Malicious actors are using AI too, and they can find vulnerable applications faster than ever. 🔒 When you have created something amazing and you want the world to see it, stop. Ask your agentic agent, Is this vulnerable? If so where? How do we store usernames and passwords? Could we do better? Only YOU are responsible for what happens to your users. And if you get hacked, you're not going to be selling whatever you have for long. ☠️

(1) https://www.youtube.com/watch?v=s96JeuuwLzc&t=405s
(2) https://www.kaspersky.com/blog/vibe-coding-2025-risks/54584/