2025
A Better Programmer
I wish I was a better programmer, this way AI can’t pull shit on me. I have caught it doing stupid stuff. I am still trying to sort out the issues from last night. Had an internal admin page where I can monitor the site, now broken. I should have reverted back to the checkpoint that i created before migrating the web app :(. But I stuck it out, trying to get the admin page to work again.
Freaking Claude!
So I setup an upgraded version of AI Personality Shift, a neat little web app that changes with different words:
Got it working in dev, it was a pain but not a big deal. Setup a new chat because I know Claude would have a message length issue. Setup the prompt, telling the specs of my setup, gave it all the files. Claude can up with a bullet proof migration plan. It created 5 migration scripts to backup, migrate, replace and test. Ran all the scripts, said everything was working. Couldn’t connect to burnati.com. It trashed the site. It stripped SSL out of the nginx configuration for the web server. Three hours later, went through two chats, now its fixed. The site runs on AlmaLinux 10 a Hyper-V VM, I took a checkpoint before all this, I could have reverted but I wanted Claude to fix it.
AI absolutely can get it wrong, presuming things (I gave it the nginx config file, that it stripped out 443?), but without it I couldn’t do what I have done so far with burntai.com.
Again!
Right in the middle of implementing a new feature, just when I got it trained, it knew exactly what needed to be down, now I need to retrain the next chat :(.
Claude Greetings
I love Claude! I love the prompt when I go to claude.ai
I have been really getting in to building with Claude, like spending most of my waking hours coding with it.
The different greetings it gives you. I don’t have an image of all them, but I will.
“Jeff Returns” “Back at Again”
Addicted to AI
I have a binge thing with coding alongside AI — time stands still, days blur together. I swore it was Wednesday, but… nope, it’s Thursday.
I create dreams and ideas, and then I actually create them. To be able to type a single sentence and have it turn into a working game? WTF. I built an entire website that I think is pretty damn cool.
It gives me an endorphin rush like a video game, but 100 times stronger — because I’m not just playing, I’m making. I can say, “Build a video game with these specs,” and it happens. I can chat with AI to design a web app or troubleshoot an error, and the thing actually comes to life.
This isn’t just code. It’s creativity on demand. And I can’t get enough.
30 Years in IT — Now Partnering With the Machine
After three decades in IT, I’ve gone from racking servers to building with AI. I’m not writing the code — the machine is — but I test, tweak, and make sure nothing catastrophic slips through. Every prompt feels like a roll of the dice: did I get it right, or will it break something? I’m not a coder, but this AI journey is making me one. And yes — I’m glad there’s a dev site to test in first.
BurntAI.com and the “Goldfish Memory” Upgrade
If you’ve followed my journey building BurntAI.com, you know I’ve been leaning heavily on Anthropic Claude for development. Why? Simple — it spits out functional code with less prompting. I can drop in a request, and it gives me something that’s 80–90% ready to paste into my project. That’s invaluable when you’re cranking out features for an AlmaLinux 10 server at warp speed.
But there’s always been one giant frustration: Claude’s memory is like a goldfish. Ask it about something we discussed yesterday, and it’ll blink back at you like, “Who are you again?”
Meanwhile, other chats — even ones I barely use — can “remember” details. Claude? Not so much. At least… until now.
August 11, 2025 — The Big Announcement
Claude 4.1 rolled out a new feature: Search and reference chats. Think of it as memory’s slightly less impressive cousin. It’s not true recall, but it lets Claude search past chats and pull relevant ones into the conversation.
Of course, it’s MAX plan only — but lucky for me, with everything I’m building on BurntAI.com, I qualified.
To turn it on:
Go to claude.ai.
Click your profile picture.
Go to Settings.
Find Search and reference chats and flip the slider to ON.
Now Claude can pretend it remembers me.
But… Not So Fast
The key word here is search. It’s not true “memory” — Claude doesn’t know me, it just looks through our old conversations for matching topics. And like any search engine, it sometimes gets things hilariously wrong.
Case in point: I’ve only ever built BurntAI.com on AlmaLinux 10. I’ve told Claude this in countless chats. Yet the other day, while migrating a dev feature to production, it confidently announced:
“Since you’re running Debian/Ubuntu in production…”
Uh… what?
Apparently, even with the power to reference past chats, Claude can still invent alternate realities. It’s like having an assistant who takes great notes but sometimes flips to the wrong page in the notebook.
My Takeaway
This is progress. Being able to search old chats saves me from endless copy-paste archaeology. But it’s still not the true memory I’m hoping AI assistants will have in the future — where they can accurately track long-term projects without hallucinating my OS.
For now, I’ll keep using Claude as my main dev partner for BurntAI.com, goldfish brain and all. Because even if it sometimes confuses AlmaLinux for Debian, it still writes some pretty awesome code.
-Jeff
Building BurntAI.com with AI: The Back-and-Forth Battle
Building BurntAI.com with AI: The Back-and-Forth Battle In my first post, I shared how I went from decades in IT to diving headfirst into building BurntAI.com. Now I want to talk about how I’ve been building it — and the challenges that come with working alongside AI as your coding partner.
My Go-To: Claude from Anthropic From the start, Claude has been my main co-developer. I like it because its default style is simple: I ask for something, and it just makes the code. No overexplaining, no “are you sure you want to do that?” — just code.
The process looks like this:
I describe the feature I want.
Claude writes the HTML, CSS, or JavaScript.
I run it.
Something breaks.
I paste the error back in.
Claude fixes it.
It’s a loop that feels like real pair programming — fast, iterative, and creative.
Hitting the Context Limit Wall The big frustration? Claude’s maximum length limit.
I’ll be halfway through solving a complicated bug, deep into a multi-file feature… and suddenly, it’s out of room. Conversation over. No more replies. All the context — gone.
That means I have to:
Start a new chat.
Re-explain the project and the bug.
Paste only the necessary code (to save space).
Hope it picks up where we left off.
Sometimes it does. Sometimes I get a totally different approach that I have to untangle.
Trying Other AI Platforms I’ve also worked with Gemini and Perplexity. They’re smart, but getting them to match Claude’s “just make the code” flow takes careful prompting. Some AI tools want to explain every step in detail. Others like to redesign my feature completely.
They’re great for fresh ideas and code reviews — but when I just need the code, Claude still wins.
Lessons Learned Along the Way AI is an amazing co-developer — but not a perfect one.
Every AI platform has its own “personality.”
Prompt engineering matters just as much as coding skills.
Always save your work — AI’s memory can vanish in an instant.
BurntAI.com is still evolving. I’m adding new features, fixing bugs, and finding more easter eggs to hide. And while AI sometimes feels like an unreliable genius, I wouldn’t have been able to build this site without it.
Jeff
Long Time In IT, First Time Developer
My First Post: From IT Engineer to AI Web Creator For over 30 years, I’ve been deep in the world of IT — working as a LAN Administrator, Systems Administrator, and Systems Engineer. My toolkit was filled with PowerShell scripts, batch files, and all kinds of automation. But one thing I never really touched? Building websites and web apps.
That changed in the last six months.
With the rapid growth of AI tools, I saw something shift — suddenly, the skills I had in automation and scripting could be supercharged. AI wasn’t just a helper; it was a creative partner. It could take my technical know-how and help me build things I’d never tackled before.
Then my six-month contract ended. I found myself unemployed.
Instead of jumping straight back into the job hunt, I went heads down into a project I’ve come to love: BurntAI.com.
It’s a post-apocalyptic-themed AI hub packed with interactive experiments, real-time AI status monitoring, quirky games, and hidden easter eggs. I’ve been using Anthropic’s Claude as my main coding partner — because Claude tends to just make the code — with Gemini and Perplexity filling in when I need a second opinion.
The process is a mix of rapid creation and frustrating roadblocks. Claude’s great until it hits the context length limit right in the middle of solving a tricky problem. Then it’s back to re-explaining, breaking the problem into smaller chunks, and pushing forward.
But here’s the thing: Even with those limits, I can build more now than I ever could before. The speed, the creativity, the sheer ability to take an idea from my head to something people can click on — it’s addictive.
I’ve discovered something important: I’m not just an IT engineer anymore. I’m a creator.
And this is only the beginning.
Jeff