A few months ago, I was the skeptic in corporate meetings rolling my eyes at AI demos. "It's just resummarizing stuff," I'd think, watching presentations about magical data consumption and perfect outputs. The originals always seemed obscured, making me more skeptical, not less.

But something shifted when I started using it for my own problems instead of watching other people's demos.

The Tee Time Problem That Changed Everything

I had this tea time booking problem that had been nagging me for months. The local course releases times sporadically, and by the time I checked manually, the good slots were gone. I'd tried setting reminders, checking religiously—nothing systematic worked.

So I asked ChatGPT: "Can we write a bot that scrapes tee times and alerts me when spots open up?"

A few hours of back-and-forth later, we had working code running on a Raspberry Pi. Not perfect code—I'm not a developer—but functional code that solved my actual problem.

That's when it clicked. This wasn't about the AI being magic. It was about having a systems thinking partner that could help me break down problems, suggest approaches, and iterate on solutions. Ask questions about ideas. Learn about components I'd never heard of.

The tee time bot was just the beginning.

When Wix Wanted $350 to Keep My Site Running

A few weeks after the bot project, Wix emailed asking for $350 to renew my website. Three pages. Mostly static. Barely any traffic.

"Bullshit," I thought. "I can build this myself."

I found Eleventy and Tailwind CSS, got a repo going, loaded it onto the Pi. Fed ChatGPT screenshots of my current site and asked it to recreate the look. Deployed the whole thing to GitHub and Render.

Cost went from $350/year to $12/year. More importantly, I understood every piece of the system.

The $30/Month Google Problem

Looking at my expenses, I realized I was paying Google $30/month for email and document storage. But here's the thing—I'd started converting everything to markdown files anyway because they work better with AI tools. Markdown is cleaner, more structured, easier to process than Google Docs' formatting soup.

So I moved email to ProtonMail, downloaded everything from Google Drive, converted it all to markdown, and stored it in a content folder on the Pi. Saved another $30/month.

Now all my projects—code, content, documentation—live on the same device where AI can actually access them.

The Memory Problem

I tried using ChatGPT's memory feature for a while. Turned it on, started context engineering, feeding it background on projects and preferences. It worked great initially.

But memory sprawl is real. The system starts remembering everything—conversations, tangents, half-formed ideas that weren't meant to stick. You tell it something once and it becomes part of your permanent record, influencing every future interaction.

I'd have to ask it to dump its memory into a markdown file, edit out the irrelevant parts, wipe the slate clean, and reload the curated version. Even then, it seemed to carry baggage from other conversations.

The solution isn't better memory management. It's designing systems where you control exactly what context gets loaded into each session.

What I'm Building Toward

Right now I'm looking at MCPs—ways to connect Claude desktop to my local systems. The goal isn't fancy automation. It's eliminating the copy-paste overhead that keeps AI at surface level.

Imagine this: I need to write an email to my former boss. Instead of copying our previous emails, finding relevant project details, and pasting context into a chat window, I just tell Claude: "Draft an email to Mike about the Q3 results project."

It already knows our communication style from past emails. It knows what projects we worked on. It knows the context without me having to assemble and paste it every time.

That's the difference between using AI as a chat tool and building it into your operating system.

The Real Unlock

Most people are using AI like a better search engine or writing assistant. Copy, paste, prompt, copy the output back. That's surface level.

The unlock happens when you stop thinking about AI as a tool you use and start thinking about it as infrastructure you build on. When your data, your projects, your communication history all live in connected systems that AI can access directly.

You're not becoming a prompt engineer or a context engineer. You're becoming someone who builds systems that remove friction from getting real work done.

I don't think people realize how close they are to building exactly what they want. It just requires taking the time to understand what means what, learning how the pieces connect, and building systems instead of workflows.

The chat window is the beginning, not the destination. The real power comes when you connect AI to everything else you're already doing.

That's what I'm working on building. One markdown file, one API connection, one small system at a time.