I stayed up until 3am last night talking to an AI. Not because I had to. Because I couldn't stop.

I've been a developer for years. I've seen hype cycles come and go. Blockchain was going to change everything. NFTs were the future. The metaverse was inevitable. I've learned to be skeptical when the tech world collectively loses its mind over something new.

ChatGPT is different. I'm writing this at 7am on four hours of sleep because I genuinely believe we just crossed a line that we can't uncross.

What I Actually Did Last Night

It started simple. I asked it to explain a Kubernetes networking concept I've been struggling with. The answer was clearer than anything I'd found on Stack Overflow or in documentation. Not just correct, but structured in a way that built understanding from first principles. Within weeks, I had largely replaced Stack Overflow with ChatGPT for most of my daily questions.

Then I asked it to write a Python script to parse some messy CSV data I'd been putting off. It wrote it in seconds. Working code. With error handling. With comments explaining each step.

Then I asked it to refactor a React component I'd been meaning to clean up. I pasted in my ugly code and it handed back a clean version with proper hooks, better state management, and an explanation of why each change was an improvement.

At that point I was fully awake. I started throwing harder problems at it. Algorithm questions. System design scenarios. Debugging sessions where I described symptoms and it narrowed down the cause. It wasn't perfect every time, but the hit rate was shocking.

Why This Feels Different

I've used AI coding tools before. GitHub Copilot is useful for autocomplete. But ChatGPT isn't autocomplete. It's a conversation. You can give it context. You can ask follow-up questions. You can say "that's not quite right, the constraint is X" and it adjusts.

The thing that got me was asking it to explain its own code. I said "walk me through why you chose this approach over the alternative" and it gave a thoughtful comparison of the trade-offs. That's not pattern matching. That's something new.

Previous AI hype was always about potential. "Imagine what this could do someday." ChatGPT is useful right now, today, for actual work. I used it to solve a real problem I'd been stuck on. That's not a demo. That's a tool.

What Worries Me

I'd be lying if I said this doesn't make me a little nervous. Not about losing my job tomorrow, but about the trajectory. If this is where AI is in early 2023, where is it in 2025? In 2028?

It also makes mistakes. Confident, articulate mistakes. It told me a Python library had a method that doesn't exist. If I hadn't known better, I would have wasted an hour debugging phantom code. The failure mode isn't "I don't know." The failure mode is "here's a completely wrong answer delivered with total confidence." That's dangerous for junior developers who can't spot the errors.

And the ethical questions are real. This thing was trained on all of our code, our writing, our knowledge. The people who created that training data aren't getting compensated. That's a conversation the industry needs to have.

Where I Think This Is Going

I think we're at the beginning of the biggest shift in software development since the internet itself. That sounds dramatic. I know. But I just watched an AI write better code than I write on a tired Friday afternoon, then explain why it made the choices it did.

I'm going to spend the next few weeks really digging into this. Testing its limits. Finding where it breaks. Building things with it. I'll share everything I learn here. (Update: I did exactly that, and wrote about it in my 2023 year in review.)

But my gut feeling, on four hours of sleep, is that everything about how we write software is about to change. Not in five years. Starting now.

If you haven't tried it yet, stop reading this and go try it. You'll understand.