A rollercoaster of pride, panic, and AI gasliting

I code to relax. Some people do yoga, some meditate—I open VS Code at 11 PM like a gremlin-raccoon. Don’t judge.

I work full-time at Microsoft, and between family time and my other hobbies, I somehow keep starting new side projects. I’ve got VocalCat, an AI interviewer for SOMOS.tech, some fediverse experiments, and a Mastodon fork where I play with identity and static blog integrations. And last night, against all common sense, I started another one.

I couldn’t help it. The itch was too strong. I spent hours before sleeping, just thinking about the implementation, and suddenly, I blacked out and woke up at 1 AM with a new repo and a half-baked ActivityPub prototype in .NET.

Now, if you know the fediverse, you know that Microsoft isn’t exactly its favorite company, and for a reason. If you are not in the fediverse, you should know that Microsoft is one of the less “appreciated” companies there, and with reason. The Microsoft of 20 years ago is not the same as today. I am proud to work at the Microsoft of today, but I don’t blame someone who has reservations about it. I would probably have them as well if I hadn’t seen things from the inside. Still, using .NET for an ActivityPub project is basically me speedrunning how fast I can get ignored by the community. Whatever. I was on a mission.

So, I did what any reasonable person would do: I called in the AI reinforcements. Copilot, Claude, ChatGPT, GitHub Copilot, you name it—I use them all like a chaotic symphony of artificial intelligence. Planning travel, reformatting documents, fixing grammar, mainly coding. Sometimes I get better answers by just asking directly to ChatGPT, sometimes I use the inline GitHub Copilot, sometimes the Cursor IDE. And if you’re about to tell me “Did you try X prompt?” in the comments—don’t. I know what I’m doing (most of the time).

Anyway, I asked Copilot for some help cleaning up my prototype, and the response made me look like a complete clown.

It regurgitated my own shitty code.

I mean, not just similar code—my code. My messy magic strings, my lazy namespace choices, my shortcuts, my missing activitypub implementations. I hadn’t even used Copilot for the first implementation. But it was open-sourced, got slurped into some LLM training set, and now it was getting thrown back at me like a loogie that missed its target.

So now, I’m sitting here with two conflicting emotions:

  1. Pride – I made it. My code is out there, influencing AI. I am immortal (in the most insignificant, background-noise kind of way).
  2. Existential crisis – If the AI is just regurgitating my own mess, how the hell am I supposed to trust it to know better than me?!

Look, I don’t blindly trust AI, but I do check its suggestions. And I hate to admit it, but often it is right. Many times, I’ve grumbled and copy-pasted its fix because it was better than my first draft. But sometimes? Sometimes it’s just like getting code reviews from an overly confident junior engineer who barely skimmed the requirements.

Moral of the story? AI copilots are cool, but they are not wizards. They’re just fancy autocomplete with an attitude. The Agile Manifesto says “Individuals and interactions over processes and tools,” and damn, was that reinforced for me today. AI is a tool.

So, yeah. Trust your instincts, use AI wisely, and if your own garbage code comes back to haunt you—well, congratulations, you’ve contributed to the collective digital chaos.

What about you? Have you ever had an AI throw your own code back at you? How did it make you feel, and what did you learn from the experience? Share your stories in the comments below!