🦄 Confession time: when I first started writing this post, I was mad. Not mildly irritated, not “eh, I’ll just wait for the next patch” annoyed. I mean genuinely fed up with Copilot, ChatGPT, and everything GPT-5 in general. And if I make it to the end of this post in one piece, maybe you'll understand why.

---

The Funeral 👻

First, let me start this out by stating that OpenAI has officially buried the excuse that GPT can’t code. We had the funeral over the weekend. All its friends were invited. Laid to rest. Cremation done. Trees planted. Memorial service complete.

It’s done — GPT-5 can code. Full stop.

There's no room for debate. No more hacks with pages of instructions or GPT hanging out on the sidelines because it's busy staring at the plan. OpenAI nailed it with an absolutely solid model.

Plus VS Code's release came with some more experimental features that turns the already-solid GPT-5 in to a “send a prompt, make a plan, deep thinking, Implementation Specialist at your service” experience. That being said, I still haven't had nearly enough time with the full model yet.

🦄 Why, you ask? Because my brilliant self left my personal account logged in while working last week. That account? It's the cheap $10 one. I burned through all my premium requests before I even noticed. 🫥

So yes, GPT-5 (the real one) passed its early test runs beautifully — small to medium features, planning and implementation, no complaints. But the deep dive it deserves? That’s still coming — stay tuned.

What was I doing instead of the GPT-5 deep dive? Playing the lead role in the horror story of Full House starring Kimmy Gibbler and the never-ending wall of text. 🧗‍♀️😑

---

Duct Tape, Please? 🤐

GPT-5-Mini is not bad. Let’s clear that up. For small, targeted context? For quick features? For implementing a clearly defined task with accuracy and precision? It’s great. Honestly. Kimmy Gibbler is a surprisingly solid coder.

But the output.
Oh. My. God. The. Output. 🫨😲😵‍💫

I'm talking about walls and walls of text. Waterfalls of “look at me, I wrote a test!” status updates flooding my chat window like it's trying to set a world record. No, Kimmy, I don’t need a play-by-play novella about the journey you took to create a three-line function.

🦄 Sure — I started out at least trying to skim what it was saying. But my sanity didn't last long into that battle. Seriously, if I read another scroll of verbose rambling I might have retreated to some dark corner until next Tuesday.

---

The Standoff

At one point, it got so bad that I set up a stand-off. I told GPT something along the lines of, “Go check on that workflow. If it works, give me a checkmark. That’s it. If you say anything else, I’m turning you off forever.”

It complied. One lonely ✅ appeared.

🦄 Victory? Maybe. For about thirty seconds. Then the rambling resumed. 🤦‍♀️

Here’s what I didn’t realize at first: I had also turned on every single experimental preview feature in VS Code. That’s my usual routine — release notes drop, I’m flipping switches left and right.

I thought I was enabling a neat little planning checklist feature. Which to be fair, I did do that... I also omitted its counterpart to internalize that chatter. Which meant the Mini chaos doubled — and nothing I added to the prompts or instructions made a bit of difference. 😫

🦄 I even tried to tame it with a new Principal Pragmatist chat mode. Didn’t matter. Kimmy kept talking. Nonstop floods of text testing my sanity at every turn. Thanks, Kimmy.

---

Leatherface Joins the Party

GitHub Copilot is only part of the equation, though. Because just when I thought I’d found a rhythm there, here comes ChatGPT starring as Leatherface, chainsaw roaring, doing a happy dance all over my carefully constructed blog draft. Shredding paragraphs. Scattering virtual scraps across the room. Obliterating what had been a perfectly flowing system of prompts and edits.

🦄 I mean, this post normally would have taken me a couple of hours. Instead, three days later, I'm still wrestling Leatherface for control of my own sentences.

Commands escalated. From firm. To censored. To questioning life choices. And finally, full-on ultimatum. I even took a few hours detour with Gemini — proof that I'm willing to go to war with an LLM over word count. (I'm also still losing that battle!)

The moral of the story? While GPT-5 is a great coder — it is more than terrible at creative writing. So, why do we have a single model replacement for 2 completely different problems, OpenAI?

---

Where We Are Now

So where does that leave things?

• GPT-5: rock-solid coder, worthy of the “Implementation Specialist” title, but still waiting for its fair turn in the spotlight with me.
• Mini: great in short bursts, but do not let her narrate. Especially not with every experimental feature flipped on. Else plan for duct tape.
• ChatGPT: still a chainsaw in a library butchering my every attempt at creative writing.

And me? Still here. Still writing. Still dragging you through the chaos with me, because apparently this is how I process the fact that we had a funeral for “GPT can’t code” and I celebrated it by duct-taping Gibbler and dodging Leatherface.

Sounds like a victory, right? Well, sort of. At least the tombstone is carved. 🫶👻

---

🛡️ Responsible AI Footnote

All opinions here are my own. Real-world AI results depend on setup, context, and the occasional experimental toggle you probably shouldn’t have turned on in the first place. Test responsibly.

🪓 Approved by Leatherface (ChatGPT).