GPT-5 is driving everyone nuts.

And honestly? I get it.

Here's what happened & how to fix it:

The Puppet Master Problem

OpenAI is playing puppet master with something they call "model switching."

They decide which model you get.
Not you.
Never you.

Their logic sounds reasonable on the surface: "We'll make it easier on users by automatically selecting the best model for each task."

The reality?

It's a flaming pile of inconsistency that's making smart people feel like they're losing their minds.

One minute you're getting genius-level insights that make you feel like you've got the world's best thinking partner sitting right next to you.

The next minute you're getting intern-quality work that makes you wonder if someone switched your AI with a Magic 8-Ball.

The Emotional Toll

Some folks describe it like this: "It feels like we just had our best teammate quit without notice. And now we have the intern pretending to be them."

That hits different, doesn't it?

Because that's exactly what's happening to your workflow, your confidence, and your trust in these tools.

You've built processes around having a reliable thinking partner. You've learned to lean on AI for the heavy lifting. You've integrated these tools into the core of how you work.

And now? Now you're playing Russian roulette every time you hit enter.

The Deeper Issue

But here's what really bothers me about this whole situation...

It's not just about inconsistent outputs.
It's about control.
It's about agency.
It's about keeping the human in the loop instead of pushing us further out.

Every time a company makes a decision "for our convenience" that removes our ability to choose, we lose a little bit of our humanity in the process.

I love these tools. I really do.
But I also love my people.

And when tools start making decisions for us instead of with us, we're heading down a path that doesn't end well.

Your Game Plan (The Real Solutions)

Here's how to take back control:

1) Become a Prompt Engineer Again (*insert sad trombone*)

I know. I know.
We thought those days were behind us.
We thought we'd moved past the need for perfect prompt engineering.

But consistency requires consistent input now.

Use the exact same prompt structure every time:
- Start with context
- Be specific about the output format you want
- Include examples when possible
- End with clear instructions

It's annoying. It feels like going backwards.
But it's the reality we're dealing with right now.

2) Demand What You Want (And Mean It)

Don't just hope for good results.
Demand them.

Add "think deep" to your prompts when you need the heavy hitter.
Start with "/think" or tell it to use "high effort."
Be explicit: "Use your most advanced reasoning for this task."

Sometimes it works. Sometimes it doesn't.
But at least you're trying to signal what you need instead of leaving it to chance.

3) Take Back Control (My Personal Favorite)

This is where it gets interesting.

First, ask ChatGPT to dump everything it knows about you. Export your conversation history. Save your custom instructions. Don't lose that context you've built up.

Then move to a tool that lets YOU choose the model:

Level 1: The Easy Switch
- Claude & Gemini (different companies, your choice)
- Both have their strengths
- Both let you know what you're getting

Level 2: The Power User Move
- TypingMind (my current go-to)
- Lets you switch between models mid-conversation
- You control the experience

Level 3: The Next Level Approach
- Cursor and Windsurf for "vibe" everything

Fun fact: If you connect with an API tool, you can still select your favorite OpenAI models. You just get to choose when and how to use them.

The goal isn't to abandon AI.
The goal is to use it on your terms.

Because at the end of the day, you're not just trying to get better outputs from your tools.

You're trying to build a business you love.
You're trying to create work that matters.
You're trying to make an impact that lasts.

And that requires you to stay in the driver's seat.

Keep building,

-Dan

P.S. What's your experience been with the model switching chaos? Hit reply and let me know. I read every response, and your insights often spark my next deep dive into these topics.