Hey Reader

October's almost done.

And my AI coding setup is COMPLETELY DIFFERENT than it was 30 days ago.

Here's what changed (and what you should steal).

#1 - I Broke Up With GPT-5 Codex

Look, GPT-5 is brilliant. Codex too.

But they're SO SLOW that it kills the vibes.

I still use them every day, but not for coding anymore. Just for planning.

Think of it like having a wise old architect who takes forever to draw blueprints. Great for strategy, terrible when you need to build fast.

So I canceled my ChatGPT subscription entirely. The only reason I had it was for Codex.

But I still use GPT-5 every day, because...

#2 - Cursor's Plan Mode Is Awesome

Cursor dropped this feature called "plan mode" and I genuinely don't know how I lived without it.

I use it 10-20 times a day. No exaggeration.

Here's how it works:

I let GPT-5 (High) create the master plan (it takes its sweet time, asks clarifying questions, thinks things through).

Then I hand that plan to Claude Sonnet 4.5 (non-thinking) to execute.

It's like having a slow genius map out the strategy, then a fast worker bee get it done.

#3 - New Trend: Model Orchestration

This is the big shift nobody's talking about.

You can't use just one AI model anymore and expect the best results.

Now you need a thinking model (= GPT-5) and an execution model (= Sonnet 4.5).

That's why standalone tools like Codex or Claude Code don't work anymore.

Some people use Claude Code AND Codex together, which is totally feasible if that's your vibe.

But I prefer to have one single tool, which for me is Cursor.

What has changed for you this month?

Talk next week,
Rob

​