Hey Reader,

I owe you an update from Week 1 of my Cursor vs Claude Code challenge.

Here's what nobody told me.

Opus Isn't "The Best"

I know, I know.

Everyone says Opus is the greatest AI coder ever made.

"Alien Tech"

And look, it IS incredible at some things.

It's fast. It's eager. It executes plans like a maniac.

Working with Opus feels like pair programming with another person.

I've watched it run for 30-45+ minutes straight without breaking a sweat.

But here's the problem...

Opus Has Amnesia

As your project grows, Opus starts forgetting things.

It loses track of what your app already does.

It creates duplicate code everywhere.

It's like working with someone who has goldfish memory.

Great energy. Zero recall.

And that creates no only bugs, but also problems down the line.

This is the part nobody talks about.

GPT-5 Is "Slow" On Purpose

Remember how everyone complains GPT-5 takes forever?

Turns out that's not a bug. It's the whole point.

GPT-5 isn't slower underneath. It's more careful.

It doesn't just start hammering away.

It stops. It thinks. It checks if something already exists in your codebase.

It considers edge cases. Thinks through different paths.

Carefully making sure a code doesn't already exist somewhere else.

Think of it like this:

Opus is the friend who immediately grabs a hammer and starts swinging.

GPT-5 is the friend who reads the IKEA instructions first.

Both have their place.

The Brutal Truth

After one week, I'm more confident than ever:

The best AI builders use BOTH models.

Sure, "proper" engineers can get away with Opus 4.5.

But they read code, fix bugs manually, and understand every change.

You don't. This is hard for non-technical professionals.

So you need both - use GPT-5 when it counts and accept the slower pace.

There's no one best model.

Anyone who says otherwise is selling you a lie.

What's Next?

The challenge is clearly morphing.

It's no longer just "30 days of Claude Code."

Now I'm asking: Can I live without Cursor entirely?

Using tools like Conductor with Claude Code and Codex instead.

Codex has blown me away, especially now that it supports skills.

Most skills are garbage. But some are really good.

They enable things Cursor can't even do yet.

So I'm really excited for the next 3 weeks.

Talk soon,
Rob

P.S. If you're confused about which AI model or tool to use for YOUR app: that's exactly what I help you with inside my AI Builder's Blueprint 🫰

​