Hey Reader,

You're in trouble if you're non-tech and use tools like Claude Code, Codex, and Droid

I've also tested the recent Chinese models, and have a final opinion on Cursor's Composer-1.

Here's everything that happened this week.

The Terminal Trap

Terminal AI coders are lightning fast. Super fun. And that’s exactly the problem.

They’re SO easy to use that I stopped paying attention.

Think of it like cooking an instant microwave meal vs. actually cooking.

The microwave is faster, but you have no idea what’s actually in your food.

I built entire apps in these terminals and loved every second of it.

That’s until I realized that I had genuinely no clue what they were doing under the hood.

Not the code (I still don’t read code). But the actual application logic. The structure. The architecture.

And when something broke, I’d have been lost because all I could ask was “explain this” and “fix that”…

So I spent two weeks rebuilding my Codex and Claude Code projects in Cursor.

Not because Cursor is “better.” But because a proper IDE forces you to see the whole picture.

You don’t need to write code. But you DO need to understand what your app is actually doing.

The terminal tools aren’t bad - they’re just dangerous if you’re not already technical.

Use a proper IDE, such as Cursor or Windsurf. Your future self will thank you.

Launch Your Idea 10x Faster

I opened 3 mentorship spots last week, and I have only ONE spot left for 2025.

Here’s what you get:

• Weekly 1-on-1 Zoom calls - Build YOUR app with me
• Private Slack - Direct access to me 24/7
• MVP planning - We plan the minimum version together
• Your screen, my experience - I’m guiding every decision
• Save months of time - No need to repeat mistakes I’ve already made

​​Apply now​ - 1 Spot Left

Last chance to build this year.

New spots only open up in 2026.

The (Chinese) Model Reality Check

While I was rebuilding, I also tested GLM 4.6 and Kimi K2 Thinking (the two big Chinese AI models everyone’s hyping).

Bottom line: They’re not ready.

Can you save money with them? Sure. Should you? Absolutely not.

The U.S. models (Claude, GPT, even Gemini) are still miles ahead in actual building.

I say this with zero bias - if the China models are better, I’m the first to use them.

Don’t get distracted by cheaper options that’ll cost you way more time.

But there’s more…

I’ve also tested something called Polaris Alpha - rumored to be part of GPT 5.1.

It's really bad. Like, models from 6+ months ago bad.

I really hope this isn’t what OpenAI is planning to release. If it is… yikes.

Final Take on Cursor’s Composer-1

After about three weeks of daily use, here’s my brutally honest review.

I went back and forth here because I’m always putting myself in your shoes.

And that’s the problem: Composer-1 is a fantastic model if you're more technical.

I use it every single day, because I know that my gut will tell me when something is off.

The deep technical skills I’ve acquired over 20 years help me tame this model and make it really good.

BUT… if I didn’t have deeper technical skills, I’d be way more careful with it.

Think of it like a sports car. It’s fast and fun, but you need to know how to handle it.

Sonnet 4.5 and GPT-5 are more like reliable sedans - they just work.

My takeaway here remains that this is a great model to experiment with and to use if you want to be fast.

Especially right now, because it is free for some users (only if you received the promo email).

But don't trust it as much as you'd trust frontier models.

Talk soon,
Rob

P.S. - Getting asked about Black Friday. The Blueprint will not be on sale. It's $497 because that's what it costs to run weekly calls and show up in Discord daily. I’m not going to discount that. If you’re serious about shipping, don't wait → Start Building Today.