You are building something with GPT 5.5.

For two days, it feels sharp.

It understands the project.

It follows your style.

Then one morning, it feels nerfed.

More refusals. Weaker code. Strange tone.

You did not change anything.

The model changed.

Or maybe Claude Fable 5 becomes your favorite model.

Then access gets restricted.

Or the tool routes your task to another model.

Or the best open-weight model is strong, but too huge to run on your laptop.

That is where people get the wrong answer.

They say:

"Use local models."

Nice idea.

But local models are not always good enough.

The strongest open-weight models are often cloud models in real life.

Kimi K2, for example, is an open-weight MoE model with 1 trillion total parameters and 32 billion active parameters. That is not a normal laptop setup.

So the real shortcut is smaller:

Build the project as a folder.

Then bring any model into that folder.

I am Alex, welcome to ShortCu8 by Innov8.

Lets Dive Deep 🐰

Your project is the folder.

The model is only the worker you bring in today.

If your project only lives inside one Claude chat or one Codex thread, switching models is painful.

If the project lives in files, switching is boring.

That is what we want.

Boring switching.

For coding, create this:

Use AGENTS.md for how the AI should work.

Use MASTER_CONTEXT.md for what the project is.

Use PLAN.md for the current task.

Use DECISIONS.md for choices already made.

Use ACCEPTANCE_CHECKLIST.md for what done means.

Now the model can be swapped.

Today:

Codex with GPT.

Tomorrow:

Claude Code with Fable or Opus.

Next week:

OpenCode with DeepSeek, Kimi, Qwen, or another hosted open-weight model.

For product/design: use this:

This is not fancy.

That is why it works.

When you start a project, ask the model:

When the model makes a big decision, ask:

When the project changes direction, ask:

When the model starts acting weird, do this:

That is the move.

You are not hoping the next model remembers.

You are giving it the memory.

Open-weight models are useful backup workers

Small local models can help with private notes, simple summaries, rough drafts, and offline work.

Big open-weight models can be much stronger, but most people will access them through cloud/API.

So the practical stack is:

Main model for best output.

Backup model when the main one gets nerfed.

Hosted open-weight model when closed models become annoying.

Small local model for simple/private/offline tasks.

Same folder for all of them.

Now go create something great.

We read every reply - just reply to this email and let us know how we can improve !

#AD1

Picture this. A buyer opens ChatGPT and asks for a recommendation in your category. Your competitor's name comes up. Yours doesn't. And that buyer never makes it to your website.

That's happening right now in markets everywhere. And most teams don't know it's happening because it never shows up in their analytics.

HubSpot AEO shows you exactly where your brand stands in AI search, where competitors are getting recommended instead of you, and tells you specifically what to fix. No expertise needed.

Try it free for 28 days. Just $50 a month after.

#AD2

Your accounting should keep pace with your business. BELAY's Financial Experts handle books, payroll, and reporting so you can lead with confidence. Download the Free Guide