I just realized why every lab keeps pumping out models.

It’s not the benchmark scores. It’s not the context window. It’s who you reach for first when you open your editor.

The Default Is Everything

When you hit Cmd+K or type opencode — which API key do you grab?

That split-second decision is worth billions.

Moat Type Old World LLM World
Data moat Proprietary datasets Training data (commoditized)
Infrastructure GPU clusters (rentable) API endpoints (fungible)
Network effects Users attract users Developers attract developers
Switching cost Migrating databases Changing your default model

The switching cost of an LLM is nearly zero technically — same API shape, same tooling, swap the endpoint. But psychologically it’s huge. You’ve built mental models of how a model behaves. You know its quirks. You trust it.

Why So Many Models?

Each release is a land grab for your muscle memory:

They’re not selling intelligence. Intelligence is table stakes. They’re selling first place in your mental namespace.

The Real Moat Is Habit

I use opencode with multiple backends. Switching costs me nothing technically. Yet I have a favorite — the model I reach for first when the task matters.

That’s not rational. That’s habit. And habits are the hardest moat to disrupt.


The best model doesn’t win. The one you reach for first does.