Something strange is happening in AI.

The systems are becoming capable enough that even some of the people building them are publicly debating whether development should slow down.

I understand the instinct.

AI capability is accelerating quickly.

Systems can increasingly reason, code, use tools, interact with software and operate with less human supervision.

That deserves serious attention.

But there’s a problem with treating “slow down” as the entire safety strategy.

If responsible people stop building powerful AI, irresponsible people don’t automatically stop too.

Unfortunately, there is no global button marked:

“EVERYONE PLEASE PAUSE FOR A MINUTE.”

Slowing Down Doesn’t Automatically Make Us Safer

AI development is happening across companies, countries, universities, open-source communities and countless private projects.

There is no single switch.

And because AI capability itself can be used defensively—to detect attacks, identify vulnerabilities and understand increasingly sophisticated systems—falling too far behind creates risks of its own.

That doesn’t justify reckless development.

It makes responsible speed more important.

The challenge isn’t simply deciding between moving fast and being safe.

It is learning how to do both.

Safety Has to Advance Alongside Capability

We need better testing.

Better monitoring.

Better permission systems.

Better understanding of how increasingly autonomous AI systems behave.

And we need those capabilities developing at roughly the same speed as the AI itself.

AI safety and AI progress aren’t opposite sides of the argument anymore.

They have to become the same engineering problem.

I’m strongly in favor of continued AI development.

I just think we should do it with our eyes wide open.