Era of Agents: The Rise of Open Weights LLMs

Hey guys. About two weeks ago, Anthropic released their new Claude model, Fable 5. It was an exciting moment — they’d been marketing it for so long — and then the US government blocked it. It was very interesting to watch, because the government actually forced Anthropic to pull a model that was already released. It had been out for something like two days.

Later, OpenAI’s GPT 5.6 was put on hold, as far as I understand — the public release of the model was paused while that same US government evaluated its safety and security, or whatever they call it. And more recently, I read in the news that China is now weighing whether to restrict foreign access to their most advanced LLMs as well.

So basically, the two greatest powers in the world decided to use their strong hand. And suddenly all of us realized how vulnerable and how dependent we are. What do we do? Where does that leave us?

It’s actually very simple, and I’m optimistic — because I believe this puts us in a better position. This was a geopolitical decision, so the response has to be geopolitical too. A commercial company cannot fight with a government. Yes, Anthropic tried, but by the end of the day they just caved.

So the response has to come from all these smaller players. And by smaller players I don’t necessarily mean small countries — they could be large countries too — but essentially the ones with fewer resources, the ones who can’t train a flagship model by themselves. And because there are so many of them, they can’t easily cooperate to train one together either. So they have no choice. If they want to preserve some kind of technological independence — to hold onto the dream of being independent one day — they have to support open weights models now. Open source agents, protocols, libraries, all of it.

And here’s the interesting twist: the dream of a small company having its own model, the dream of your own small LLM running on your own personal device, suddenly has a decent chance of coming true. Commercial models keep getting more powerful and commercial agents keep getting more capable — but that very pressure is what puts the rest of us in a better position to have smaller, powerful models, and smaller, powerful personal agents.

I’ll leave you with an analogy. Decades ago, this is exactly how Linux went from a hobby project to the operating system running the whole internet. More or less the same setup. So I’m very bullish on this, and I hope you are too. See you guys. Bye-bye.