Two weeks ago, the US government switched off the most powerful AI model in the world.
Fable 5 was gone overnight with no warning.
I'd built my entire coding workflow around it.
Multiple agents, verification loops, the lot.
And it was just gone.
I made two videos about this over the last couple of weeks, and if you want the full deep-dive, they're linked at the bottom. But there's one analogy from the second video that I think every builder reading this needs to hear.
Back in the day, if you built your whole audience on Twitter or Instagram and the platform banned you or the algorithm changed, it was over.
You had no recourse.
Your audience belonged to them.
That's why smart creators started building email lists.
Because an email list is yours. Nobody can switch it off. Nobody can change the algorithm. And nobody can ban you from your own subscriber list.
Right now, most of us are building on closed AI models (like Anthropic’s Claude and OpenAI’s GPT) the same way people used to build on social media platforms.
We're renting access to intelligence that lives in someone else's building.
And as we just saw with Fable 5, that access can disappear with a single government letter.
Why GLM-5.2 changes this.
Four days after Fable 5 got banned, a lab in Beijing released GLM 5.2, an open-weights model that scores within about 5% of Opus 4.8 on coding benchmarks.
Open weights means you can download it, run it, fork it, and no company or government can switch it off.
But for 99% of people, running it locally doesn't make sense right now.
The hardware costs tens or hundreds of thousands of dollars.
But that's not the point.
Here's why you should care even if you never touch it.
A free, near-frontier model existing on the open market does three things:
It puts a floor under the whole industry. Every closed lab now has to price and behave knowing there's a free alternative that nobody can switch off.
It kills the "permanent underclass" narrative. The idea that frontier AI would become too expensive for normal people to use just got a lot harder to argue.
It shifts where the value is. Access to intelligence is no longer the edge, because everyone has it now. The edge is what you actually build with it.
Or as I put it in the video: models commoditise labour, not building. Your ability to think, to have taste, to know what to build and why, is the thing AI can't replace.
And it's worth more now than it was a year ago.
What to actually do about this.
You don't need to run local models tomorrow, but you should probably stop building your entire workflow around a single provider.
Test an open alternative this week.
Even just to know it exists and roughly how it performs.
GLM-5.2 is available through Fireworks, through Vercel's AI Gateway, and through a bunch of other providers.
This way you can use it inside Codex, Claude Code, or whatever works for you.
(none of these are paid sponsorships, this newsletter is always 100% sponsor-free)
Think of it like your email list.
You don't stop posting on X just because you have a newsletter.
But you make sure that if X disappeared tomorrow, you'd still have a way to reach your audience.
Same thing with AI. Keep using Claude, keep using Codex, but make sure that if one of them disappeared tomorrow, you wouldn't be starting from zero.
Full videos:
→ We need to talk about Fable → [watch]
→ Why GLM-5.2 is a bigger deal than you think → [watch]
Question for you: did the Fable 5 ban affect your workflow? Reply and let me know.
See you next Tuesday.
You've got this.
- Rob
P.S. It’s my birthday today! I just turned 25 :)
