Sometimes the news takes over and ruins your weekend. This was one of those weekends.
I got access to Anthropic’s amazing new Fable 5 model starting last Thursday late afternoon/evening. I went into build mode immediately.
Within just a few hours I rebuilt a bunch of systems for my new startup, Rustproof. Then Saturday morning it was gone. Because we’re not allowed to have nice things, Anthropic’s ongoing tensions with the Administration have reignited and led Anthropic to pull access to the model globally.
This newsletter isn’t about who is right or wrong or who is to blame in that dispute (although if you ask me both are at fault for different reasons). The message is: the frontier keeps moving and the pace is accelerating.
I have to assume that we will get access to a model like Fable 5 again after some agreement between the Administration and Anthropic. We aren’t going to stop AI progress while trying to race ahead of the Chinese and OpenAI will probably have a similar model soon.
The rest of this post was written Friday and based on the assumption that we’ll all have access to these capabilities. And those capabilities are another game changer in a year of game changers. We went from a 10-20% increase in capability with Opus 4.8’s release just 2-3 weeks ago to what feels like a 2-5x increase in capability with Fable as of last Thursday.
WTF?!
Can this continue?
Who benefits?
Here's what surprised me, though. The jump doesn't lift everyone. It sorts people.
For regular everyday work, Fable is overkill, and usually counterproductive. Ask it to write a memo or stand up a simple report and the gain barely shows up. Sometimes the output gets worse. The model is too big for the shape of the job.
It's like asking a Nobel Prize–winning physicist to help with your TikTok. Could they help? A little. Would they be worse at it than your 13-year-old niece? Absolutely.
You need work big enough for the model to stretch.
I've been living in that kind of work at Rustproof. We're building an entire system to provision and govern AI agents for our clients: what agents exist, what they're allowed to touch, who can run them, whether they ran, what they produced. It’s a lot of moving parts. A year ago it would have taken months to years and multiple expensive engineers. Five days ago it was weeks of work: a week on the core, then days of testing, hunting errors, chasing edge cases, cleaning up.
With Fable 5 that loop collapsed again. My off the cuff estimate is that I was moving where between 2-3x faster and maybe up to 5x on certain tasks. I was getting complex builds in a single session, off a single plan.
I usually build with one model and review with another: build with Claude, tear it apart with Codex, or the reverse. There's always a list of holes the reviewer pokes in the other's work. With Fable the list nearly vanished. The comments from 5.5 went from paragraphs to closer to a single line, "This is a well-designed way to do X."
On Thursday, I got bumped up to first class. Everything was roomy and sleek. (Imagine my chagrin when they bumped me back to economy on Saturday.)
What's slow isn't the tool but the builder
Everyone is going to be able to buy this so access is not the bottleneck.
What's slow is the willingness to move. And that's not about skill.
The motivated builder moves at full speed because that’s what they do.
Others might be motivated but won’t be willing to cannibalize the exact model their current profit sits on top of. So many people are going to freeze.
You'll hear that freeze dressed up as calm. "It doesn't really change my day-to-day." "Let's wait and see." "Most people aren't even using it well yet." All technically true. All the sound of an incumbent talking itself into runway it doesn't have.
Because the threat just jumped forward in time. The day you start competing with a motivated builder who has this stuff got a lot closer this week.
The moat is the velocity
But let’s remember that this isn’t one and done. You can’t hire a contractor, build something and then coast. Fable 5 is proof that the frontier will keep jumping, and any single build advantage decays with those jumps.
Every edge you get will have a half life that is shortening all the time. The 5x from Fable will be normal soon. The window in your vertical will close. None of it lasts.
Except one thing. When the frontier jumps again, and the arbitrage reopens, whoever is positioned to grab it the day it lands wins. So the edge that compounds instead of decaying isn't any model or any tool. It's adoption velocity itself: being the person, or the team, built to absorb each jump the moment it arrives, while everyone else is still arguing about whether it's real.
That's the moat. Not what you build. How fast you turn the next capability into something live.
And it's the one advantage the frozen incumbent cannot copy, because copying it means becoming the thing they are structurally built not to be: an organization with nothing to protect and every reason to move.
So the question Fable left me with isn't "what did you build in those 36 hours." It's the one worth sitting with this weekend:
When the frontier jumps again (and it will) are you built to move?