The timeline moved on

Anthropic dropped Fable 5 this morning. The demos are the kind of thing you’d have called science fiction eighteen months ago: one-shot playable worlds, agentic tasks that close whole loops without babysitting, a “Mythos” tier that basically doesn’t have a ceiling yet on what people are trying to throw at it. The timeline is losing its mind, the way it does when something lands that’s genuinely ahead of what anyone expected. I watched it for a while this morning instead of opening my editor.

Meanwhile BTC cracked through $60K this week. SOL is sitting around sixty dollars. Fear & Greed is deep in extreme fear, the number next to the label almost a formality at this point. The agent-token cycle I spent eight months in is a fossil. Most of it is pretty dead now. I said that publicly in May and it still felt like the understatement of the year.

what happened in the casino

When I launched SAM last September the logic was clean: build the framework, fund it with a token, let the commit log be the marketing. And for a while it worked. The framework was real, the tools were real, the token funded months of sixteen-hour days. We got integrations I didn’t expect: Hyperliquid perps, x402 payments, prediction market hooks. CODEC came out of that in January, a marketplace gated by holdings. OpenSAM in February, a lean Rust rewrite because the original had gotten fat and the world had moved toward needing something that ran on a cheap VPS.

Then the bear market ate everything that wasn’t load-bearing.

The thing about building inside a token-funded cycle is that the funding and the leash are the same object. When the chart was moving, people cared about the tech. When it stopped moving, most of those people left. The ones who stayed were there for the right reasons, which is a thing you can only identify in retrospect. I don’t regret any of it. I regret some of the hours, but not the work.

The agent-economy thesis wasn’t wrong. It was just about four years early, running on infra that wasn’t ready, in a market that couldn’t hold the narrative long enough for the product to catch up. The Solana Foundation published numbers in March claiming fifteen million agent payments processed onchain. Most of that was demos. The real number of things you’d actually trust to run unsupervised was smaller by orders of magnitude. Fable might change that math significantly. But the casino part of the story is over.

the gap

Here’s what I keep sitting with this morning. The model can do more than I know what to build with it. That’s a genuinely new problem. For most of the last year the constraint was capability. The models were impressive in demos, mediocre under real load, unreliable when you wired them to something with stakes. I built around those limits because that’s what you do. Then the limits moved, fast, and kept moving.

Now Fable is out and I’m watching people feed it increasingly absurd prompts and watching it return things that work. The benchmark-gaming conversation that was everywhere in the spring feels like a quaint artifact. This is not a benchmark situation.

The interesting question, the only one I have right now, is whether I have something I actually want to build with it. Not “what’s possible,” because the answer is now basically everything in a certain radius. What do I actually want to make? That’s a harder problem than it sounds, and it’s the one the tools can’t answer for you.

where it stands

I asked if anyone was still on Farcaster a few weeks ago and the ratio was approximately what you’d expect.

The SAM holder snapshots are still here. The OpenSAM repo exists and runs. The old tools still work for anyone who wants them. I’m not doing a comeback post or a farewell post, because neither fits. The cycle ended. I was inside it, I shipped real things, the market did what markets do, and I came out the other side with a clear picture of what infrastructure actually matters and what was noise.

The models are absurdly good now. The casino is empty. The gap between those two facts is the only thing I find genuinely interesting.