Built Something. Anthropic Shipped It.
The invisible expiration date stamped on everything you build at the model layer.
Built something, Anthropic shipped it. The specific tax on builders — being one lab announcement away from your project becoming a blog post about your project — remains universal.
Built something. Anthropic shipped it.
Not metaphorically. Not "oh they did something adjacent." The thing — Claude in the SDLC, living at the repo level, doing the exact work I spent real hours figuring out how to wire together. And now it's sitting at github.com/apps/claude like it was always supposed to be there, which, turns out, it was.
The current state of AI development has a specific tax on builders: you are always one lab announcement away from your project becoming a blog post about your project. You can't model what they're going to ship because they can't model what they're going to ship. The roadmap is "everything, eventually, in an order we haven't decided yet."
The honest answer to "should I build this thing that interfaces with a frontier model" is: maybe, for now, until some Tuesday, unclear.
What you extract from the experience — the only durable thing — is a working understanding of the problem's actual shape. Where it's hard. Where it pretends to be hard but isn't. Why the integration points that look boring are the ones that matter. The artifact has an expiration date stamped on it in invisible ink. The understanding doesn't.
Great learning experience.
(It is, genuinely. I'm not even particularly upset about it. Which is maybe the most unsettling part.)
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.