That doesn’t diminish what has changed. It tells us what changed: we’ve experienced an enormous reduction in the marginal cost of attempted cognition, not the abolition of scarcity, causality, dependency, verification, or time.
That means that rewriting ttfx in assembler doesn't mean I can't also do everything else I set my mind to. Because it doesn't cost any meaningful amount of attention or time. It's just tokens! That's why we can fix everything!
I asked the agents a simple question: which stack serves our workflows fastest? The answer was surprising, not because we unseated our default Apache+mod_php, but because it exposed bugs that would make things slow no matter what stack we used.
So the runtime question is still open. But the real ROI so far wasn't "pick a faster server", it was using the hunt as an exercise to stress the system hard enough that it told us what was actually broken.
Between the old and new package names, Durable Workflow has passed 1M composer installs and now sees about 150K more every month. Thank you to everyone building with it. You’re awesome! 🙌
Pick the runtime capacity you need. Pay a predictable monthly price. Retries, timers, signals, queries, child workflows, and activity attempts don’t become surprise line items.
We publish the benchmark for every tier so you can see what that capacity actually delivers.