Comment by drag0s

one example where non-thinking matters would be latency-sensitive workflows, for example voice AI.