What does it mean that agent reliability matters more than agent sophistication?

A simple agent that fires consistently beats a complex agent that fires intermittently.

Example: You design a sophisticated morning routine: journal for 20 minutes, meditate for 15, review your goals, do a cold plunge, then read for 30 minutes. It works twice. Then you sleep in once, skip the plunge because it's cold, drop the journaling because you're behind on email, and the whole sequence collapses. Your colleague writes one sentence in a notebook every morning before coffee. She has done it for 300 days straight. Her single-sentence agent has produced more cumulative insight than your five-step system produced in its two appearances.

Try this: Pick one cognitive agent you've tried to install that keeps failing — a review habit, a decision protocol, a daily reflection. Strip it down to the absolute minimum version that you could execute in under two minutes, in any context, with zero preparation. Run that version every day for one week. At the end of the week, compare: did the stripped-down version fire more often than the original? If yes, you've just proven the lesson empirically.