Every agent runs the same loop. Not the magical sense. The boring sense. Perceive context. Decide what to do. Act. Observe the result. Repeat.
Your AI executive assistant is doing this all day, in the background, with your inbox and your calendar.
So are you, with the assistant.
You perceive what the assistant did: a draft reply, a calendar move, a flagged follow-up. You decide whether it landed. You act: approve, edit, redirect. You observe the result. The reply went out. The meeting moved. The thread closed. Repeat.
Two loops, interlocking. That’s the work.
This is the part most people miss when they imagine “an AI assistant.” They picture one loop running flawlessly without them. That’s not the goal. The goal is two loops that get tighter every week. The assistant learns your taste. You learn what to delegate next. The vendor running the loop with you is part of the loop.
A few things that go wrong inside the loop:
Over-specifying. Writing the assistant a paragraph when a sentence would do. If you knew every step, you’d just do it yourself. Describe the what and let the assistant figure out the how.
Under-specifying. “Handle this.” Handle it how? The assistant needs enough context to perceive the problem. Garbage in, garbage out still applies, even when the loop is fast.
Not reading the output. The assistant shows you what it did. Read it. The observe step is where the loop tightens. Skip it and you stay stuck running the same correction every week.
When both loops are running well, the work feels like thinking at the speed of typing. You make a decision. The assistant moves. You move on.
That’s not automation. That’s the loop.