2026-05-17

#21 Build With Intention | The Agent Loop that Changes Everything

Hello Reader,

Today, I want to talk about the agent loop.

At their core, AI agents are just while loops with memory and tools.

Intention

What makes AI Agents special is one word. Loop.

An agent is a loop. It plans. It acts. It observes what happened. It reflects on whether it's done. And then it does it again, until the work is finished.

A prompt is a request and a response. You ask, it answers, it stops. Nothing happens until you come back.

It's the reason agents can do things that prompts can't.

Insight

Here's the loop in plain language.

Plan. The agent decides what to do next. Not the whole job. Just the next step.

Act. It does the step. Calls a tool. Writes a file. Sends a message. Reads a page.

Observe. It looks at what happened. Did the tool return what it expected? Did the file save? Did the answer make sense?

Reflect. It asks itself a quiet question. Am I done? Or is there a next step? If there is, the loop continues.

That's it. Four moves, on repeat, until the work is finished.

I'll give you a real example.

Muse is the system I built to help run this newsletter. One agent captures ideas during the week. Another researches them. Another shapes the strategy. Another writes. Another publishes. They share a memory layer, so they're not starting from scratch each time.

Here's what matters. None of them are doing a single task and stopping. Each one is running its own loop. The capture agent doesn't just catch one idea on Tuesday. It runs all week, watching, noticing, saving. The strategist doesn't write one plan. It revisits, adjusts, and reconsiders as new information comes in.

When the loop is running, the work happens whether I'm at my desk or not. That's the unlock.

Action

Here's what I want you to try this week. Two small steps.

Step one. Look at your week and ask one question.

Which of my workflows ends when I stop?

Pick anything. Reviewing applications. Writing a status update. Researching a competitor. Following up with leads. Whatever it is, ask yourself honestly: Does this stop the moment I stop, or could it keep running?

Some workflows genuinely need you every time. Strategy. Decisions. Real relationships. Keep those.

But others end only because you're the loop. You're the one planning, acting, checking, deciding to do it again. Take you out, and it stops.

Those are candidates.

Step two. Pick one and map it as a loop.

Take one of the candidates and write down what it would look like as four moves.

What's the plan step? What's the action? What does the agent observe to know if it worked? What does it reflect on to know if it's done?

You don't have to build it. You just have to see it. Once you can sketch a workflow as a loop, you will understand agentic AI intuitively.