Designing the Agent You Will Keep Using

You have now built or watched four different agents: a file organizer, a comparison shopper, a competitor watcher, and a store builder.

Those were demos. They taught you the mechanics.

But the agent that actually changes your life is not a demo. It is the one you build for your own specific situation and keep running for months.

This chapter is about designing that agent.

No new tools. Just thinking.

Start with Friction, Not Features

Do not ask, "What cool thing can an agent do?"

Ask, "What do I do every week that I hate doing?"

The best agents remove recurring friction. Boring, repetitive, every-week tasks are exactly where agents shine, and they are where you will feel the benefit most.

Make a list:

• What do you do every Monday?
• What do you do every morning?
• What do you do every time a certain kind of email arrives?
• What do you do every time you start a new project?

Those repeated motions are your best agent candidates.

Run each candidate through the green/yellow/red filter from Section 1.4.

Category	What it means	Best use
Green	Repetitive, rule-based, and low-stakes if something goes slightly wrong.	Perfect for agents: research, summarizing, organizing, drafting, and monitoring.
Yellow	Useful, but requires your judgment or has some risk.	Good with supervision, especially for tasks involving money, customers, or published content.
Red	High-stakes, hard to undo, or requires serious human judgment.	Do not automate. Keep these tasks under your control.

Your first keeper agent should be a green-light task. Build trust before handing over anything yellow.

Design It with the Anatomy You Already Know

Every agent you keep should use the prompt structure from Chapter 4.1:

Prompt part	Question to answer
Goal	What outcome do you want every time it runs?
Constraints	What apps, formats, sources, and guardrails should it follow?
Done condition	How does the agent know the task is finished?

Then decide how the agent should run:

Run type	How it works
On demand	You trigger it manually when needed.
Scheduled	It runs on a set clock, such as daily or weekly.
Dispatch	You trigger it from your phone, and it runs on your desktop.

The Keeper-Agent Test

A good keeper agent passes these checks:

• It solves real, recurring friction, not a one-off problem.
• It is a green-light task you are comfortable handing over.
• It uses connectors you already have set up.
• You can describe what done looks like in one sentence.
• You would genuinely miss it if it stopped working.

If your idea passes all five, build it. That is your agent.

Start Small, Expand Later

Do not try to build a mega-agent that runs your whole life on day one.

Build one agent that does one thing reliably. Use it for a couple of weeks. Then expand it, or build a second one.

The people who get the most out of agents are not the ones with the most complex setups. They are the ones with a few simple agents they actually trust and use.