Skip to content
Agent Delegation·2 min·English

The 5 rungs I used to build an agent that runs my campaigns

How you build authority for an agent one rung at a time: a small task, a limited action, repetition, a skill, and only then a scheduled agent.

The 5 rungs I used to build an agent that runs my campaigns

What you get

A staged template for handing real authority to a code agent, without asking it to replace a person on day one

Who it is for

For anyone who asks a code agent for a task that is too big, gets refused, and wants to know how to actually build that authority.

Why it works this way at all

When you ask a code agent to "replace my campaign manager," that is a huge, irresponsible request. There is no definition of what it can touch, what it cannot, and how you know it succeeded. The refusal is not a bug, it is a sign the request is too big. The fix is not to fight the refusal. The fix is to build authority one rung at a time: each rung proves itself before you move to the next.

Rung 1: one small, checkable task

Don't ask for "manage the campaigns." Ask for something specific and small you can verify by eye: "find every campaign with the worst performance this month." That is a read-only task, there is no risk, and it is easy to check if the answer is right.

Rung 2: a limited action on what is already approved

After the first task has worked a few times, let the agent act, but only on what it already identified: "cut the budget by 20% on the campaigns you flagged." Not "manage budgets," a defined action on a list that is already approved.

Rung 3: repetition and consistency

Run the same task several times under different conditions (good campaigns, not just bad ones; different days). The goal is to make sure the agent did not just get lucky once, and that the process is consistent.

Rung 4: turning it into a skill

Once the process works consistently, ask the agent (or build it yourself) for a skill that documents the exact process: which campaigns to check, what threshold defines "bad," how much to cut or raise. Now the agent does not need you to re-explain every time.

Rung 5: a scheduled agent

Only after the skill has proven itself is it time to run it as a scheduled agent, running on its own at a fixed time (say, 11pm every night) without you triggering it manually. This is the last rung, not the first.

The simple rule

You do not ask an agent to replace a person on day one. You let it prove itself one rung at a time, until it actually earns the authority. This works for campaigns, and for any other authority you want to hand an agent.

How to use this now

Start on the first rung today. One small, checkable task. The rest builds itself.

AI-native products, workshops, and automations. Built from everywhere.

© 2026 Daniel Goldman