This agent loop runs itself

I want to share a pattern I've been quietly building around for the past several months.

It's changed how I work with agents. I now use it for almost every recurring job in my business.

I call it The Night Shift.

Here's the shift in thinking: Most of us still work with AI the same way we always have. We open a chat. We prompt. We wait. We review. We prompt again.

Even with skills, project knowledge, and memory, the shape of the work is the same. The agent only moves when we move.

So a while back, I asked a different question: Not how do I prompt better. But what if my agents had recurring shifts in my business?

What if, while I sleep, the work just gets done? And I only show up for the parts that need my judgment?

That's the whole idea behind The Night Shift. I unpacked it with real examples in my new video this week. But here's the gist. It has three parts:

1. A shared interface.

This is where the work lives. Where state gets tracked. Where the agent and I communicate.

It can be as simple as a markdown file with notes and checklists. Or for more complex processes, a custom app with a UI for me and an API for the agent.

The key is one source of truth that both of us (my agents and I) can read and write to.

2. Me, the human in the loop.

But not in the way you might expect. I'm not in there all day. I drop in for short, focused sessions. Two minutes, sometimes twenty.

I review the work. Leave a comment. Approve something. Check a box. Then I get back to my real work.

3. The agent with a Skill, on a schedule.

The agent runs a Skill, a step-by-step process I've written for it, on a recurring schedule.

Each time it runs, it picks up where we left off. Acts on my latest feedback. Pushes the work forward. Surfaces new items for me to review next time.

That's it. Three parts. A shared interface, me, and an agent on a schedule.

A few examples running in my business right now:

• SEO meta tags review. An agent reviews every page on buildermethods.com every two weeks. Flags stale or sub-optimal tags. Fixes them automatically, or waits for my approval on high-value pages.
• Open-source PR reviews. An agent reviews contributions on my free tools each week and queues them up for me to merge, close, or follow up on.
• Kit email sequences. An agent reviews my email sequences in Kit and surfaces what needs to be updated, rewritten, or replaced to ensure they have up-to-date info about my content and Builder Methods Pro.

Now I'm using this model across everything in my business. I'm transforming my entire content development pipeline (multiple tools, multiple agents, same 3-part loop).

How to use it

The real work in this approach isn't what the agent does every night. The real work is what you do once, upfront.

Designing the interface. Writing the skill. Thinking through the process. That's a new (or not-so-new?) muscle we have to build as operators in this era. It's where our actual leverage lives.

So here's a question worth asking when you're about to sit down and do something today:

1. Have I done this before?
2. Will I need to do it again?

If the answer to both is yes, that's a candidate building a process, an interface, and an agent to handle it using The Night Shift model.

See those real examples in my new video this week where I unpack The Night Shift model in-depth. Then leave a comment with any questions or feedback!

Keep building.