The Meta Programming Shift

Recently I’ve had to relearn how I work.

I’ve been using agentic AI for most of 2026. And for a while, the workflow was simple — tell agents what to do, review the output, adjust, repeat.

It felt like progress. But it was essentially project management, with me still as the bottleneck. Reliant on my attention, my availability, my capacity.

So I shifted. I stopped directing agents and started writing rules.

Not a single mega-prompt for one task. Rules that govern how agents behave across dozens of tasks, without me watching.

It’s a subtle change that compounds quickly. When you direct an agent, you’re trading your time for its output — one to one. When you write rules, you’re programming a system that runs whether you’re there or not. One to many. Every new agent inherits the same operating logic.

There are three levels to this:

Level 1: Delegation. Give it a task, get a result. This is where most coaches hear “AI” and stop.

Level 2: Orchestration. Running multiple agents, managing workflows, reviewing outputs. More powerful, but you’re still the bottleneck.

Level 3: Meta-programming. Writing the rules that make levels one and two run themselves.

The interesting question isn’t whether agents can do the work. It’s whether you can score the output well enough to trust it without reviewing everything.

Or whether you decide some bits you want to do anyway — because you enjoy them.

That’s the part most people skip. The goal isn’t to automate yourself out of your business. It’s to automate everything except the parts that make it yours.