The Augmented | Future of Human Resilience

I've seen smart people freeze in the middle of a burning building because they didn't have a clear order. I've also seen them get killed because some cowboy in the command post gave them an order that made zero sense for the conditions on the ground.

On the fireground, we don't "prompt" each other. We command. We communicate. We close the loop.

If you’re still "prompting" your AI agents, you’re not a leader. You’re a hobbyist. And in a high-stakes professional environment, that’s how you end up with a mess of hallucinations and wasted compute.

Here is how you actually manage an AI team. It’s not tech. It’s Incident Management 101.

##1. Span of Control: Don’t Overload the Brain

In the NSW Fire Brigades, we had a strict rule: Span of Control. One person manages three to seven people. Any more, and your situational awareness collapses. You stop leading and start drowning in noise.

The same applies to AI. I see people trying to build "God-prompts" — one massive block of text that tells an AI to do ten different jobs at once. Research this, write that, check the code, format the email, and make it funny.

That is a span of control failure.

An AI model’s "attention" is a finite resource. If you give it seven tasks, it’ll do two well, three poorly, and lie about the rest.

On my laptop, I run seven distinct agents. Gem handles the deep research. Ritz writes the copy. Gee builds the code. Each one has a single, defined role. I don't ask Gem to write poetry. I don't ask Ritz to debug Python. I keep the span of control tight so the output stays sharp.

##2. Commander’s Intent: Tell Them WHAT and WHY, Not HOW

When a Captain sends a crew into a structure, they don't say, "Put your left foot forward, then your right, then turn the nozzle 15 degrees." That’s micromanagement. It fails the moment the smoke gets too thick to see your own hands.

Instead, they give **Commander’s Intent**.

"We are going in to search the primary bedroom and vent the roof. Our goal is life preservation. If the floor feels soft, you pull out immediately."

The crew knows the *outcome* and the *constraints*. They use their local intelligence to figure out the *how*.

Stop giving your AI a list of 50 instructions on where to put commas. Give it the intent. "Your intent is to convince a skeptical operations manager that their current fire compliance is a liability. Focus on the legal risk, not the hardware. Tone is professional but urgent."

When the AI understands the *mission*, it can adapt. When it’s just following a script, it’s useless the moment the context shifts.

##3. The Accountability Loop: No Passive Acceptance

On the fireground, if I give an order, the recipient repeats it back. "Primary search, vent roof, copy." Then they report when it’s done. That’s a closed loop. Without it, you have no accountability.

Most people treat AI like a vending machine. You put in a prompt, out comes a product, and you use it. That’s an open loop. It’s dangerous.

In my workflow, the Human is the Lead. Always.

When an agent gives me an output, that’s not the end. That’s the beginning of the Accountability Loop. I interrogate it. I ask why it made a specific call. I check it against the Commander’s Intent.

If it’s wrong, I don't just "re-prompt." I hold the agent accountable to the original brief. "You focused on the hardware. My intent was the legal risk. Go back and fix the focus."

##Human in the Lead

The metaphor is simple: You are the Incident Controller. The AI agents are your crews.

They are faster than you. They can go places you can’t. They don’t get tired. But they have no soul, no skin in the game, and no local "feel" for the building.

If you step back and let the AI run the scene, the building burns down. If you micromanage every step, you might as well do it yourself.

Stop prompting. Start commanding. Be the human in the lead.

— J.