The Agency After Prompts: How Agencies Win When AI Makes Execution Cheap

This summer I spent an afternoon with a small agency talking about AI. I brought a deck called The Agency After Prompts, and I'm sharing it here because the conversation it started hasn't stopped for me.

Every question they asked was really the same question, the one every client is thinking whether they say it out loud or not:

"Why should I pay an agency when AI is faster and cheaper?"

The agencies that answer it well will thrive. The ones that answer it poorly will become vendors.

The wrong answer: competing with AI on execution

The instinctive response is to compete with the machine: faster production, more content, prompt expertise, lower costs. I understand the pull, because it feels like keeping up.

But it's the exact battlefield where AI gets stronger every month. If your strategy is to out-execute the thing whose whole job is execution, you've picked the one fight you're guaranteed to lose.

The right answer: sell what stays scarce

Clients were never really buying execution. They were buying your judgment, your taste, your history of imagination, your perspective, and the relationship they trust when the stakes are high. And sometimes the courage to tell them what they need to hear rather than what they want to hear.

AI makes research, writing, production, analysis, and iteration abundant. When something becomes abundant, whatever stays scarce gets more valuable. Taste, judgment, original insight, trust, and human connection become the premium assets.

That's why I think great agencies get more valuable in an AI world, not less. The catch is that they have to stop pricing the thing that just got cheap.

How I actually use AI: from assistant to "little employees"

I'm not an AI expert, and I didn't set out to become one. I wanted more leverage. My own path went in four steps: doing everything myself, then using AI as an assistant, then hiring it like an employee, and now treating it more like an operating system.

The biggest shift came when I stopped using AI as software and started building what I called "little employees" in the room. I have a brand strategist, a research analyst, a startup evaluator, and an investment analyst. Each has a role and its own knowledge base, and each gets better as that knowledge grows.

Underneath all of them is memory. I collect newsletters, workflows, frameworks, marketing books, and experiments, and instead of letting them disappear into bookmarks, everything goes somewhere I can actually ask questions of. That's what lets me connect a framework from a healthcare project to a problem in fast food. Intelligence is abundant now; experience and memory are what compound.

Two questions from the room

What's the difference between a Claude Project and a skill?

A project is more like an employee: a role with trained knowledge. A skill works across projects. Your writing style is a skill, and so is the difference between your LinkedIn voice and your email voice.

Is it really better to build separate AI employees for separate jobs?

Yes, and the advantage is avoiding slop. The more specific the role, the better the work. There's a practical reason too: AI context degrades. Over a week of use, a single catch-all chat starts forgetting what you told it at the start. One person in the room said, "I didn't realize that happened." Most people don't.

Where agencies should start with AI: a one-year roadmap

Don't start with agents. Start with whatever you do every week that's repeatable.

My example was proposals. The same content goes into every proposal, which makes it a perfect first AI employee. Someone in the room said, "Man, I'd love to get out of the business of creating proposals." That's the right instinct.

The order I'd recommend:

• Week 1: Granola and Claude. Record your meetings and start asking questions of them.
• Month 1: Build an agency brain. Every meeting, brief, and proposal becomes searchable institutional memory. The most expensive waste inside most agencies is lost intelligence, not money.
• Month 2: Hire three AI employees.
• Quarter 1: Launch your first agent.
• Year 1: Build a product.

And when you're stuck, ask the AI how to do the thing. "Hey Claude, how should I write a prompt for this?" It will teach you the process.

The principles that hold up

The specific tools in this deck will keep changing month to month. The principles underneath them are what I'd build an agency on:

• Judgment over speed
• Memory over prompts
• Specific employees over one do-everything chat
• Start with what's repeatable, and let it compound

I closed the talk with the line I believe most:

The agencies that lose will use AI to cut costs. The agencies that win will use AI to increase imagination.

The full deck

The complete deck, The Agency After Prompts, is free to use. Take it, steal from it, and share it with your team. The tool slides have a short shelf life; the rest doesn't.

Working through this inside your agency?

If you're running an agency or a marketing team and working out where AI fits, I'd like to hear where you're stuck. I help agencies and brand teams figure out what to keep, what to hand to AI, and what to build next. Get in touch, see my work, or subscribe on Substack for the next one.