I Stopped Writing Prompts. Here's What Replaced Them.

How a markdown file, a Sunday afternoon, and zero coding changed how I think about AI in real estate.

Most people in real estate are collecting prompts.

“Use this one for listing descriptions.” “Try this one for buyer emails.” “Here’s my mega-thread of 50 prompts for agents.”

I get it. I started there too. And prompts work. If you’re an agent, a manager, or a marketer who went from staring at a blank screen to generating a solid first draft in 30 seconds, that’s real progress.

But I want to show you what comes next. Because the distance between “using AI with good prompts” and “building AI into how you work” is where things get genuinely interesting. And you don’t need to be technical to cross that line.

I’m not a developer. I can’t write code from scratch. But I can work alongside AI to build real things. Last Sunday, I spent the afternoon setting up a local knowledge base using a vector database and Claude. The Sunday before that, I built a custom workflow that triages my inbox by urgency and stakeholder priority before I’ve had my coffee.

I’m also not an AI expert. I’m an operator. What I’ve found is that there’s a whole space between “I use ChatGPT sometimes” and “I’m a developer” that almost nobody in real estate is talking about. A space where you build real systems, real workflows, real infrastructure, using plain English and AI as your co-builder.

That’s what this post is about.

The Prompt Ceiling

Here’s the problem with prompt libraries: every time you use one, you’re still doing the work.

You find the right prompt. You paste it into a chat window. You feed it context about this specific property, this specific client, this specific situation. You review the output. You tweak. You paste the result somewhere else.

That’s not automation. That’s a more organized version of doing everything yourself. You’re still the middleman between the AI and your work product.

I hit this ceiling a while back myself. I had dozens of saved prompts. They were good. But I was spending almost as much time managing the prompts as I was saving by using them.

Then I found something that changed the equation entirely.

What a “Skill” Actually Is

Claude, the AI I use daily (and wrote about in my last post), has a feature called Skills.

The concept is simple: A prompt is something you type every time. A skill is something your AI already knows how to do.

You write a plain-English instruction file, a single markdown document called SKILL.md, that describes a workflow. What triggers it. What steps to follow. What rules to respect. You upload it once. From that point forward, Claude follows those instructions automatically whenever the right context shows up.

No code. No JSON. No API keys. Plain English in a text file.

Here’s one distinction worth understanding early: a Project is where your AI stores what it knows about you. A Skill is where it stores what it knows how to do. Projects give your AI context. Skills give it capabilities. Both matter. But if you’re just getting started, Projects come first.

A skill has four parts:

• Description: What this workflow does.

• Trigger Phrases: When it should activate.

• Procedure: The steps, in order.

• Rules: Guardrails so it doesn’t do something you’ll regret.

That’s it. If you can write a checklist for a new hire, you can write a skill.

But Here’s the Thing Most People Miss

Before you ever build a custom skill, Claude already has powerful capabilities built in that most users never discover.

This is where I want every agent, manager, and marketer reading this to pay attention, because what I’m about to describe is available to you right now, today. No setup. No uploads. No technical anything.

Claude can create actual files.

Not text in a chat window that you copy and paste. Actual, downloadable, ready-to-use documents.

Tell Claude: “Create a listing presentation for 47 Maple Drive, a 4-bedroom colonial in Westport listed at $2.85M. Include a market overview, pricing strategy, and 90-day marketing timeline.”

You get a PowerPoint file. A real deck that opens in PowerPoint or Google Slides. Formatted slides. Speaker notes. Ready to customize and present.

Tell Claude: “Here’s my last 6 months of closed sales data. Build me a client-ready market report with median price trends and days-on-market by neighborhood.”

You get an Excel spreadsheet with working formulas. Or a formatted Word document. Or both.

One honest caveat here: Claude can build the deliverable, but it can’t pull live MLS data on its own. That’s the integration layer the industry is still building. I’ve been exploring a platform called Repliers that connects MLS data directly to Claude through an MCP server, essentially a bridge that lets your AI talk directly to an external data source without you copying and pasting anything in between. It’s early, but I’ve gotten it running. When this layer matures, the gap between “AI can help me write” and “AI can build my entire listing launch package from live MLS data” closes fast.

Tell Claude: “Write a just-listed email for my sphere, a property description for MLS, and a social media caption. The property is a waterfront contemporary in Rowayton with 180-degree harbor views.”

You get a Word document with all three, formatted and ready to use.

This isn’t future technology. This is what the tool can do for you today.

What You Should Build First (And It’s Not a Skill)

If you take one thing from this article, let it be this:

Build a Project.

I use Claude, but if you’re a ChatGPT user, set up a project there. The concept is the same. Just get started. The tool matters less than the habit.

A Project is a workspace where you upload documents that give the AI permanent context. Think of it as giving your AI a filing cabinet about your business.

Here’s what to put in it:

• Your last 15-20 listing descriptions and marketing emails

• Your bio and personal brand positioning

• Your active listings with property details and talking points

• Market data for your farm area

• Any templates you reuse: buyer guides, listing presentations, CMA frameworks

• Examples of emails or messages that sound like you

Now every conversation inside that Project starts with Claude already knowing who you are, how you write, what you’re working on, and what your market looks like.

You don’t need to engineer anything. You just talk.

“Write the description for the new listing on Harbor Road. You know my style.”

And it does. Because it’s read 20 examples of your writing and understands your voice, your level of formality, the way you describe architectural details, your tendency to lead with lifestyle over specs (or the reverse).

“Draft a follow-up email to the Millers after Saturday’s showing. They loved the primary suite but were concerned about the commute. Be encouraging but honest.”

Claude writes in your voice, references the specific property and the specific concern, and produces something you’d actually send. Maybe with a tweak or two. But the heavy lifting is done.

“I have a listing appointment Thursday. The sellers interviewed two other agents that I compete with daily. Help me prep three talking points that differentiate my approach.”

Now Claude is pulling from your brand positioning, your track record, your market knowledge, all uploaded in the Project, to help you prepare for a competitive pitch.

This takes about 20 minutes to set up. The return is immediate and it compounds over time as you add more materials.

The Builder Category

There’s a narrative in this industry that you’re either “technical” or you’re not. You either code or you don’t. You’re an AI person or you’re still figuring out what an LLM is.

I think that’s wrong, and I think it’s holding people back.

I spent last Sunday building a vector database on my laptop. I didn’t write the code. I described what I wanted, Claude wrote the scripts, and we troubleshot together when things broke. By the end of the afternoon, I had a working local knowledge base: a perfect memory of every conversation I’ve had across Claude, ChatGPT, and Perplexity, searchable and queryable from inside my AI workflows. No more losing context between tools or starting over every conversation.

I couldn’t have done that alone. But I also couldn’t have done it if I didn’t understand what I was trying to build and why. The AI handles the syntax. I handle the architecture and the intent.

That’s the new category. Not “coder.” Not “user.” Builder.

The more I study the foundational technology, the more convinced I am that this category is where most ambitious professionals will land. You don’t need to understand backpropagation to build a skill that saves your team 10 hours a week. But understanding how these models process context, retrieve information, and generate output makes you a dramatically better builder.

What This Means for Residential Real Estate

The brokerages that figure this out first aren’t just going to save time. They’re going to operate at a level of consistency and speed that becomes a value-add for their agents.

When a prospective agent sees that your platform can produce a custom listing presentation in minutes, that your onboarding is seamless, that your market reports generate themselves, that your AI tools actually know the agent’s voice and market, that’s a value proposition they can feel. Not another CRM demo. Not another training portal. Something that tangibly makes their business better from day one.

And agents should be asking about this. Not “do you have AI tools?” Everyone will say yes. The better question: “What have you built?”

The answer tells you everything about how seriously a firm’s leadership is thinking about what comes next.

Where I’m Going With This

I don’t have this figured out. Nobody does. There’s a saying I’ve carried with me for years: you’re living life in beta. Nothing is ever fully complete. It may be shipped to production, but an update is right around the corner. That’s all of us right now with AI. The tools change monthly. The capabilities expand weekly. The best you can do is stay in motion, build what you can with what’s available today, and be ready to adapt when the next update drops.

That’s partly why BrokerageOS exists.

But I’ve seen enough to know that the gap between “uses AI” and “builds with AI” is smaller than most people think. It starts with a Project that knows your business. It scales with Skills that encode your best practices. And it compounds as you invest the time to understand what these tools can actually do.

Someone needs to explore this openly. To share what works, admit what doesn’t, and pull others forward.

That’s the job I’m signing up for. I hope you’ll come along.