I've Watched Agents Fail at AI for Three Years. Here's What's Actually Going Wrong.
I train agents on technology every single day. Have for 15 years. And for the last three years, I've had a front-row seat to every way a real estate agent can get AI wrong.
There are exactly two ways it happens. Agents either use it blindly and never check what comes out the other side, or they're so intimidated they don't touch it at all. I haven't seen a third category. Those are the two failure modes, and the industry keeps feeding both of them.
The Training Industrial Complex Isn't Helping
Every conference now has an AI session. Every coach has an AI module. Every brokerage has sent out some version of "here are 10 prompts to try." I've seen the materials. Most of it is useless.
Not because the information is wrong. Because it stops too early. It teaches agents to produce AI output without teaching them to evaluate it. So you get a luxury listing description that sounds like it was written by someone who has never been inside a home. Or a market summary with numbers that are just wrong. And the agent posts it because the training said "use AI to save time" and they did.
That's not efficiency. That's a liability.
The agents who avoid it entirely aren't being irrational either. They've watched colleagues embarrass themselves with bad output. They don't trust the tools, and honestly, with the training most of them received, that skepticism is earned.
The Problem Isn't the Technology
I want to be clear about something. The AI tools themselves are genuinely capable. That's not the issue.
The issue is that the tools are general-purpose and real estate is specific. When you hand an agent a general AI interface and tell them to go figure it out, you're asking them to become a prompt engineer on top of everything else they already do. Most of them have no interest in that, and they shouldn't have to.
I've been a licensed agent for 21 years. I know what the job actually involves. You're managing clients, managing timelines, managing the emotional weight of what is usually the biggest financial transaction of someone's life. The last thing you need is another tool that requires a learning curve before it does anything useful.
What actually works is the opposite approach: build the AI into a workflow that's already familiar, keep the interface simple, and make the output specific enough that the agent can immediately tell if something's off. Remove the friction before the agent ever touches it.
What I Built Instead of Another Framework
A couple of years ago I stopped trying to train agents on AI tools that weren't built for them. I started building tools that were.
The first one is ListingSpotter. It helps agents and transaction managers locate their advisor's listings anywhere in the world. Sounds simple. It is simple. That's the point. Summit Sotheby's is already using it, and the reason it works isn't because it's technically impressive. It's because it fits exactly how agents actually work, not how a developer thinks they work.
I'm also building out tools specifically for high-rise and luxury condo marketing, where the data and the pitch are completely different from a single-family listing. Those aren't done yet, but they're coming from the same place: a specific problem, a specific workflow, a specific solution.
None of this came from reading about AI. I use these tools myself. I've built them inside the same industry I've worked in for over two decades. That's not a credential I'm listing to impress anyone. It's just the reason the tools don't have the gaps that most off-the-shelf solutions do.
What Mid-to-High-End Agents Actually Need
If you're a luxury agent, your clients notice everything. A mediocre listing description isn't just a missed opportunity, it's a signal about your attention to detail. Bad output that goes unchecked doesn't just waste your time, it costs you credibility.
So the bar is higher. And that means the tools and the training both need to clear a higher bar.
What I see working for agents at this level is not more AI education. It's custom tools that do one thing well, fit into an existing workflow, and produce output that the agent can actually stand behind. Specific enough to be useful. Simple enough to actually get used.
That's what I'm building. Not a course, not a prompt library, not a framework. Product.
The Honest Version of Where This Goes
I'm focused on Sotheby's agents right now because I know that world. I know the brand standards, I know the client expectations, I know what breaks at scale because I've watched it break at scale for 15 years.
But the problem isn't unique to one brokerage. Every luxury agent in the country is sitting somewhere between "using AI badly" and "not using it at all." The gap I'm building into exists everywhere.
I'll expand when the tools are right. Not before.
If you're a mid-to-high-end agent who has tried AI, hated what came out, and isn't sure what to do next, that's exactly who I'm writing for. Stick around.