On August 11, 2026, I stood in Dallas at the third annual HousingWire AI Summit, held at the George W. Bush Presidential Center, and heard the sharpest answer to the question I have been chewing on for years: why do our agents refuse to use the tech we pay for? (HousingWire) The room was full of brokers and mortgage leaders who had actually shipped AI, not just pitched it. Nobody did a victory lap. Several of them talked openly about what failed.
I was there to sit on a panel about why AI projects fail and how brokers should think about building versus buying. I have lived that question from both sides. I have been in this business 22 years. I run REMAX Victory + Affiliates in Dayton, Ohio, and I am building an AI operating system for brokerages. The first version of it failed. Not because of AI. Because it was not anything anyone wanted. I built it to solve my problem, not theirs.
Agents communicate every way there is: they call, they text, they email, they DM. I wanted to streamline that, so I built a tool where an agent could ask an AI a question trained on our standards of practice, our manuals, our onboarding process, and state licensing law. It worked. Nobody used it.
About six months in, I demoed it to another broker and he asked what my adoption was. I had to say it out loud: really, really low. We burned it down and started over. This time we met agents where they already were, meaning a phone number they can text instead of another app to download, and a social feed they are already in every day where the assistant answers questions in line in the thread. We rebuilt it as ToryOS. Adoption now runs around 90 percent, and two of my heaviest users are women in their seventies who never touched anything else we rolled out.
That is the cheap version of a lesson a lot of us pay thousands for. Which brings me to the best answer I heard all day.
What Was the Best Answer I Heard in Dallas?
Josh Miller, the CEO and co-founder of Epique Realty, gave the sharpest lesson of the day, and it is the one I said on camera I would adopt in the next 90 days. His point was that brokerage AI adoption is not the problem. Agent adoption is. Nearly every broker in that room has spent thousands of dollars on technology their agents refuse to use. (HousingWire)
His test for whether a tool will actually get used is that one of three things has to be true. Either the agents already want it and are already using something like it, or they are already paying for something like it out of their own pocket, or there is a real incentive attached to using it. If none of those three is true, you are not launching a tool. You are trying to create demand that does not exist.
The proof he cited made the floor quiet. The industry average for logging into the CRM brokerages provide is roughly one in five agents. Epique's CRM adoption rate stands at 89 percent, and the reason is not a better CRM. It is that they put free leads inside it, so if you do not log in, you do not get them. (HousingWire)
He also made a point that reframed my whole history with this. Brokers were the ones who handed agents websites and email in the early 2000s because agents could not build those themselves, and brokers are going to be the ones who hand agents AI the same way. That is the role. We are not users of the tech we buy; we are distribution for it.
I heard that as a name for the scar my failed first build left me. It took me most of a year and a painful demo to learn the expensive way that nobody wanted the tool I was proud of. Miller stated the same lesson as a clean three-condition test, and that made it usable instead of just a mark on my quarter.
Of course, some brokers will argue that agents should just be told what to use. You are the broker. You pay for the tools. You have every right to mandate them. But if your agents are independent contractors, that argument collapses before lunch. You cannot mandate anything that does not put money in their pocket or save time they already value. The three-condition test is not a theory. It is what the market structure of this industry demands.
What Surprised Me?
Two things landed harder than I expected, and both said something about adoption.
The first came from a mortgage side speaker who is training non-technical business people to build their own tools. He refused to treat it as a hobby. No hour a week on the side. They pulled their best people completely offline for a full week of immersive boot camp, homework at night, and every person had to build something and demo it to pass. His point was blunt: an hour of training a week gets you nowhere. The most valuable AI people in your company already work there, they are just business people who understand where the problems are.
That one stung, because I have been guilty of exactly the wrong version of it. I have used this stuff personally for two years and never properly taught my own staff.
The second was a warning about the age we are in. The same speaker argued that we are living through a period of subsidized AI, that today's cheap tokens are being funded by investors and will not stay cheap forever. His practical answer was unglamorous: give people a budget and track what you spend.
And there was one scene I keep turning over, from an MLS session that got deep into the technical weeds. A speaker described walking into a brokerage office of about fifty people, asking who was using AI, and ten hands went up. All of them were uploading MLS data into a public chatbot with no governance at all. That is happening in your office right now, whether you have a policy or not.
What Am I Actually Changing?
None of this is worth anything if I do not ship it. So here is what I am actually doing in the next 90 days, and what any broker in that room, or reading this, can do this week.
First, I am running our internal AI training like a boot camp, not a lunch and learn. Real time, block-scheduled, people pulled offline, homework at night, and everyone builds something and demos it to pass.
Second, I am applying the three-condition test to everything in our stack before we roll anything else out. If our agents do not already want it, would not pay for it themselves, and have no incentive to use it, it does not ship. That includes the rebuilt assistant platform, and it holds for every tool a vendor pitches us. I have wasted enough money proving the test is real.
Third, the free exercise I gave the room, because it costs nothing: write down 25 questions your agents asked you in the last month. That list is your build list. It will tell you more than any vendor demo will, because those questions are demand that already exists. Your next tool is already in that notebook. All you have to do is meet your agents where they already are.
So that is my question to you, and I mean it as a peer, not a pitch. What are the 25 questions your agents asked you in the last month?
Discussion