I got into this back in 2014 wanting to build something as basic as a document filler for real estate transactions, a tool that barely existed in our space at the time. What I learned almost immediately was that the paperwork was the least of it. The real work of a deal lived everywhere else, in the scheduling, the calendars, the follow-up, and the handoffs between people, and almost nothing on the market tied any of it together. So when someone tells me artificial intelligence is about to transform real estate, I tend to ask a blunt question in return. Transform which part, exactly? Because most of the money being spent on AI in this industry right now is aimed at the wrong part of the business, and the numbers are starting to show it.
The Pilot Trap
The adoption story looks like a triumph if you only read the headline. Delta Media’s 2026 survey found that 97 percent of brokerage leaders say their agents are now using AI, up from 80 percent just two years earlier. Almost nobody is sitting it out. But adoption and results are two very different measurements. JLL’s 2025 Global Real Estate Technology Survey of more than 1,500 senior decision-makers found that while the overwhelming majority of firms have started AI pilots, only 5 percent report hitting most of their program goals, and more than 60 percent admit they are not yet organizationally or technically ready to execute. Deloitte’s 2026 outlook tells the same story of enthusiasm cooling into realism, with 27 percent of firms still wrestling with implementation problems like technical issues, missing expertise, and resistance to change, and only a sliver reporting that AI has been genuinely transformative. As Deloitte put it, organizational readiness and change management may matter as much as the technology itself.
I have watched this happen up close more times than I can count, and I want to be fair to the brokers doing it, because every one of those tools is chasing a real pain point. The listing-description generator is about getting sharper marketing copy for a client’s property. The website chatbot is about giving buyers who are searching a faster, more useful response so they stick around. The transaction-summary tool is about staying on top of the endless list of tasks a broker has to complete to meet their fiduciary duty to a client. The lead-scoring add-on is about understanding where someone actually is in their home-search journey so you can meet them there. Good instincts, all of them. The problem is what happens next. A brokerage signs up for all four in the same quarter, and by the next quarter none of them are part of how anyone actually works. The tools were not bad. They were dropped on top of a workflow that was already broken, so each one became one more thing to check instead of one less thing to do. A pilot that does not remove work will not survive a busy Tuesday.
The Wrong Layer
Look at where the AI is actually being pointed and the problem gets obvious. A 2026 Real Brokerage agent survey found that 88 percent of agents use AI for marketing and content creation, while only 58 percent use it for client communication. Brokers have gotten very good at pushing a listing live quickly, and much less good at following up on the early touch points and questions that come in from potential clients right after. In my experience the lag on those first inquiries is rarely a matter of hours. They can sit for weeks before anyone responds in a consistent way. The fix is not another content tool. It is setting a clear standard across the whole transaction lifecycle for what a client can expect from their agent, and then holding the workflow to it.
This is the part that frustrates me. Content generation makes a great demo and does very little for the operational core of a brokerage. That core is the unglamorous machinery of getting a lead to the right person before it goes cold, getting a new agent set up and productive without eating a manager’s entire afternoon, and keeping every person who touches a deal looking at the same version of the truth. Fix those and the gains compound across every transaction you do. Generate a prettier email and all you have is a prettier email.
Why “Just Add an Agent” Falls Apart
Here is where I part ways with a lot of the current hype, though I understand exactly how brokers end up here. They are reaching across into other industries and technologies to try to serve their clients better, and in the process they exhaust themselves pouring time and money into inefficient directions. General-purpose agents are the clearest example. The popular move right now is to take one of these agents, wrap it around your existing pile of disconnected tools, and hope it behaves. The trouble is that you are handing human-level decisions to something that was never designed to make them, and the only lever you have to keep it in line is prompting. So people write longer and longer prompts trying to build guardrails into the thing, burning tokens and adding cost, and it still slips. And there is no clean way around that. Even the protocols you add to restrict what the agent is allowed to touch have to live somewhere, in the prompt or the project guardrails, which means they eat the same token bandwidth and pile on the same cost. You end up funneling a general-purpose tool into a narrow shape it was never built for, hamstringing it and calling it automation.
The alternative is less glamorous and far more reliable. Instead of turning one general agent loose across a dozen apps, you build the automation itself, defined workflows that run on a single connected source of truth and do not depend on an agent making decisions on the fly. The model still has a job, but it is an assist role, tweaking and optimizing a workflow, personalizing a message, pulling the right reference point at the right moment. It rides on top of the automation instead of being the automation. That order matters, because the moment you hand the whole process to an agent, context starts slipping and client relationships fall through the cracks, and in this business your past clients are the business. Build the workflow first and the data stays intact. Everything a deal touches passes through the same system. There is no patchwork of logins, no handoffs where information falls on the floor, no five tools each holding a slightly different version of the same record. I came to that conclusion about six years ago, and I will be honest, it is a little strange to now watch industry reports arrive at the same place and describe it as a discovery.
What AI Still Can’t Touch
Then there is the part of this job that AI does not touch, and in my view should not. This is a relationship business, and the data on that is not subtle. NAR’s 2025 Profile of Real Estate Firms found that 46 percent of brokerage sales volume comes from repeat business and another 44 percent from past-client referrals. Ninety percent of the work comes from people who already trust you.
You cannot prompt your way to that trust. A home is the single largest store of wealth most American families will ever hold, and when someone hands you that decision, they are paying for judgment, not for a chatbot. They are paying for the person who talks a nervous first-time buyer off the ledge the night before closing, who reads a multiple-offer situation and knows which seller wants the highest number and which one wants certainty, who catches the line in an inspection report that is technically minor and practically a dealbreaker for this particular family. The recent panic that agentic systems will automate away the junior work misses the point. The value in this business was never the junior work. It was the judgment that the junior work was quietly training people toward.
A Test Before You Buy
So when a vendor shows me something new, I have stopped asking whether the AI is impressive. I ask three things instead, and I would suggest any operator do the same. Does it remove a task, or does it just add another screen to check? Does it live inside the systems my people already use all day, or does it demand a detour they will quietly abandon? And when it gets something wrong, can I say clearly who is accountable? If a product cannot answer those three, it does not matter how advanced the model underneath it is. It will end up in the graveyard of pilots with all the others.
The Real Divide
Over the next couple of years, the gap between brokerages will widen, but not along the line most people expect. Nearly everyone has adopted something now, so it will not separate the firms that use AI from the firms that do not. It will separate the ones that used AI to strip out operational drag from the ones that used it to bolt a layer of novelty onto the same old mess. Refusing to rethink the workflow underneath the tool is its own kind of self-sabotage, an artificial handicap you are choosing to carry.
The winners will look unremarkable from the outside. Their agents will simply spend more time with clients and less time chasing paperwork, and the technology making that possible will be close to invisible. That is the tell. In this business, the AI that matters is not the AI your clients can see. It is the AI that quietly makes sure you never drop the one thing that would have cost them their trust.
