The Automation Line: Mapping Where AI Actually Stops in Real Estate
Every AI conversation in this industry eventually lands on the same anxious question: which jobs disappear first. A recent piece of research, shared by Apollo’s chief economist Torsten Slok and drawn from a forthcoming book by economists Luis Gariçano, Jin Li and Yanhui Wu, gives that question a much more useful shape. Their framing is not about job titles at all. It is about task structure, and it happens to map almost perfectly onto how I think about property intelligence.
The core insight is this: AI adoption is not driven by how complex or relationship-heavy a task looks from the outside. It is driven by how “clean” the task is. Clean tasks have clear inputs, clear rules and a checkable output. Messy tasks involve incomplete information, competing incentives and judgment calls that cannot be fully specified in advance. As a task gets messier, the cost of letting AI run it without a human checking the work rises faster than AI’s raw capability improves. That gap is the moat.
Mapped onto real estate brokerage, the clean layer is easy to name: lead qualification, listing copy, basic comparative market analyses, scheduling, document assembly and first pass market research. These are exactly the workflows that property intelligence platforms have been chipping away at for years. What stays scarce, and becomes more valuable as the clean layer automates, is pricing judgment, negotiation strategy, local context and client counseling. In other words, the intelligence layer does not replace the interpretive layer. It clears space for it.
As AI takes on more entry-level tasks, organizations risk losing the pathways that turn novices into experts.
That quote, attributed to Deloitte’s research on what it calls a broken skills ladder, is the part of this story that data people should sit with longest. Dashboards and automated CMAs are genuinely useful. They also happen to be the exact repetitive tasks that once trained junior agents to read a market. If a platform absorbs that training ground, the next generation of analysts and advisors needs a different route to develop judgment, not just faster software.
There is also a signal worth tracking for anyone building or buying property intelligence tools. Commissions rebounded after the NAR settlement rather than collapsing, which suggests the market is still paying, quite deliberately, for the messy work agents do. That is not an argument against automation. It is an argument for being precise about what you automate. Megateam structures, where junior staff handle AI assisted admin while a senior agent owns the client relationship, look like a sensible answer to the skills ladder problem, and they are worth watching as a model for how data tools and human expertise actually combine in practice.
The takeaway for KG Data readers is not that AI is overhyped or underhyped. It is that the line between what a system can safely automate and what still needs a person is not fixed. It moves with how messy the task is, and mess, in housing markets exposed to shifting insurance and climate risk especially, is not going away anytime soon. The smartest property intelligence tools will be the ones designed around that line, not against it.
Source: The Real Deal
